Every record, as text — the same data the orbit map plots above.
Cybersecurity · Hybrid · Off the LLM map (in-house non-LLM ML)
Foundational models: in-house non-LLM ML only
Stack:
Compliance: end-to-end encryption, zero-knowledge architecture, SOC 2 Type 2
1Password’s production architecture is centered on a zero-knowledge, end-to-end encrypted vault service, where plaintext is encrypted on the client before reaching 1Password’s servers. Publicly available sources confirm hybrid cloud deployment patterns for components such as the SCIM bridge, but they do not reliably expose the company’s full internal database stack or all backend infrastructure details.
Cybersecurity · AWS · Native (own / self-hosted weights)
Foundational models: Abnormal Behavioral Foundation Model — Native (own / self-hosted weights) (primary); Abnormal ABX / behavioral ML models — Native (own / self-hosted weights)
Stack: PostgreSQL, Amazon S3 (data lake/object storage), Elasticsearch/OpenSearch (log/search, inferred)
Compliance: SOC 2 (inferred from enterprise email security norm), ISO 27001 (inferred), GDPR-aligned data processing (inferred)
Abnormal runs a cloud-native, API-first behavioral AI platform that integrates with cloud email and collaboration suites (e.g., Microsoft 365, Google Workspace) outside the mail flow, ingesting user, email, and identity telemetry into proprietary behavioral models and a data lake to drive detection, response, and human-risk scoring.[1][5][7][8][9] Its core is a proprietary behavioral foundation model and related ML pipelines hosted in its own cloud environment, exposed through its SaaS console and APIs rather than via public/open-source model distribution.[7][9]
Healthcare · AWS · Native (own / self-hosted weights)
Foundational models: Abridge clinical (own) — Native (own / self-hosted weights) (primary); GPT — Direct API
Stack: PostgreSQL, vector
Compliance: HIPAA, SOC 2
Ambient clinical-documentation AI with its own medical models plus partner LLMs, grounded in clinician audio.
Creative · Multi-Cloud · Native (own / self-hosted weights)
Foundational models: Adobe Firefly (family of diffusion and generative models, including Image 3 and Firefly Foundry variants) — Native (own / self-hosted weights) (primary); Custom/bespoke Firefly Foundry models (brand-tuned generative models) — Native (own / self-hosted weights); NVIDIA-based Firefly stack (CUDA-X, NeMo, Cosmos stack for Firefly models) — Native (own / self-hosted weights)
Stack: Amazon S3 (data lake/object storage for AI training and inference)[14], Amazon FSx for Lustre (high-performance file storage for GPU training workflows)[14], Managed storage and data services on AWS (EBS, EFA, etc., used as part of AI training platform)[14], Azure-based managed data stores and PaaS services for Adobe Experience Platform and Commerce (multi-cloud deployment on Azure and AWS)[1][2][7][12], GCP sovereign regions for Adobe Experience Manager Managed Services (data residency-focused deployments)[5], Proprietary Adobe data services layered on top of these cloud providers (Experience Platform and Experience Cloud foundations)[6][15]
Compliance: Digital sovereignty controls leveraging AWS, Azure, and GCP sovereign regions for Adobe Experience Manager Managed Services (jurisdictional control, data residency, added encryption and controls)[5], Multi-region, high-availability architecture across AWS or Azure Availability Zones for Adobe Commerce on cloud (disaster recovery and resilience)[7], Content provenance and authenticity via Content Credentials using C2PA standards for AI-generated content (Firefly-related trust and safety)[4]
Adobe runs a **multi-cloud** architecture primarily across AWS, Azure, and selected GCP sovereign regions, with Adobe Experience Platform and Experience Cloud services deployed on both AWS and Azure, and Firefly’s GPU-heavy training and inference infrastructure built on AWS. Core data and AI workloads rely on managed storage, compute, and networking primitives from these clouds, with Adobe layering proprietary platforms such as Experience Platform, Sensei, and Firefly on top.[1][2][3][5][6][7][12][14][15]
FinTech · On-Prem · Off the LLM map (in-house non-LLM ML)
Foundational models: in-house non-LLM ML only
Stack: PostgreSQL
Compliance: PCI DSS Level 1, SOC 2, ISO 27001
Unusually for its scale, runs its own data centers (not a hyperscaler) on a single Java platform over PostgreSQL for full control of the payment path. RevenueProtect fraud scoring is in-house ML.
FinTech · AWS · Off the LLM map (in-house non-LLM ML)
Foundational models: in-house non-LLM ML only
Stack: PostgreSQL, Kafka
Compliance: SOC 2, PCI DSS
BNPL underwriting on large in-house risk ML; off the foundational-LLM map at the core.
AI/LLM · Multi-Cloud · Native (own / self-hosted weights)
Foundational models: Jamba (base / instruct family, hybrid SSM-Transformer MoE) — Native (own / self-hosted weights) (primary); Jamba 1.5 Large — Cloud-hosted (Bedrock/Vertex/Azure); Jamba 1.5 Mini — Cloud-hosted (Bedrock/Vertex/Azure); Jamba (open-source weights, Apache 2.0) — Native (own / self-hosted weights)
Stack:
Compliance:
AI21 Labs develops and serves its own Jamba family of LLMs and related services via its AI21 Studio API and private deployments, while also distributing Jamba models through third‑party clouds such as Azure and upcoming NVIDIA APIs.[5][6][7] Public materials describe deployment options across public cloud and private/on‑prem environments but do not reveal a full production stack beyond this high‑level multi‑cloud posture.[3][5][6]
Marketplace · AWS · Direct API
Foundational models: GPT-4-class — Direct API (primary)
Stack: MySQL (Vitess), HBase, Elasticsearch
Compliance: SOC 2, PCI DSS, GDPR
AWS-native marketplace that evolved a Rails monolith into services over sharded MySQL (Vitess) + HBase. Search/pricing are in-house ML. Created Apache Airflow; AI customer-support uses LLMs.
Data · Multi-Cloud · Middleware / wrapper
Foundational models: partner LLM — Middleware / wrapper (primary)
Stack: PostgreSQL
Compliance: SOC 2
Open-source data integration; connectors + AI assist via partner LLMs.
Productivity · AWS · Direct API
Foundational models: GPT-4-class — Direct API (primary); Claude — Direct API
Stack: MySQL, Redis
Compliance: SOC 2, HIPAA, GDPR
Low-code database whose 'spreadsheet-database' is backed by MySQL with heavy Redis caching on AWS. Airtable AI/Cobuilder routes to OpenAI and Anthropic.
AI/LLM · Hybrid · Native (own / self-hosted weights)
Foundational models: Pharia 1 / Pharia-1-LLM-7B — Native (own / self-hosted weights) (primary); Luminous (legacy family) — Native (own / self-hosted weights)
Stack: object
Compliance: sovereign, EU-compliant, access control, monitoring
Aleph Alpha’s current product layer appears to be PhariaAI, a sovereign enterprise stack that includes knowledge capture, development, operation, access control, and monitoring components. Public materials also show first-party API access and containerized deployment patterns, but the exact production datastore and hosting mix are not fully disclosed, so the infrastructure is best characterized as hybrid rather than purely on-prem or purely cloud.[2][5][10]
Healthcare AI · AWS · Native (own / self-hosted weights)
Foundational models: Ambience clinical (own) — Native (own / self-hosted weights) (primary); GPT — Direct API
Stack: vector
Compliance: HIPAA, SOC 2
Ambient documentation with its own clinical models plus OpenAI, grounded in encounter audio.
Analytics · AWS · Direct API
Foundational models: Anthropic Claude (and other Bedrock foundation models) — Cloud-hosted (Bedrock/Vertex/Azure) (primary); GPT-family (historical/experimental, inferred) — Direct API
Stack: Amazon S3, Amazon DynamoDB, Kafka (streaming), Redis/ElastiCache, proprietary columnar analytics store
Compliance: SOC 2 (inferred), ISO 27001 (inferred), GDPR (inferred), CCPA (inferred)
Amplitude runs a high-throughput event analytics platform on AWS with Kafka-based ingestion, DynamoDB/Redis for state and caching, and a proprietary columnar datastore optimized for behavioral analytics, plus AI Agents powered via Amazon Bedrock. Core production services are containerized and orchestrated with Kubernetes and Terraform-based infrastructure-as-code.[5][7]
Defense · Hybrid · Off the LLM map (in-house non-LLM ML)
Foundational models: in-house non-LLM ML only
Stack: Proprietary internal telemetry stores (Lattice platform), Data lake/warehouse for long-term sensor/log storage (likely cloud + on-prem, exact vendors undisclosed), Streaming/message bus layer (Kafka/Pulsar-like per system design guidance, not confirmed as production)
Compliance: ITAR (International Traffic in Arms Regulations) – required for US defense exports and autonomous weapons systems, DoD cybersecurity controls (e.g., NIST SP 800-53 / FedRAMP-aligned baselines for handling defense telemetry), SOC 2–style controls for hyperscale manufacturing and software-defined facilities (inferred from defense SaaS norms), Defense-specific secure manufacturing and export controls for Arsenal-1 and international facilities
Anduril’s production architecture is built around its proprietary **Lattice** autonomy platform as the software backbone for autonomous systems and the **Arsenal** software-defined manufacturing stack, integrating sensor fusion, decision logic, and production control across a mix of cloud and tightly controlled on‑prem deployments.[10][12][13] Public material focuses on capabilities and manufacturing rather than concrete cloud/database vendor disclosures, so specific underlying services (e.g., exact clouds and DB engines) remain intentionally opaque.[3][10][12]
AI/LLM · Multi-Cloud · Native (own / self-hosted weights)
Foundational models: Claude 3.5 family (Opus / Sonnet / Haiku variants) — Native (own / self-hosted weights) (primary); Claude 3.x / Claude Instant (legacy/previous generations) — Native (own / self-hosted weights)
Stack: PostgreSQL[13], vector databases (e.g., Qdrant)[4][13], Redis/Valkey (cache)[5][9][13], object storage (cloud blob storage, e.g., S3/GCS)[1][13]
Compliance: SOC 2 (inferred, typical for enterprise AI SaaS; not explicitly confirmed)[2][3], ISO 27001 (inferred, typical for scale enterprise workloads; not explicitly confirmed)[2][3], privacy-preserving sandbox perimeter for agents (policy/architecture control rather than formal cert)[2][11], network, application, and API security enforcement via distributed infrastructure (Akamai)[12]
Anthropic runs **Claude** and related services on a safety-first **multi-cloud** compute fabric spanning AWS Trainium2, Google TPUv7 and NVIDIA GPUs, fronted by Kubernetes‑based microservices (API gateways, orchestration, rate limiting, caching, and safety filters), with state held in PostgreSQL, vector stores, Redis, and cloud object storage.[1][10][13] Production offerings like Claude API and Managed Agents expose a fully managed orchestration and agent runtime, while emerging self‑hosted sandboxes move tool execution into customer infrastructure but keep Claude inference, routing, and session state on Anthropic’s cloud.[2][3][11]
Productivity · Hybrid · Direct API
Foundational models: GPT-4 class — Direct API (primary)
Stack: MySQL, Amazon Aurora (MySQL-compatible), Snowflake or similar cloud data warehouse (inferred)
Compliance: SOC 2 (inferred), ISO 27001 (inferred), GDPR, CCPA
Asana runs a large-scale, stateful, microservice-style backend with autoscaling infrastructure and a dedicated Production Infrastructure pod for observability and platform reliability, layered behind a multi-region, hybrid cloud deployment. Its data stack separates transactional MySQL/Aurora stores from analytical event pipelines and a cloud data warehouse used by growth and analytics teams.
Telecom · Hybrid · Native (own / self-hosted weights)
Foundational models: own ML — Native (own / self-hosted weights) (primary); partner LLM — Cloud-hosted (Bedrock/Vertex/Azure)
Stack: own
Compliance: SOC 2
Telecom; in-house network ML plus partner LLMs (Ask AT&T on Azure OpenAI) for employees.
DevTools · AWS · Cloud-hosted (Bedrock/Vertex/Azure)
Foundational models: GPT-4-class — Cloud-hosted (Bedrock/Vertex/Azure) (primary)
Stack: PostgreSQL, DynamoDB, Kafka
Compliance: SOC 2, ISO 27001, FedRAMP (Moderate)
Jira/Confluence migrated to an AWS-native multi-tenant platform (Micros) on PostgreSQL + DynamoDB. Atlassian Intelligence wraps OpenAI behind a tenancy/governance layer.
Design/Eng · AWS · Native (own / self-hosted weights)
Foundational models: Autodesk AI (own) — Native (own / self-hosted weights) (primary); partner LLM — Cloud-hosted (Bedrock/Vertex/Azure)
Stack: PostgreSQL, own
Compliance: SOC 2, ISO 27001
Design/CAD cloud with its own generative-design models plus partner LLMs for assistants.
HR · AWS · Direct API
Foundational models: GPT — Direct API (primary)
Stack: MySQL
Compliance: SOC 2
SMB HR platform; AI assist on OpenAI over employee data.
AI-infra · Multi-Cloud · Native (own / self-hosted weights)
Foundational models: Llama — Native (own / self-hosted weights) (primary)
Stack: object
Compliance: SOC 2, HIPAA
Model-inference platform serving open models (Llama, etc.) on autoscaling GPU infra.
Biotech SaaS · AWS · Direct API
Foundational models: GPT — Direct API (primary)
Stack: PostgreSQL
Compliance: SOC 2, GxP
R&D cloud for life sciences; AI assistants on OpenAI over experimental data.
Retail · GCP · Cloud-hosted (Bedrock/Vertex/Azure)
Foundational models: Gemini — Cloud-hosted (Bedrock/Vertex/Azure) (primary)
Stack: own, MySQL
Compliance: SOC 2, PCI DSS
Retailer using Google Cloud Gemini for support + shopping assistants over its catalog.
FinTech · AWS · Direct API
Foundational models: Claude — Direct API (primary); GPT — Direct API
Stack: MySQL
Compliance: SOC 2, PCI DSS
Cash App Money Bot + Square Managerbot run on Anthropic Claude (Sonnet) and OpenAI GPT via Block’s open-source goose agent framework (Claude is goose’s default on Databricks).
Banking (EU) · Hybrid · Native (own / self-hosted weights)
Foundational models: Mistral — Direct API (primary); own ML — Native (own / self-hosted weights)
Stack: own
Compliance: SOC 2, GDPR
European bank; in-house risk ML plus Mistral for EU-sovereign generative AI.
Travel · Hybrid · Native (own / self-hosted weights)
Foundational models: GPT — Direct API (primary); own ML — Native (own / self-hosted weights)
Stack: own, Cassandra
Compliance: SOC 2, PCI DSS
Travel marketplace; huge own ranking/pricing ML plus OpenAI for the AI Trip Planner.
Storage · AWS · Direct API
Foundational models: GPT — Direct API (primary); Gemini — Cloud-hosted (Bedrock/Vertex/Azure); Claude — Direct API
Stack: MySQL, HBase
Compliance: SOC 2, FedRAMP, HIPAA
Box AI is explicitly multi-model — routes across OpenAI, Google and Anthropic over enterprise content.
FinTech · AWS · Direct API
Foundational models: GPT-4-class — Direct API (primary); Claude — Direct API
Stack: PostgreSQL, Kafka
Compliance: SOC 2, PCI DSS
AWS-based corporate-card/spend platform on PostgreSQL with Kafka eventing. Brex Assistant uses OpenAI/Anthropic over its own financial data and an agentic spend layer.
FinTech (EU) · Hybrid · Direct API
Foundational models: GPT — Direct API (primary)
Stack: PostgreSQL
Compliance: PCI DSS, GDPR
European neobank; Finn AI assistant on OpenAI over account data.
Social/Consumer · Hybrid · Native (own / self-hosted weights)
Foundational models: Doubao (ByteDance foundational LLM, Chinese & international markets) — Native (own / self-hosted weights) (primary); Monolith (large-scale recommendation / ranking model family) — Native (own / self-hosted weights); Astra-Global (multimodal large language model for robot navigation) — Native (own / self-hosted weights)
Stack: ByteHouse (cloud-native data warehouse, ClickHouse-derived)[6], ByteHTAP (internal HTAP system)[9], HDFS (offline storage in recsys pipelines)[8], Kafka / ByteMQ (messaging + streaming backbone, storage/compute separated)[1][8], Redis / RedisCell (caching, counters)[10], MySQL (common OLTP store, prior record corroborated)[8]
Compliance: Data governance with access controls, retention policies, privacy safeguards in streaming architecture[3], Strict compliance boundaries for global-scale systems (region-specific requirements, moderation, privacy) inferred from design guides[5][13]
ByteDance runs a large-scale hybrid architecture with proprietary cloud-native data and streaming systems (ByteHouse, ByteHTAP, ByteMQ/Flux) atop commodity storage and compute, tightly integrated with recommendation and multimodal AI stacks for products like TikTok/Douyin and robotics.[1][3][6][7][8][9] Real-time streaming (Kafka/ByteMQ), HTAP, and on-device AI infrastructure (Pitaya) form the core substrate for online training, inference, and global content delivery.[1][2][3][8][10][11]
Design · AWS · Native (own / self-hosted weights)
Foundational models: Leonardo (own, acquired) — Native (own / self-hosted weights) (primary); GPT-4-class — Direct API
Stack: MySQL, DynamoDB
Compliance: SOC 2, ISO 27001, GDPR
Massive consumer design platform on AWS with MySQL + DynamoDB and a WASM-accelerated rendering pipeline. Magic Studio blends in-house generative models with partners (incl. acquired Leonardo.ai).
FinTech · AWS · Off the LLM map (in-house non-LLM ML)
Foundational models: in-house non-LLM ML only
Stack: PostgreSQL
Compliance:
Carta operates a proprietary, cloud-based equity and private capital management platform, most likely hosted on AWS using a relational database such as PostgreSQL. Public information focuses on product capabilities rather than disclosing detailed production architecture, so deeper specifics cannot be reliably verified.
AI-hardware · On-Prem · Native (own / self-hosted weights)
Foundational models: Llama — Native (own / self-hosted weights) (primary)
Stack: object
Compliance: SOC 2
Designs wafer-scale AI chips and serves open models (Llama) at record speed on its own systems.
DevSecOps · Multi-Cloud · Off the LLM map (in-house non-LLM ML)
Foundational models: in-house non-LLM ML only
Stack: object
Compliance: SOC 2, FedRAMP
Secure-by-default container images + supply-chain security; off the foundational-LLM map at the core.
Consumer AI · GCP · Native (own / self-hosted weights)
Foundational models: Kaiju Large (110B) / current C‑series production models — Native (own / self-hosted weights) (primary); Kaiju Medium (34B) — Native (own / self-hosted weights); Kaiju Small (13B) — Native (own / self-hosted weights)
Stack: AlloyDB for PostgreSQL[2][4], Cloud Spanner[4], Google Cloud Memorystore for Redis Cluster[15], Object storage (Google Cloud Storage, implied for model checkpoints and data)[3][7], Analytics warehouse (BigQuery)[10]
Compliance: Unknown (no explicit public certifications like SOC 2, ISO 27001, HIPAA found in retrieved sources; standard GCP security/compliance underlies infra)
Character.AI runs a proprietary end‑to‑end stack on Google Cloud, training and serving its own Kaiju/C‑series LLMs on GPU and TPU infrastructure (including H100 clusters) with heavy inference optimization, backed by managed databases (AlloyDB, Spanner), Redis, and BigQuery on GCP.[1][2][3][4][7][10][11][15] The production platform uses dense transformer models (Kaiju Small/Medium/Large) trained and hosted on GCP, with int8-optimized inference and custom serving infrastructure for low‑latency conversational workloads.[3][7][11]
FinTech · Hybrid · Off the LLM map (in-house non-LLM ML)
Foundational models: in-house non-LLM ML only
Stack: own, PostgreSQL
Compliance: PCI DSS, SOC 2
Global payments processor on its own infrastructure; large in-house fraud/risk ML; off the foundational-LLM map at the core.
EdTech · AWS · Native (own / self-hosted weights)
Foundational models: own ML — Native (own / self-hosted weights) (primary); GPT — Direct API
Stack: MySQL
Compliance: SOC 2
Study platform blending its own content ML with OpenAI for its learning assistant.
FinTech · AWS · Cloud-hosted (Bedrock/Vertex/Azure)
Foundational models: Claude (family: Claude 2/3, exact variant not specified) — Cloud-hosted (Bedrock/Vertex/Azure) (primary)
Stack: Snowflake, MySQL, DynamoDB
Compliance:
Chime runs a proprietary fintech platform on AWS, using S3, Airflow and Spark-based data pipelines that write to Snowflake and then propagate to MySQL and DynamoDB for operational and analytics use.[1][5] AI capabilities, including call summarization, are integrated via Amazon Bedrock into this core architecture rather than being a standalone LLM-native product.[5]
Database · Multi-Cloud · Middleware / wrapper
Foundational models: partner LLM — Middleware / wrapper (primary)
Stack: ClickHouse
Compliance: SOC 2, GDPR
Open-source columnar OLAP DB; AI features plug partner LLMs over its analytics engine.
Legaltech · AWS · Direct API
Foundational models: GPT — Direct API (primary)
Stack: PostgreSQL
Compliance: SOC 2
Legal practice-management cloud; Clio Duo AI on OpenAI over matter/document data.
Infrastructure · On-Prem · Native (own / self-hosted weights)
Foundational models: Llama / Mistral (open models via Workers AI) — Native (own / self-hosted weights) (primary)
Stack: D1 (SQLite), Durable Objects (SQLite-backed), Workers KV, R2, Cloudflare Queues (Durable Objects-backed), Internal analytics/logging stores (likely ClickHouse/columnar, not publicly detailed), Postgres (limited/internal use, e.g., legacy/control-plane; inferred)
Compliance: Zero Trust architecture, Service isolation, Least privilege access, Anycast edge security model, Secure access service edge (SASE) via Cloudflare One, Strong credential-based access control
Cloudflare runs its own Linux-based servers and private backbone as a globally anycasted edge and core network, with most products (CDN, Zero Trust, R2, KV, Queues, Agents, Containers) built on an internal proxy chain and the Workers/Durable Objects developer platform.[3][9][10] Control planes and schedulers (for Containers, Workers AI, agents, etc.) are themselves implemented on Cloudflare’s stack, indicating a predominantly self-hosted, on-prem architecture rather than dependence on hyperscale public clouds.[7][8][10]
AI-coding · AWS · Direct API
Foundational models: Claude — Direct API (primary); GPT — Direct API
Stack: PostgreSQL, vector
Compliance: SOC 2
Autonomous SWE agent (Devin) orchestrating Claude + GPT over its own long-horizon planning and code-execution sandboxes.
AI/LLM · Multi-Cloud · Native (own / self-hosted weights)
Foundational models: Command (own) — Native (own / self-hosted weights) (primary)
Stack: object storage, vector store
Compliance: SOC 2, GDPR
Enterprise-focused LLM lab training its own Command/Embed/Rerank models, served across clouds (Google Cloud, Oracle, AWS) and deployable in-VPC for data-sensitive customers. North platform targets RAG.
FinTech · AWS · Direct API
Foundational models: GPT-4-class — Direct API (primary)
Stack: MongoDB, PostgreSQL, DynamoDB
Compliance: SOC 2, ISO 27001, SOC 1
AWS-native crypto exchange on MongoDB/PostgreSQL with DynamoDB for high-throughput paths; strong HSM/key-management and on-chain infra. Uses in-house risk ML and LLMs for support.
Telecom/Media · Hybrid · Native (own / self-hosted weights)
Foundational models: own ML — Native (own / self-hosted weights) (primary); partner LLM — Cloud-hosted (Bedrock/Vertex/Azure)
Stack: own
Compliance: SOC 2
Cable/media; own voice-remote + network ML plus partner LLMs for assistants.
Proptech · AWS · Direct API
Foundational models: GPT — Direct API (primary)
Stack: PostgreSQL
Compliance: SOC 2
Real-estate brokerage platform; agent AI tools on OpenAI over listing/CRM data.
Data Platform · Multi-Cloud · Middleware / wrapper
Foundational models: partner LLMs — Middleware / wrapper (primary)
Stack: Apache Kafka, ksqlDB
Compliance: SOC 2, HIPAA, PCI DSS, ISO 27001
Managed Apache Kafka (Confluent Cloud) deployed across AWS/Azure/GCP with the Kora cloud-native engine; adds stream processing via Flink. Built around open Kafka with proprietary cloud tooling.
AI/LLM · Multi-Cloud · Native (own / self-hosted weights)
Foundational models: Contextual RAG models (own) — Native (own / self-hosted weights) (primary)
Stack: vector
Compliance: SOC 2
RAG-native enterprise platform training its own grounded language models.
AI-infra · On-Prem · Off the LLM map (in-house non-LLM ML)
Foundational models: in-house non-LLM ML only
Stack: object
Compliance: SOC 2, ISO 27001
GPU hyperscaler renting NVIDIA capacity to AI labs; infrastructure layer, runs customers models not its own.
EdTech · AWS · Direct API
Foundational models: GPT — Direct API (primary)
Stack: MySQL
Compliance: SOC 2
Online learning; Coursera Coach + course generation on OpenAI over its catalog.
Robotics · AWS · Native (own / self-hosted weights)
Foundational models: RFM-1 (own robotics foundation) — Native (own / self-hosted weights) (primary)
Stack: object
Compliance: SOC 2
Warehouse robotics on RFM-1, its own robotics foundation model (team since joined Amazon).
CX AI · GCP · Native (own / self-hosted weights)
Foundational models: Cresta (own) — Native (own / self-hosted weights) (primary); GPT — Direct API
Stack: vector, PostgreSQL
Compliance: SOC 2
Contact-center AI with its own real-time coaching models plus partner LLMs over call data.
Cybersecurity · AWS · Cloud-hosted (Bedrock/Vertex/Azure)
Foundational models: Charlotte AI (CrowdStrike internal orchestration layer over multiple fine-tuned LLMs) — Cloud-hosted (Bedrock/Vertex/Azure) (primary); Multiple fine-tuned LLMs for security use cases (names not publicly disclosed) — Cloud-hosted (Bedrock/Vertex/Azure); Embedding models for semantic search over Falcon data — Cloud-hosted (Bedrock/Vertex/Azure)
Stack: Distributed NoSQL store for Threat Graph and telemetry (proprietary, AWS-hosted)[5], Amazon S3 for large-scale object storage of security telemetry and detections[5], Amazon OpenSearch Service (including vector store for embeddings)[5], Amazon EMR (data processing over scalable storage, likely HDFS/S3-backed)[5], Streaming/ETL pipelines for telemetry ingestion (precise tech not publicly confirmed; Kafka/Kinesis inferred by architecture)[5]
Compliance: SOC 2 Type I[3][14], SOC 2 Type II[3][14], ISO/IEC 27001:2022[3][14], ISO/IEC 27017[3], ISO/IEC 42001:2023 (AI management system)[3][14], ISO 22301:2019[3], PCI DSS v4.0.1[3][10], FedRAMP High authorization for Falcon Platform for Government[3][14], GovRAMP (CrowdStrike public-sector program)[3], HIPAA (for applicable offerings)[10], CSA STAR Level 2[3][10], GDPR-aligned controls and attestations[10], NIST-aligned controls (including TX-RAMP listing)[10]
The Falcon platform is a cloud-native, single lightweight-agent architecture built on AWS that streams endpoint and cloud telemetry into CrowdStrike’s distributed NoSQL and object stores, processed by large-scale analytics and ML pipelines (e.g., EMR, SageMaker) to power detections, Threat Graph, and higher-level services like Next-Gen SIEM and Charlotte AI.[2][4][5][6][13][15] It operates as a multi-tenant SaaS service hosted on AWS regions, integrating deeply with AWS-native services and APIs while protecting workloads across on-prem, AWS, GCP, Azure, and other environments.[2][4][6][7][8][11][12][15]
Autonomous · GCP · Native (own / self-hosted weights)
Foundational models: Cruise driving ML (own) — Native (own / self-hosted weights) (primary)
Stack: own
Compliance: SOC 2
GM self-driving running its own perception/planning ML; restructured toward personal-vehicle autonomy.
DevTools · Multi-Cloud · Native (own / self-hosted weights)
Foundational models: Cursor proprietary code models (fine-tuned frontier family) — Native (own / self-hosted weights) (primary); Claude family — Direct API; GPT family — Direct API
Stack: turbopuffer (multi-tenant DB for encrypted files and Merkle trees), Pinecone (vector database for documentation embeddings), Warpstream/Kafka-compatible streaming, Redis/ElastiCache (caching and session state)
Compliance: end-to-end encryption of workspace files, Merkle-tree based integrity for workspace sync, enterprise-grade security posture, privacy modes for code and data handling
Cursor runs a multi-cloud, event-driven architecture with most CPU/backend services on AWS, major GPU inference clusters on Azure plus newer GPU clouds, and specialized AI workloads on GCP, fronted by stateless gateways, queues, and GPU worker fleets for LLM inference and agents.[6][10] Workspace data and embeddings flow through turbopuffer, Pinecone, and Kafka-compatible streams, with Temporal-based orchestration for agent workflows and large multi-GPU clusters (including Fireworks AI-hosted proprietary fine-tuned models) serving code-focused LLMs.[4][6][7][8][10]
Cybersecurity · Hybrid · Native (own / self-hosted weights)
Foundational models: Self-Learning AI (own) — Native (own / self-hosted weights) (primary)
Stack: own
Compliance: SOC 2, ISO 27001
Self-Learning AI built in-house; models the network rather than calling an LLM.
Data Platform · Multi-Cloud · Native (own / self-hosted weights)
Foundational models: DBRX — Native (own / self-hosted weights) (primary); Llama family (e.g., Llama 2 / Llama 3) — Native (own / self-hosted weights); Other open-source foundation models (e.g., MPT, Falcon, etc.) used via Databricks Model Serving — Native (own / self-hosted weights); Proprietary frontier models (e.g., OpenAI GPT family) when customers integrate via external APIs from notebooks/jobs — Middleware / wrapper
Stack: Delta Lake (transactional lakehouse tables on cloud object storage)[6][10], Apache Spark (distributed compute engine for ETL/ML/streaming)[11][12], Unity Catalog (centralized governance and metadata layer)[1][5], Cloud object storage (Amazon S3, Azure Data Lake Storage, Google Cloud Storage) as primary data store[6][4][12], Medallion (Bronze/Silver/Gold) data modeling pattern on top of Delta tables[2][12]
Compliance: Separation of control plane (Databricks-managed backend services) and compute/data plane (customer cloud account) for isolation and governance[3][4][7][11], Network isolation via VPC/VNet injection, CIDR planning, and private connectivity for data plane resources[1][5][7], Workspace- and catalog-level access control with Unity Catalog (role-based access, row/column-level policies)[1][5][10], Encryption of data in transit and at rest via underlying cloud provider primitives (KMS/Key Vault, TLS)[5][10], Governance features including lineage tracking, masking, and auditability built into lakehouse stack[1][10]
Databricks runs a two-layer architecture where a Databricks-managed control plane hosts the UI, APIs, metadata, and orchestration services, while customer workloads execute in a compute/data plane inside the customer’s AWS, Azure, or GCP account (or Databricks’ serverless account) against cloud object storage using Spark and Delta Lake.[3][4][9][11] Around this core, Databricks positions an open-core lakehouse stack (Delta Lake, Spark, MLflow) with proprietary governance, serverless, and AI platform services, integrated into medallion-style production patterns.[1][2][6][12]
Observability · Multi-Cloud · Cloud-hosted (Bedrock/Vertex/Azure)
Foundational models: Claude — Cloud-hosted (Bedrock/Vertex/Azure) (primary)
Stack: PostgreSQL, Cassandra, Kafka, Elasticsearch
Compliance: SOC 2, HIPAA, FedRAMP, ISO 27001
Runs primarily on AWS with significant GCP presence; ingests trillions of points/day through Kafka into Cassandra and custom time-series stores. Bits AI / Watchdog use in-house anomaly ML plus LLMs.
FinTech · AWS · Direct API
Foundational models: GPT — Direct API (primary)
Stack: PostgreSQL
Compliance: SOC 2
Neobank; in-house underwriting ML plus OpenAI for support/assistant.
Data · Multi-Cloud · Middleware / wrapper
Foundational models: partner LLM — Middleware / wrapper (primary)
Stack: PostgreSQL
Compliance: SOC 2
Analytics-engineering standard (dbt); Copilot AI is model-agnostic over the semantic layer.
CX AI · AWS · Direct API
Foundational models: Claude — Direct API (primary); GPT — Direct API
Stack: vector, PostgreSQL
Compliance: SOC 2
AI customer-support agents orchestrating Claude + GPT over a company knowledge base.
HR · AWS · Direct API
Foundational models: Claude — Direct API (primary); GPT-4-class (unspecified variant) — Direct API
Stack: PostgreSQL, SQL (unspecified relational), NoSQL (unspecified)
Compliance: SOC 2 (inferred, typical for payroll/HR SaaS), ISO 27001 (inferred), GDPR, Global payroll & tax compliance (multi-country)
Deel runs a proprietary, multi-tenant SaaS platform (app.deel.com / api.letsdeel.com) on AWS with separate sandbox and production environments, using relational and NoSQL databases behind REST APIs consumed by customers and integrations.[3][10][4] AI is incorporated primarily at the application and workflow layer to automate compliance, payroll, and HR processes, rather than as a foundational LLM-native architecture.[6][4]
AI/Translation · Hybrid · Native (own / self-hosted weights)
Foundational models: DeepL models (own) — Native (own / self-hosted weights) (primary)
Stack: own
Compliance: GDPR, SOC 2, ISO 27001
European MT leader training its own translation + writing models on its own EU infrastructure.
FoodTech · AWS · Native (own / self-hosted weights)
Foundational models: own ML — Native (own / self-hosted weights) (primary); GPT — Direct API
Stack: own, PostgreSQL
Compliance: PCI DSS, SOC 2
Food delivery; own dispatch/ETA ML plus OpenAI for support.
Hardware/Enterprise · Hybrid · Native (own / self-hosted weights)
Foundational models: Llama — Native (own / self-hosted weights) (primary); Cohere — Cloud-hosted (Bedrock/Vertex/Azure)
Stack: own
Compliance: SOC 2, ISO 27001
Dell AI Factory packages on-prem open models (Llama) + partners (Cohere) for enterprise/sovereign deployment.
Communications · GCP · Direct API
Foundational models: GPT-4-class — Direct API (primary)
Stack: ScyllaDB (ex-Cassandra), MongoDB, Elixir/Rust services
Compliance: SOC 2, GDPR
Runs on Google Cloud; publicly migrated trillions of messages from Cassandra to ScyllaDB and uses Elixir + Rust for real-time gateway/voice. AI features (Clyde and successors) built on OpenAI.
Media · Hybrid · Native (own / self-hosted weights)
Foundational models: own ML — Native (own / self-hosted weights) (primary); partner LLM — Cloud-hosted (Bedrock/Vertex/Azure)
Stack: own
Compliance: SOC 2
Media/parks; large in-house recommendation/VFX ML plus partner LLMs for assistants; cautious GenAI posture.
Healthcare (EU) · Hybrid · Direct API
Foundational models: Mistral — Direct API (primary)
Stack: own, PostgreSQL
Compliance: GDPR, HDS
European health-booking platform; AI assistants built on Mistral for EU data-sovereignty.
Marketplace · AWS · Direct API
Foundational models: GPT — Direct API (primary)
Stack: Aurora, CockroachDB
Compliance: SOC 2, PCI DSS
In-house logistics ML plus OpenAI for support + merchant tooling.
Storage · Hybrid · Direct API
Foundational models: GPT-4-class — Direct API (primary)
Stack: MySQL (Edgestore), custom blob store
Compliance: SOC 2, HIPAA, FedRAMP, ISO 27001
Famously repatriated bulk file storage off AWS onto its own 'Magic Pocket' exabyte infrastructure while keeping some cloud for edge/compute. Metadata in sharded MySQL (Edgestore). Dropbox Dash AI uses OpenAI.
EdTech · AWS · Direct API
Foundational models: GPT — Direct API (primary)
Stack: MySQL, Redis
Compliance: SOC 2, GDPR
Birdbrain in-house ML sets lesson difficulty; GPT-4 (Direct API) powers Roleplay + Explain-my-Answer.
Search/Data · Multi-Cloud · Native (own / self-hosted weights)
Foundational models: ELSER (own) — Native (own / self-hosted weights) (primary); partner LLM — Middleware / wrapper
Stack: Elasticsearch
Compliance: SOC 2, GDPR
Ships ELSER, its own retrieval model, and plugs partner LLMs for RAG.
Gaming · Hybrid · Native (own / self-hosted weights)
Foundational models: own ML — Native (own / self-hosted weights) (primary); partner LLM — Cloud-hosted (Bedrock/Vertex/Azure)
Stack: own
Compliance: SOC 2
Game publisher; large in-house animation/matchmaking ML plus partner LLMs for tooling.
AI/Voice · Multi-Cloud · Native (own / self-hosted weights)
Foundational models: Eleven v3 speech synthesis — Native (own / self-hosted weights) (primary); Multilingual v2 speech synthesis — Native (own / self-hosted weights); Flash v2.5 speech synthesis — Native (own / self-hosted weights); Turbo v2.5 speech synthesis — Native (own / self-hosted weights); Eleven Music — Native (own / self-hosted weights); Scribe v2 Realtime (speech processing / transcription) — Native (own / self-hosted weights)
Stack: PostgreSQL, object storage
Compliance: SOC 2 (inferred), ISO 27001 (inferred)
ElevenLabs exposes proprietary speech, audio, and agent models via HTTPS REST, SSE streaming, and WebSocket APIs behind api.elevenlabs.io, backed by credit-based authentication and multi-tenant SaaS infrastructure designed for high concurrency and enterprise scalability.[1][7][9] Core value is provided by in-house deep learning voice and audio models rather than a foundational LLM as the primary engine.[2][5]
Gaming · Hybrid · Native (own / self-hosted weights)
Foundational models: own ML (MetaHuman/animation) — Native (own / self-hosted weights) (primary)
Stack: own
Compliance: SOC 2, COPPA
Unreal Engine + Fortnite; in-house generative/animation ML (MetaHuman) on its own infra.
Healthcare · Hybrid · Cloud-hosted (Bedrock/Vertex/Azure)
Foundational models: GPT — Cloud-hosted (Bedrock/Vertex/Azure) (primary)
Stack: InterSystems Cache, own
Compliance: HIPAA, SOC 2
Dominant EHR on its own InterSystems stack; generative features run on Azure OpenAI via the Microsoft partnership.
Biotech · AWS · Native (own / self-hosted weights)
Foundational models: ESM3-large (98B frontier protein LLM family) — Native (own / self-hosted weights) (primary); ESM3-medium (7B) — Native (own / self-hosted weights); ESM3-small-open (1.4B open version) — Native (own / self-hosted weights); ESM C 600M / 300M protein models — Native (own / self-hosted weights)
Stack: object storage (S3 or equivalent, inferred), domain-specific biological data stores (inferred)
Compliance:
EvolutionaryScale operates ESM3 as a proprietary, GPU-accelerated protein LLM stack on AWS, exposing commercial access primarily via AWS SageMaker, AWS HealthOmics, and soon Amazon Bedrock, while distributing a smaller open version of ESM3 and earlier ESM models with code and some weights on GitHub and Hugging Face for non-commercial and open use.[3][5][7][11] Internal architecture details beyond this cloud/HPC and PyTorch-centric GPU cluster setup are not publicly documented.[3][10][12]
Travel · AWS · Direct API
Foundational models: GPT — Direct API (primary)
Stack: own, MySQL
Compliance: SOC 2, PCI DSS
Travel platform; early OpenAI customer (Romie / trip planning) over its own travel-graph ML.
Wholesale · AWS · Native (own / self-hosted weights)
Foundational models: own ML — Native (own / self-hosted weights) (primary); GPT — Direct API
Stack: MySQL
Compliance: SOC 2
Wholesale marketplace; own recommendation ML plus OpenAI for merchandising assistants.
Design · AWS · Direct API
Foundational models: GPT-4-class — Direct API (primary)
Stack: PostgreSQL, DynamoDB
Compliance: SOC 2, GDPR
Browser-based design tool whose moat is a custom C++ rendering/document engine shipped to the browser via WebAssembly. Metadata in sharded PostgreSQL on AWS; multiplayer via bespoke real-time sync servers.
Robotics · Hybrid · Native (own / self-hosted weights)
Foundational models: Helix (current VLA stack: 7B vision-language + high-frequency visuomotor policy) — Native (own / self-hosted weights) (primary)
Stack: proprietary manufacturing/operations data infrastructure (MES/PLM/ERP/WMS stack)[2], custom telemetry and training data pipelines for Helix VLA (vision, language, proprioception)[7][9]
Compliance: Not publicly specified; likely standard enterprise practices but no explicit SOC2/ISO27001/IEC 62443 claims in accessible materials[1][2][7][9]
Figure’s production architecture is centered on its in-house **Helix** vision-language-action stack running on vertically integrated humanoid hardware and a proprietary BotQ manufacturing/software infrastructure, with robots streaming high-bandwidth data for continuous training and fleet improvement[1][2][7][9]. Core control and reasoning have moved off external LLMs to fully in-house Helix models, with cloud services used as supporting infrastructure rather than as foundational AI engines[7][9].
AI Infra · Multi-Cloud · Native (own / self-hosted weights)
Foundational models: Llama — Native (own / self-hosted weights) (primary); Mistral — Native (own / self-hosted weights)
Stack: object
Compliance: SOC 2
Fast inference for open models on its own serving stack.
Data · Multi-Cloud · Middleware / wrapper
Foundational models: partner LLM — Middleware / wrapper (primary)
Stack: PostgreSQL
Compliance: SOC 2, HIPAA
Managed data movement; AI features via partner LLMs over pipeline metadata.
Logistics · AWS · Native (own / self-hosted weights)
Foundational models: own ML — Native (own / self-hosted weights) (primary); GPT — Direct API
Stack: PostgreSQL, own
Compliance: SOC 2
Freight forwarding; in-house logistics ML plus OpenAI for ops assistants.
E-Commerce (India) · GCP · Native (own / self-hosted weights)
Foundational models: own ML — Native (own / self-hosted weights) (primary); Gemini — Cloud-hosted (Bedrock/Vertex/Azure)
Stack: own, MySQL
Compliance: PCI DSS, SOC 2
Indian commerce giant (Walmart-owned); own recommendation ML plus Gemini-based shopping assistants.
FinTech (Africa) · AWS · Native (own / self-hosted weights)
Foundational models: own ML — Native (own / self-hosted weights) (primary); GPT — Direct API
Stack: PostgreSQL
Compliance: PCI DSS, SOC 2
Pan-African payments; in-house fraud ML plus OpenAI for support.
SaaS (India) · AWS · Native (own / self-hosted weights)
Foundational models: Freddy (own) — Native (own / self-hosted weights) (primary); GPT — Direct API
Stack: MySQL
Compliance: SOC 2, ISO 27001
CX/ITSM suite; Freddy AI blends its own models with OpenAI over customer data.
IT (Japan) · Hybrid · Native (own / self-hosted weights)
Foundational models: Takane (own) — Native (own / self-hosted weights) (primary); Cohere — Cloud-hosted (Bedrock/Vertex/Azure)
Stack: own
Compliance: ISO 27001
Builds Takane, its own enterprise Japanese LLM (with Cohere), for regulated/government use.
Biotech · AWS · Native (own / self-hosted weights)
Foundational models: own bio foundation models — Native (own / self-hosted weights) (primary)
Stack: own
Compliance: SOC 2
Cell-programming foundry training its own biological/sequence models over massive lab data.
DevTools · Azure · Cloud-hosted (Bedrock/Vertex/Azure)
Foundational models: GitHub Copilot (GPT-4-class) — Cloud-hosted (Bedrock/Vertex/Azure) (primary)
Stack: MySQL (Vitess), Redis, Elasticsearch
Compliance: SOC 2, ISO 27001, PCI DSS
GitHub runs on a large-scale, multi-service production platform hosted on Microsoft Azure, with operational patterns documented by GitHub/Microsoft materials and independent technical writeups. The data layer is commonly described as MySQL at scale with Vitess plus Redis and Elasticsearch, while GitHub also publishes substantial open-source infrastructure and developer tooling.
DevTools · GCP · Off the LLM map (in-house non-LLM ML)
Foundational models: in-house non-LLM ML only
Stack: PostgreSQL, Redis, Elasticsearch
Compliance: SOC2, ISO 27001, FedRAMP, HIPAA, GDPR
GitLab.com is primarily hosted on Google Cloud Platform in the us-east1 region, with most application services running on multiple Kubernetes clusters and core datastores (PostgreSQL, Gitaly, Redis, Elasticsearch) running outside Kubernetes. Secret management uses Google KMS for GCP services and Chef Encrypted Data Bags for other host secrets.
Enterprise Search · GCP · Direct API
Foundational models: GPT — Direct API (primary); Claude — Direct API; Gemini — Cloud-hosted (Bedrock/Vertex/Azure)
Stack: vector, Elasticsearch
Compliance: SOC 2, ISO 27001
Enterprise assistant that lets customers pick across OpenAI, Anthropic and Google over a permissions-aware index.
Super-app · AWS · Native (own / self-hosted weights)
Foundational models: own ML — Native (own / self-hosted weights) (primary); GPT — Direct API
Stack: own, MySQL
Compliance: SOC 2, PCI DSS
SEA super-app; large in-house logistics/fraud ML plus OpenAI for assistants.
Observability · Multi-Cloud · Direct API
Foundational models: GPT-4 class models (Grafana Assistant backend) — Direct API (primary)
Stack: Grafana Mimir, Grafana Loki, Grafana Tempo, Prometheus, Object storage (AWS S3, GCS, Azure Blob)
Compliance: SOC 2, ISO 27001, GDPR, HIPAA (partial/BAA for specific offerings)
Grafana Cloud runs a SaaS observability platform built on its open source engines (Mimir, Loki, Tempo, Prometheus) deployed across major public clouds with object storage backends like S3/GCS/Azure Blob for long‑term data. The core Grafana server is a Go backend with a React/TypeScript frontend that acts as a visualization and control plane over these distributed storage and query services.
Productivity · AWS · Native (own / self-hosted weights)
Foundational models: own NLP (own) — Native (own / self-hosted weights) (primary); GPT — Direct API
Stack: PostgreSQL
Compliance: SOC 2, GDPR
Long-standing in-house NLP models plus GPT for generative rewriting.
Productivity AI · AWS · Direct API
Foundational models: Claude — Direct API (primary)
Stack: PostgreSQL, vector
Compliance: SOC 2
AI meeting-notes app built primarily on Anthropic Claude over transcribed audio.
AI Hardware · On-Prem · Native (own / self-hosted weights)
Foundational models: Llama 3 70B (and related Llama-family variants) — Native (own / self-hosted weights) (primary)
Stack:
Compliance:
Groq builds and operates its own on-premise inference infrastructure around custom Language Processing Units (LPUs) and GroqChip hardware, exposing this via stateless, multi-region HTTP/gRPC APIs used by customers’ gateways, queues, and orchestrators.[1][3][6][7] The production patterns described in public materials place Groq as a dedicated low-latency inference backend behind customer-managed proxies, Redis queues, and Kubernetes-based orchestrators, rather than as a full cloud stack provider.[1][3][4][6]
HR/Payroll · AWS · Direct API
Foundational models: GPT — Direct API (primary)
Stack: MySQL
Compliance: SOC 2
SMB payroll/HR; AI assistant features on OpenAI over payroll/benefits data.
Legal AI · Azure · Cloud-hosted (Bedrock/Vertex/Azure)
Foundational models: GPT — Cloud-hosted (Bedrock/Vertex/Azure) (primary)
Stack: PostgreSQL, vector
Compliance: SOC 2
Legal-domain assistant built on OpenAI via Azure with bespoke legal retrieval.
DevTools · AWS · Off the LLM map (in-house non-LLM ML)
Foundational models: in-house non-LLM ML only
Stack: PostgreSQL
Compliance: SOC 2, ISO 27001, FedRAMP
Now part of IBM; HashiCorp Cloud Platform (HCP) for Terraform/Vault/Consul runs on AWS. Core tools are widely-adopted open-source (BSL-licensed) infra-as-code and secrets management.
Defense · Hybrid · Direct API
Foundational models: Mistral large language / vision‑language‑action models (family) — Direct API (primary)
Stack: PostgreSQL (inferred from typical modern Python/C++ stacks)[4], Time-series/telemetry stores (e.g., proprietary or specialized DBs for sensor data, inferred)[4], On-premise data lakes for simulation/operational data (inferred)[4]
Compliance: Defense-grade secure cloud and sovereign environments (non-public details)[4], Likely compliance with European defense and export-control regimes (inferred from role as 'Europe’s largest defence technology company')[3][6]
Helsing operates a hybrid of secure cloud and sovereign/on‑prem defense environments built around Python/C++/Rust, PyTorch, and Kubernetes for AI workloads and telemetry/simulation data processing, integrated into physical Resilience Factories for production of autonomous and EW systems.[1][3][4] Their core software stack ingests multi‑sensor and weapons‑system data to provide real‑time battlefield insights and decision support rather than public, consumer-facing LLM services.[5][6]
Analytics · AWS · Direct API
Foundational models: GPT — Direct API (primary); Claude — Direct API
Stack: PostgreSQL
Compliance: SOC 2
Collaborative data notebook; Magic AI routes to OpenAI + Anthropic over warehouse queries.
Insurtech · AWS · Off the LLM map (in-house non-LLM ML)
Foundational models: in-house non-LLM ML only
Stack: PostgreSQL
Compliance: SOC 2
Home insurer on in-house underwriting/imagery ML; off the foundational-LLM map at the core.
Healthcare · AWS · Native (own / self-hosted weights)
Foundational models: Polaris core healthcare conversation LLM — Native (own / self-hosted weights) (primary); Polaris supervisor constellation (30+ specialized healthcare LLMs, 70B+ parameters) — Native (own / self-hosted weights)
Stack: proprietary healthcare data lake (details undisclosed), vector database for embeddings and retrieval (inferred), relational/operational DB for tooling and orchestration (inferred)
Compliance: HIPAA (healthcare PHI handling, inferred from domain and clinical deployments), Enterprise security controls on AWS (NVIDIA case study mentions AWS deployment)[5]
Hippocratic AI’s production architecture centers on the Polaris constellation: a core healthcare conversation LLM surrounded by 30+ specialized supervisor, verifier, and tool-call models, many built on open-source foundations but trained and operated as proprietary healthcare agents in AWS.[1][3][5][8] The system uses online and offline LLM judges plus task-specific engines (e.g., overdose, labs/vitals, benefits, scheduling) to monitor and govern voice interactions and workflow actions at clinical scale.[1][4][8]
Observability · AWS · Direct API
Foundational models: Claude — Direct API (primary)
Stack: own columnar
Compliance: SOC 2
Observability platform; its Query Assistant is built on Anthropic Claude over telemetry.
CRM · AWS · Direct API
Foundational models: GPT-4-class — Direct API (primary)
Stack: HBase, MySQL, Kafka
Compliance: SOC 2, GDPR
AWS-native marketing/CRM suite built on a large HBase + MySQL backbone with Kafka event streaming. AI ('Breeze'/ChatSpot) wraps OpenAI plus in-house models over first-party CRM data.
AI/LLM · AWS · Native (own / self-hosted weights)
Foundational models: Llama / Mistral / open models — Native (own / self-hosted weights) (primary)
Stack: S3 object storage, Git LFS, PostgreSQL
Compliance: SOC 2, GDPR
The model/dataset hub on AWS, storing weights as Git LFS repos over S3 and running Inference Endpoints/Spaces. Stewards open libraries (Transformers, Diffusers) — the open-source center of gravity for ML.
AI/LLM · Azure · Native (own / self-hosted weights)
Foundational models: Inflection (own) — Native (own / self-hosted weights) (primary)
Stack: object
Compliance: SOC 2
Trains its own Inflection models (now enterprise-focused) on Azure supercompute.
Biotech · Multi-Cloud · Native (own / self-hosted weights)
Foundational models: Chemistry42 / own models — Native (own / self-hosted weights) (primary)
Stack: object
Compliance: SOC 2
Generative-chemistry drug discovery on its own target-discovery + molecule-generation models.
Marketplace · GCP · Direct API
Foundational models: GPT — Direct API (primary)
Stack: PostgreSQL, BigQuery
Compliance: SOC 2
Own catalog/logistics ML; an early, prominent OpenAI customer for search + Ask Instacart.
Support · AWS · Direct API
Foundational models: Claude — Direct API (primary); GPT — Direct API
Stack: PostgreSQL, MySQL
Compliance: SOC 2, GDPR
Fin support agent runs on Anthropic + OpenAI over a customer help-content index.
FinTech · AWS · Native (own / self-hosted weights)
Foundational models: GenOS (own) — Native (own / self-hosted weights) (primary); Claude — Cloud-hosted (Bedrock/Vertex/Azure)
Stack: PostgreSQL, own
Compliance: SOC 2, PCI DSS
Intuit Assist runs on its own GenOS platform with partner LLMs underneath.
Legaltech · AWS · Direct API
Foundational models: GPT — Direct API (primary); Claude — Direct API
Stack: PostgreSQL, vector
Compliance: SOC 2
Contract-lifecycle platform; AI contract review/drafting on OpenAI + Anthropic.
Biotech · GCP · Native (own / self-hosted weights)
Foundational models: AlphaFold 3 / AlphaFold-derived models — Native (own / self-hosted weights) (primary)
Stack: object storage, cloud data platform (unspecified), scientific/HPC data systems (unspecified)
Compliance:
Public materials indicate Isomorphic Labs runs its production AI/drug-design engine on Google Cloud, using Google Cloud AI Hypercomputer with TPU/GPU fleets orchestrated by Google Kubernetes Engine for autoscaling inference workloads. The company publicly emphasizes its own drug-design engine and AlphaFold-derived models rather than exposing a foundational LLM-centric stack.
Marketing AI · AWS · Direct API
Foundational models: GPT — Direct API (primary); Claude — Direct API
Stack: PostgreSQL
Compliance: SOC 2
Marketing copy platform orchestrating OpenAI + Anthropic over brand context.
AgTech · AWS · Native (own / self-hosted weights)
Foundational models: own ML (See & Spray/autonomy) — Native (own / self-hosted weights) (primary)
Stack: own
Compliance: ISO 27001
Precision-ag + autonomy on large in-house computer-vision/ML models running on equipment edge.
EdTech · GCP · Direct API
Foundational models: GPT — Direct API (primary)
Stack: own, PostgreSQL
Compliance: SOC 2, COPPA
Khanmigo tutor built on OpenAI GPT-4 over its learning content, with heavy safety scaffolding.
FinTech · AWS · Direct API
Foundational models: GPT-4 class (OpenAI Chat Completions) — Direct API (primary)
Stack: Amazon DynamoDB, PostgreSQL, Oracle Database, Apache Kafka (as log/event store), Hadoop HDFS, Amazon S3, Amazon Redshift, Riak, RabbitMQ (message broker)
Compliance: AWS financial services security and compliance alignment (core banking on AWS, collaboration with AWS compliance/security teams)[10], Bank-grade security controls inferred from licensed bank status and AWS FSI reference architectures[10]
Production is built as microservices on AWS (including EKS/ECS) with event-driven flows via Kafka and a data platform spanning DynamoDB, S3, Glue/Spark, Hadoop/HDFS and Redshift, plus legacy Riak/RabbitMQ in some paths.[1][2][5][6][8][9][10][11] Klarna’s AI customer service stack layers agent orchestration (e.g., LangGraph) and RAG on top of OpenAI GPT-4-class models, integrated with internal APIs and Zendesk.[3][4][7]
Marketing · AWS · Native (own / self-hosted weights)
Foundational models: own ML — Native (own / self-hosted weights) (primary); GPT — Direct API
Stack: MySQL, Cassandra
Compliance: SOC 2, GDPR
Marketing automation; own predictive ML plus OpenAI for content over its customer-data platform.
AI Framework · AWS · Middleware / wrapper
Foundational models: partner LLMs — Middleware / wrapper (primary)
Stack: PostgreSQL
Compliance: SOC 2
Orchestration framework + LangSmith observability; sits between apps and any model.
Insurtech · AWS · Off the LLM map (in-house non-LLM ML)
Foundational models: in-house non-LLM ML only
Stack: PostgreSQL
Compliance: SOC 2
AI-first insurer running large in-house underwriting/claims ML (AI Maya/Jim); off the foundational-LLM map at the core.
Electronics (Korea) · On-Prem · Native (own / self-hosted weights)
Foundational models: EXAONE (own) — Native (own / self-hosted weights) (primary)
Stack: own
Compliance: ISO 27001
Trains EXAONE, its own foundation models, for products/enterprise on its own infrastructure.
AI Infra · Multi-Cloud · Native (own / self-hosted weights)
Foundational models: Llama — Native (own / self-hosted weights) (primary)
Stack: object
Compliance: SOC 2
Builds PyTorch Lightning + a studio to train/serve open models (Llama, etc.) on cloud GPUs.
DevTools · GCP · Off the LLM map (in-house non-LLM ML)
Foundational models: in-house non-LLM ML only
Stack: PostgreSQL, Redis, MongoDB
Compliance:
Linear’s CTO has described the production stack as running on Google Cloud with Kubernetes, using React and Node/TypeScript, with Postgres for primary storage, Redis for queuing, and MongoDB for caching scenarios. The only clearly documented AI-related capability in the provided sources is product-level “AI workflows”; no source here establishes a specific foundational LLM provider in the core production architecture, so model_links is empty.
AI/LLM · Hybrid · Native (own / self-hosted weights)
Foundational models: LFM2-2.6B — Native (own / self-hosted weights) (primary); LFM 1.3B — Native (own / self-hosted weights); LFM 3.1B — Native (own / self-hosted weights); LFM 40.3B MoE — Native (own / self-hosted weights)
Stack:
Compliance:
Liquid AI builds and serves its own Liquid Foundation Models (LFM and LFM2) with a custom hybrid liquid/convolution/attention architecture, optimized for both data-center and fully on-device deployment across CPUs, GPUs, and NPUs.[1][3][4][9][11] Public materials emphasize hardware-in-the-loop training and edge/PC deployment, but do not expose a full production cloud stack, suggesting a mix of self-hosted/model-serving infrastructure plus OEM/partner integrations rather than a single public hyperscaler.[9][11]
AI-coding · AWS · Direct API
Foundational models: Claude — Direct API (primary)
Stack: PostgreSQL
Compliance: SOC 2
AI app builder leaning on Anthropic Claude to generate full-stack apps from prompts.
Retail · GCP · Native (own / self-hosted weights)
Foundational models: Gemini — Cloud-hosted (Bedrock/Vertex/Azure) (primary); own ML — Native (own / self-hosted weights)
Stack: own
Compliance: SOC 2, PCI DSS
Home-improvement retailer on Google Cloud; Gemini-based associate/customer assistants + own ML.
Mobility · AWS · Off the LLM map (in-house non-LLM ML)
Foundational models: in-house non-LLM ML only
Stack: MySQL, DynamoDB
Compliance: SOC 2
In-house pricing/ETA ML; off the LLM map at the core.
AI-coding · GCP · Native (own / self-hosted weights)
Foundational models: Magic LTM-2 / long-context LTM models — Native (own / self-hosted weights) (primary); Magic LTM-1 — Native (own / self-hosted weights)
Stack:
Compliance:
Magic’s public evidence points to a proprietary, long-context model stack centered on its own LTM models rather than a packaged third-party LLM. The clearest infrastructure signal is a Google Cloud partnership for AI supercomputers, but the public record does not expose a full production data stack or compliance posture.
FinTech · AWS · Off the LLM map (in-house non-LLM ML)
Foundational models: in-house non-LLM ML only
Stack: PostgreSQL
Compliance: PCI DSS, SOC 2
Card-issuing platform; in-house risk ML; off the foundational-LLM map at the core.
E-Commerce · AWS · Native (own / self-hosted weights)
Foundational models: own ML — Native (own / self-hosted weights) (primary); GPT — Direct API
Stack: own, MySQL
Compliance: SOC 2, PCI DSS
LatAm commerce/fintech giant; large in-house recommendation/fraud ML plus OpenAI for assistants.
Automotive · Multi-Cloud · Native (own / self-hosted weights)
Foundational models: Gemini — Cloud-hosted (Bedrock/Vertex/Azure) (primary); own ML — Native (own / self-hosted weights)
Stack: own
Compliance: ISO 27001, GDPR
In-car MBUX assistant on Google Cloud Gemini; large in-house ADAS/vehicle ML.
FinTech · AWS · Direct API
Foundational models: GPT — Direct API (primary)
Stack: PostgreSQL
Compliance: SOC 2
Startup banking; AI assist on OpenAI over transaction data.
Platform · Azure · Native (own / self-hosted weights)
Foundational models: GPT-4 class models (Copilot, Azure OpenAI GPT family) — Cloud-hosted (Bedrock/Vertex/Azure) (primary); Phi family (small/medium models used in Copilot and Azure AI) — Native (own / self-hosted weights)
Stack: Azure SQL Database, SQL Server, Azure Cosmos DB, OneLake (Microsoft Fabric), Azure Data Lake Storage
Compliance: Azure global distributed datacenter security controls, Data residency and geographies/regions model, Availability zones for high availability and resiliency
Microsoft operates primarily as a cloud-first organization running its internal and external services on Azure, backed by globally distributed, Microsoft-operated datacenters and a hybrid-cloud capable architecture.[1][4][6] Core business, productivity, and analytics workloads are delivered via Azure IaaS/PaaS (VMs, app services, databases) plus newer unified analytics platforms like Microsoft Fabric/OneLake, integrated with shared Microsoft services such as Microsoft 365 and identity.[1][5][7]
Image AI · Hybrid · Native (own / self-hosted weights)
Foundational models: Midjourney V7 diffusion model — Native (own / self-hosted weights) (primary); Midjourney V6.x diffusion models — Native (own / self-hosted weights); CLIP or CLIP-like text encoder — Native (own / self-hosted weights)
Stack: object storage (likely S3/GCS-equivalent)[5], message queues/job state store (unspecified)[5], CDN-backed asset delivery[5]
Compliance:
Midjourney runs a proprietary latent diffusion image model pipeline behind a Discord bot and web UI, with prompts flowing through a request-handling layer into GPU clusters for model inference, then storing generated images in large-scale object storage fronted by a CDN.[5][9] Public information focuses on model architecture and system-design patterns; specific cloud vendor, database products, and formal compliance regimes are not disclosed.[1][5][9]
Collaboration · AWS · Direct API
Foundational models: GPT — Direct API (primary)
Stack: PostgreSQL
Compliance: SOC 2, GDPR
Visual canvas; AI features on OpenAI.
AI/LLM · Multi-Cloud · Native (own / self-hosted weights)
Foundational models: Mistral Large (frontier reasoning model family, including latest versions) — Direct API (primary); Mistral Small / Medium / Devstral (production-efficient and developer-tuned models, including Devstral 2 123B) — Direct API; Codestral (code-generation family, including Codestral 25.05) — Direct API; Pixtral / Pixtral Large (vision-language models) — Direct API; Mistral 7B (open-weight) — Native (own / self-hosted weights); Mixtral 8x7B (open-weight Mixture-of-Experts) — Native (own / self-hosted weights); Mistral Small 3.1 (open-weight, on-prem friendly) — Native (own / self-hosted weights); Devstral 2 123B — Cloud-hosted (Bedrock/Vertex/Azure); Mistral open-weight models (e.g., Mistral 7B, Mixtral 8x7B) on Bedrock — Cloud-hosted (Bedrock/Vertex/Azure); Medium 3 — Cloud-hosted (Bedrock/Vertex/Azure); Codestral 25.05 — Cloud-hosted (Bedrock/Vertex/Azure); Mistral open weights (European-licensed) — Cloud-hosted (Bedrock/Vertex/Azure)
Stack: object storage for model artifacts, datasets, and logs (e.g., S3-compatible across partner clouds)[10], vector databases used with Mistral embeddings and open-weight models (e.g., Qdrant, Pinecone, pgvector; typical for La Plateforme and self-hosting scenarios)[11], relational databases for accounts, billing, and orchestration metadata (likely PostgreSQL/MySQL, inferred from typical SaaS/API patterns)[10][11]
Compliance: SOC 2 Type II (claims to comply with SOC 2 Type II framework, with reports available via Trust/Help Center)[3][4][7][12][15], ISO 27001 (claims to comply with ISO 27001 framework)[3][4][7][12][15], ISO 27701 (claims to comply with ISO 27701 framework)[3][4][7][12][15], GDPR (EU-incorporated company with GDPR-focused data processing and SCCs; EU data residency by default for La Plateforme)[3][4][6][9][11]
Mistral runs a multi-cloud architecture where its hosted La Plateforme and Studio offerings sit atop partner clouds (Google Cloud, AWS, Azure, SAP, IBM, Snowflake, NVIDIA, Outscale), while open-weight models are also deployable on‑prem via standard inference containers.[10][11] The production stack is therefore split between first‑party EU‑centric hosting for regulated workloads and distribution of models through major cloud marketplaces and managed services.
Productivity · AWS · Direct API
Foundational models: GPT — Direct API (primary)
Stack: MySQL, Vitess
Compliance: SOC 2, ISO 27001
Work OS; AI blocks on OpenAI over board data.
Database · Multi-Cloud · Middleware / wrapper
Foundational models: partner embeddings/LLMs — Middleware / wrapper (primary)
Stack: MongoDB
Compliance: SOC 2, HIPAA, PCI DSS, ISO 27001
Atlas runs the MongoDB document database as a managed service across AWS, Azure and GCP. Atlas Vector Search makes it a RAG datastore; AI features integrate partner LLMs rather than a captive model.
FinTech (EU) · AWS · Off the LLM map (in-house non-LLM ML)
Foundational models: in-house non-LLM ML only
Stack: PostgreSQL
Compliance: PCI DSS, GDPR
European neobank on in-house fraud/risk ML; off the foundational-LLM map at the core.
Internet · Hybrid · Native (own / self-hosted weights)
Foundational models: HyperCLOVA X — Native (own / self-hosted weights) (primary); Seoul World Model (urban/world modeling built on NAVER street-view data) — Direct API; NVIDIA Cosmos world foundation models (used within NAVER "physical AI" stack) — Cloud-hosted (Bedrock/Vertex/Azure)
Stack: Proprietary Naver search/indexing and user data stores (on-prem, GAK Sejong), Cloud-native databases within NAVER Cloud (unspecified mix of SQL/NoSQL)
Compliance: Regional data sovereignty and localization requirements for Korea and partner regions[2][6], Enterprise and government-grade AI cloud security controls (unspecified standards)[2][7]
NAVER runs its core portal and AI services primarily on **on-premise data centers** such as the GAK Sejong facility, increasingly augmented by NAVER Cloud’s sovereign AI infrastructure built with NVIDIA DSX and AMD EPYC/Instinct platforms for large-scale training and inference.[2][4][5][6][7] The production stack is vertically integrated—NAVER controls the data centers, cloud layer (NAVER Cloud), and proprietary foundation models like HyperCLOVA X, exposed as services to internal products and external partners.[2][5][7]
Database · AWS · Middleware / wrapper
Foundational models: partner LLM — Middleware / wrapper (primary)
Stack: PostgreSQL (pgvector)
Compliance: SOC 2
Serverless Postgres (now Databricks); AI via pgvector + partner embeddings.
Streaming Media · Hybrid · Off the LLM map (in-house non-LLM ML)
Foundational models: in-house non-LLM ML only
Stack: Cassandra, MySQL/RDS, CockroachDB, EVCache, DynamoDB, Elasticsearch, S3, Redshift
Compliance: SOC 2 (inferred from large-scale AWS SaaS practices)[2][9], ISO 27001 (inferred, typical for global consumer SaaS on AWS)[2][9], PCI-DSS for payments and billing workflows (inferred)[8][9]
Netflix runs a cloud‑native microservices control plane almost entirely on AWS (EC2, RDS/MySQL, DynamoDB, Cassandra, CockroachDB, Kafka, S3, etc.) while its proprietary Open Connect CDN appliances form a separate data plane deployed at ISPs for video delivery.[2][3][6][12] Client apps (web, mobile, TV) talk to backend microservices via APIs/GraphQL, with extensive use of in‑house open‑source tooling (Zuul, Eureka, Spinnaker, EVCache) and AWS services for storage, streaming, and analytics.[3][6][8][12]
Productivity · AWS · Off the LLM map (in-house non-LLM ML)
Foundational models: in-house non-LLM ML only
Stack: PostgreSQL, Amazon RDS, Apache Kafka (Confluent Cloud), Apache Hudi, S3, Snowflake, Debezium CDC, Spark
Compliance: data residency controls (US/EU regional processing)
Notion’s core production architecture is centered on a sharded PostgreSQL system on AWS, with 96 physical Postgres instances and five logical shards per instance as of 2023. Its data lake pipeline uses Debezium CDC into Confluent Cloud Kafka, then Apache Hudi and S3 for downstream analytics, search, and AI-related workloads; the public sources do not confirm a native foundational model inside Notion’s core product.
FinTech · AWS · Off the LLM map (in-house non-LLM ML)
Foundational models: in-house non-LLM ML only
Stack: Cassandra, PostgreSQL
Compliance: SOC 2
Heavy in-house ML (acquired Hyperplane); off the LLM map at the core.
Cybersecurity · AWS · Middleware / wrapper
Foundational models: partner LLM — Middleware / wrapper (primary)
Stack: MySQL, DynamoDB, Redis
Compliance: SOC 2, FedRAMP (High), ISO 27001
AWS-native identity cloud (workforce + Auth0 customer identity) on MySQL/DynamoDB with a multi-tenant cell architecture. AI is in-house threat-detection ML plus assistant LLMs.
AI/LLM · Hybrid · Native (own / self-hosted weights)
Foundational models: GPT-4 class frontier models (e.g., GPT-4 family) — Native (own / self-hosted weights) (primary); GPT-4o and successor ChatGPT models — Native (own / self-hosted weights)
Stack: Azure Cosmos DB, PostgreSQL, Blob/Object Storage (e.g., Azure Blob), Kafka (for event/log streaming)
Compliance: SOC 2 (inferred from enterprise-grade cloud use on Azure)[15][16], ISO 27001 (inferred from Azure AI and hyperscaler environments)[15][16], Regional data residency and enterprise controls via Azure/OpenAI Deployment Company (inferred)[15][19][20]
OpenAI runs large GPU superclusters and Kubernetes-based orchestration across Azure and its own data centers in a hybrid setup, with Azure as the primary cloud and custom HPC clusters (e.g., Stargate) for training and serving frontier models.[15][16][19] Core application and API workloads use relational stores like PostgreSQL for accounts/settings and globally scalable databases such as Azure Cosmos DB plus Kafka streams for high-volume conversation, analytics, and event data.[1][8][12]
Healthcare AI · Multi-Cloud · Native (own / self-hosted weights)
Foundational models: OpenEvidence medical (own) — Native (own / self-hosted weights) (primary)
Stack: vector
Compliance: HIPAA
Medical-evidence answer engine on its own models trained over peer-reviewed literature.
Enterprise · On-Prem · Cloud-hosted (Bedrock/Vertex/Azure)
Foundational models: Cohere Command — Cloud-hosted (Bedrock/Vertex/Azure) (primary); Llama — Cloud-hosted (Bedrock/Vertex/Azure)
Stack: Oracle
Compliance: SOC 2, FedRAMP, HIPAA
Runs its own OCI; OCI Generative AI hosts Cohere + Llama for enterprise.
Telecom (EU) · Hybrid · Native (own / self-hosted weights)
Foundational models: Mistral — Direct API (primary); own ML — Native (own / self-hosted weights)
Stack: own
Compliance: GDPR
European telecom partnering with Mistral for sovereign LLMs over its own network ML.
Govtech/Data · Hybrid · Cloud-hosted (Bedrock/Vertex/Azure)
Foundational models: Customer-selected foundational and ML models integrated via AIP (no single Palantir-native LLM at core) — Cloud-hosted (Bedrock/Vertex/Azure) (primary)
Stack: Customer-provided data warehouses and lakes (plug-in architecture), Relational databases (e.g., PostgreSQL, Oracle, SQL Server) integrated via Foundry data connectors, Distributed file/object storage (e.g., S3-compatible, OCI object storage) for data lake-style storage[5][7], Palantir-managed metadata and ontology stores (proprietary)[5][7]
Compliance: SaaS deployment with isolated, secure, cloud-disconnected environments via Apollo[8], Sovereign / regulated deployments with on-prem and vendor-specific sovereign AI reference architectures (NVIDIA, Dell)[2][3][4], Enterprise-grade access control, audit, and data governance embedded in Foundry/AIP (described in architecture docs/whitepapers)[1][7]
Palantir runs three tightly integrated proprietary platforms—Foundry (data operations), AIP (generative AI), and Apollo (continuous delivery)—as a managed SaaS that can be deployed across public clouds and on-prem/sovereign infrastructure, with customers bringing their own data platforms and ML tooling into Foundry’s integration and ontology layers.[1][5][7][8] Operationally, this yields a hybrid architecture: Palantir provides the application/AI stack while underlying compute, storage, and some databases reside in customer-selected environments (AWS/GCP/Azure/OCI/private/on-prem) including recent NVIDIA and Dell sovereign AI reference architectures.[2][3][4][6]
Cybersecurity · Multi-Cloud · Native (own / self-hosted weights)
Foundational models: Custom-hosted LLMs (top frontier models, fine-tuned for internal coding copilot and SDLC use cases) — Native (own / self-hosted weights) (primary); Claude-family models (custom-trained for code generation) — Cloud-hosted (Bedrock/Vertex/Azure); Multiple proprietary security AI/ML models (~1000+ models across products such as Prisma, Cortex, PAN-OS security subscriptions) — Native (own / self-hosted weights)
Stack: Amazon RDS, Amazon Redshift, Amazon Neptune, Amazon OpenSearch Service, Redis, S3 (data lake / cold storage), Proprietary PAN-OS / Cortex / Prisma data stores
Compliance: SOC 2, ISO 27001, FedRAMP, HIPAA, PCI-DSS
Palo Alto Networks runs a multi-cloud SaaS architecture with AWS as the primary runtime for major platforms like Prisma Cloud, using services such as RDS, Redshift, Neptune, OpenSearch, Redis, and S3 alongside Kubernetes-based microservices for data ingestion, analytics, and Infinity Graph. GCP (including Vertex AI) and Azure are used selectively for AI/ML workloads and cloud-native firewall offerings, with proprietary security engines and data stores layered on top of these managed services.
Creator · GCP · Direct API
Foundational models: GPT — Direct API (primary)
Stack: MySQL
Compliance: SOC 2
Creator-membership platform; AI features on OpenAI.
FinTech · Hybrid · Off the LLM map (in-house non-LLM ML)
Foundational models: in-house non-LLM ML only
Stack: Oracle, own
Compliance: SOC 2, PCI DSS
Massive in-house fraud/risk ML on its own infrastructure; off the LLM map at the core.
AI/Search · Multi-Cloud · Native (own / self-hosted weights)
Foundational models: Sonar (family: Sonar, Sonar Pro, Sonar Reasoning, Sonar Deep Research) — Native (own / self-hosted weights) (primary); DeepSeek-R1 (via Sonar Reasoning Pro) — Cloud-hosted (Bedrock/Vertex/Azure); GPT-5 / GPT-4.x class — Direct API; Claude (Claude 4.x Sonnet / Opus) — Cloud-hosted (Bedrock/Vertex/Azure); Gemini 2.5 Pro — Direct API; Grok 4 — Direct API
Stack: Elasticsearch, PostgreSQL, vector store
Compliance:
Perplexity operates a multi-cloud architecture: it historically built and hosted its Sonar models on AWS (including Bedrock for Anthropic/Claude) while later signing a large Azure agreement to deploy models via Microsoft Foundry, indicating active use of both providers.[1][3][8][9][15] Its core production stack centers on proprietary Sonar LLMs (built on and fine‑tuned from open-source bases like Mistral and later Llama 3.x) combined with partner frontier models (GPT‑5/4-class, Claude, Gemini, Grok) selected per query.[1][2][5][6][10][11][13][14]
Robotics · Hybrid · Native (own / self-hosted weights)
Foundational models: π (robotics foundation model controlling many robot platforms) — Native (own / self-hosted weights) (primary)
Stack: Object storage (WEKA Data Platform over high-performance files/object)[2], Oracle Cloud Infrastructure storage/DB services (unspecified mix, likely OCI Object Storage and managed DB)[1], Custom distributed data infrastructure for robot learning (data pipelines between raw telemetry and training/eval)[11]
Compliance:
Physical Intelligence runs a proprietary robotics foundation model stack with large-scale training and data infrastructure, using WEKA for high-performance data and Oracle Cloud Infrastructure for hosted compute, complemented by Kubernetes-based CI/CD and GPU-heavy cloud environments.[1][2][7][11][12] The architecture centers on Python/C++ robotics services, PyTorch/JAX-based model training on cloud GPUs, and data pipelines from robot teleoperation and simulation into foundation model training and evaluation.[11][12]
AI Infrastructure · Multi-Cloud · Middleware / wrapper
Foundational models: partner embedding models — Middleware / wrapper (primary)
Stack: proprietary vector database
Compliance: SOC 2, HIPAA, ISO 27001
Managed vector database for RAG, deployed across AWS/GCP/Azure with a serverless architecture that separates storage (object store) from query compute. The index, not a model, is the product.
Social · AWS · Native (own / self-hosted weights)
Foundational models: Navigator-1 (own, open-source-based) — Native (own / self-hosted weights) (primary); Qwen (open) — Native (own / self-hosted weights)
Stack: HBase, MySQL
Compliance: SOC 2
Built Navigator-1, its own multimodal foundation model (fine-tuned from open-source bases incl. Qwen) powering the Pinterest Assistant at ~10% the cost of frontier models; big in-house recommendation ML.
FinTech · AWS · Off the LLM map (in-house non-LLM ML)
Foundational models: in-house non-LLM ML only
Stack: MongoDB, PostgreSQL
Compliance: SOC 2, ISO 27001, GDPR
AWS-hosted financial-data network connecting apps to banks; MongoDB + PostgreSQL behind a heavy data-normalization and anti-fraud ML layer. AI is in-house risk/identity models, not a public LLM brand.
AI Aggregator · AWS · Middleware / wrapper
Foundational models: partner LLMs (aggregator) — Middleware / wrapper (primary)
Stack: MySQL
Compliance: GDPR
Consumer multi-model aggregator fronting OpenAI, Anthropic, Google and open models behind one app.
AI-coding · AWS · Native (own / self-hosted weights)
Foundational models: Laguna S — Native (own / self-hosted weights) (primary); Malibu — Cloud-hosted (Bedrock/Vertex/Azure); Point — Cloud-hosted (Bedrock/Vertex/Azure)
Stack: Apache Iceberg (central data/experiment layer), Object storage over Iceberg (likely S3-compatible, inferred from AWS + Iceberg usage), Kubernetes-native config/state stores (e.g., etcd; inferred from 'stacked under a single Kubernetes orchestrator')
Compliance: Enterprise-grade, in-VPC / on-prem isolation (docs + platform marketing), Air‑gapped deployment support (AWS re:Invent talk), Customer-owned VPC / data residency controls (docs), No explicit public claims of SOC 2 / ISO 27001 / HIPAA in available materials
Poolside builds and trains proprietary foundation models via its internal **Model Factory** (Titan training stack on a ~10K GPU cluster, Apache Iceberg data layer, Kubernetes orchestration) and then deploys those models and agentic systems fully inside customer VPCs/on‑prem through the Poolside Platform, with additional distribution via AWS Bedrock and Trainium-backed inference.[1][2][7][8][9][5][6][11] The production architecture is therefore a proprietary, Kubernetes-based ML/agent stack running on AWS (and customer infrastructure) with an Iceberg-centered data plane and tightly integrated model training, evaluation, and deployment pipelines.[1][2][7][8][9][11]
DevTools · AWS · Direct API
Foundational models: GPT — Direct API (primary)
Stack: PostgreSQL
Compliance: SOC 2
API platform; Postbot AI on OpenAI over API specs/collections.
Construction SaaS · AWS · Direct API
Foundational models: GPT — Direct API (primary)
Stack: PostgreSQL
Compliance: SOC 2
Construction-management cloud; AI agents on OpenAI over project/document data.
AI Infra · Multi-Cloud · Middleware / wrapper
Foundational models: partner embeddings — Middleware / wrapper (primary)
Stack: vector
Compliance: SOC 2
Open-source vector database for RAG; model-agnostic.
Consumer AI · Multi-Cloud · Middleware / wrapper
Foundational models: partner LLMs (multi) — Middleware / wrapper (primary)
Stack: object
Compliance: GDPR
AI hardware (r1); routes to multiple partner LLMs behind a large-action-model orchestration layer.
E-Commerce (Japan) · Hybrid · Native (own / self-hosted weights)
Foundational models: Rakuten AI (own LLM) — Native (own / self-hosted weights) (primary)
Stack: own
Compliance: ISO 27001
Japanese commerce/fintech ecosystem training its own Japanese-first Rakuten AI models.
FinTech · AWS · Direct API
Foundational models: GPT-4-class — Direct API (primary); Claude — Direct API
Stack: PostgreSQL
Compliance: SOC 2, PCI DSS
AWS-native finance-automation/corporate-card platform on PostgreSQL, AI-first in positioning: LLM agents (OpenAI/Anthropic) automate expense categorization, approvals and procurement over transaction data.
FinTech (India) · AWS · Native (own / self-hosted weights)
Foundational models: own ML — Native (own / self-hosted weights) (primary); GPT — Direct API
Stack: MySQL, PostgreSQL
Compliance: PCI DSS, SOC 2
Indian payments; in-house fraud/risk ML plus OpenAI for support assistants.
Banking · On-Prem · Native (own / self-hosted weights)
Foundational models: Cohere Command — Cloud-hosted (Bedrock/Vertex/Azure) (primary); own ML — Native (own / self-hosted weights)
Stack: own
Compliance: SOC 2, OSFI
Canadian bank running large in-house risk ML; Cohere models deployed in-VPC for sovereign generative AI (Borealis AI).
Biotech · GCP · Native (own / self-hosted weights)
Foundational models: Recursion Phenom (own) — Native (own / self-hosted weights) (primary)
Stack: BigQuery, own
Compliance: SOC 2
Drug discovery on its own phenomics foundation models trained over petabytes of cellular imaging.
Social Media · AWS · Direct API
Foundational models: GPT-4-class — Direct API (primary)
Stack: PostgreSQL, Cassandra, Memcached
Compliance: SOC 2, GDPR, CCPA
AWS-hosted; a Python/Pylons heritage over PostgreSQL with Cassandra and heavy Memcached caching, moving to Go/GraphQL services. Monetizes data licensing for AI training; Reddit Answers uses LLMs.
AI/LLM · Multi-Cloud · Native (own / self-hosted weights)
Foundational models: Reka Core — Native (own / self-hosted weights) (primary); Reka Flash — Native (own / self-hosted weights); Reka Edge / reka-edge-2603 — Native (own / self-hosted weights)
Stack:
Compliance:
Reka trains and serves its own multimodal encoder–decoder frontier models (Core, Flash, Edge) on custom Kubernetes-based GPU clusters spanning multiple vendors, using PyTorch on large H100/A100 fleets and a separate A10/A100 inference stack.[2][5] Public deployment is via Reka’s own web app and API endpoints (chat.reka.ai, platform.reka.ai, showcase.reka.ai), with an OpenAI-compatible API server for Edge provided through vLLM and Hugging Face artifacts.[2][7][12]
Telecom (India) · On-Prem · Native (own / self-hosted weights)
Foundational models: BharatGPT / Jio Brain (own+partner) — Native (own / self-hosted weights) (primary)
Stack: own
Compliance: ISO 27001
Telecom/digital giant building sovereign Indian-language models (Jio Brain / BharatGPT) on its own infrastructure.
AI Infra · GCP · Native (own / self-hosted weights)
Foundational models: Llama — Native (own / self-hosted weights) (primary)
Stack: object
Compliance: SOC 2
Runs open models as one-click APIs on its own GPU fleet.
DevTools · GCP · Direct API
Foundational models: Claude — Direct API (primary)
Stack: PostgreSQL (Neon), object
Compliance: SOC 2
Cloud IDE whose Agent is built primarily on Anthropic Claude.
DevTools · AWS · Direct API
Foundational models: GPT — Direct API (primary); Claude — Direct API
Stack: PostgreSQL
Compliance: SOC 2
Internal-tools builder with multi-model AI app components.
FinTech · GCP · Direct API
Foundational models: GPT — Direct API (primary)
Stack: PostgreSQL
Compliance: SOC 2, PCI DSS
In-house risk ML plus OpenAI for support/assistant features.
Gaming · Hybrid · Native (own / self-hosted weights)
Foundational models: own ML — Native (own / self-hosted weights) (primary)
Stack: own
Compliance: SOC 2
League/Valorant maker; in-house anti-cheat + matchmaking ML on its own infra; off the foundational-LLM map at the core.
HR/IT · AWS · Direct API
Foundational models: GPT — Direct API (primary); Claude — Direct API
Stack: MySQL, PostgreSQL
Compliance: SOC 2
Workforce platform; AI agents on OpenAI + Anthropic over the employee graph.
Automotive · AWS · Native (own / self-hosted weights)
Foundational models: own ML (autonomy/vision) — Native (own / self-hosted weights) (primary)
Stack: own
Compliance: ISO 27001
EV maker training its own driver-assist/vehicle ML; off the foundational-LLM map at the core of the product.
Legaltech · AWS · Direct API
Foundational models: Claude — Direct API (primary)
Stack: PostgreSQL, vector
Compliance: SOC 2
Contract-AI platform built primarily on Anthropic Claude over legal documents.
FinTech · AWS · Cloud-hosted (Bedrock/Vertex/Azure)
Foundational models: Amazon Nova — Cloud-hosted (Bedrock/Vertex/Azure) (primary)
Stack: MySQL, Cassandra
Compliance: SOC 2
Cortex investing assistant (Digests) runs on Amazon Nova via Bedrock, chosen for ~80% lower per-token cost; core trading/risk stays in-house ML. Now also exposes agent-trading APIs.
Gaming · Hybrid · Native (own / self-hosted weights)
Foundational models: Roblox Assistant (own) — Native (own / self-hosted weights) (primary)
Stack: MongoDB, own
Compliance: SOC 2, COPPA
UGC gaming platform on its own data centers; trains its own generative models for 3D/code creation assistants.
Insurtech · AWS · Off the LLM map (in-house non-LLM ML)
Foundational models: in-house non-LLM ML only
Stack: PostgreSQL
Compliance: SOC 2
Telematics auto-insurer on in-house pricing/driving ML; off the foundational-LLM map at the core.
Video AI · AWS · Native (own / self-hosted weights)
Foundational models: Gen-4.5 — Native (own / self-hosted weights) (primary); Gen-4 — Native (own / self-hosted weights); Gen-3 Alpha / Gen-3 Alpha Turbo — Native (own / self-hosted weights); Gen-2 — Native (own / self-hosted weights)
Stack: object storage, vector database
Compliance:
Runway’s public product and developer docs show a proprietary, cloud-hosted AI platform with first-party APIs and an internal model catalog rather than exposed third-party foundational model dependencies. The available evidence supports AWS as the most likely cloud footprint, but the exact database technologies and compliance attestations are not publicly confirmed in the sources provided.
CRM · Multi-Cloud · Cloud-hosted (Bedrock/Vertex/Azure)
Foundational models: OpenAI GPT-4-class (Einstein generative features, Copilot, Agentforce) — Cloud-hosted (Bedrock/Vertex/Azure) (primary); Anthropic Claude (Einstein and Agentforce integrations) — Cloud-hosted (Bedrock/Vertex/Azure)
Stack: Oracle (primary multitenant relational database for core CRM instances), Custom multitenant metadata store on top of Oracle, PostgreSQL (in parts of Hyperforce/public cloud services, inferred from prior record and common cloud RDBMS usage)[4], Kafka (Ajna platform for event streaming and data pipelines)[8], HDFS/Hadoop (DeepSea analytics platform consuming Kafka streams)[8], Time-series store (Argus, internal monitoring/metrics)[8]
Compliance: Multitenant isolation with org-level data segregation and encryption keys[3][10], Hyperforce regional deployment for data residency and regulatory compliance[1][2][4], Zero-trust oriented access controls and strong identity/permission model (role hierarchy, permission sets, Shield encryption)[12][18], Multi-AZ high-availability with disaster recovery across paired data centers/regions[4][7][10]
Salesforce runs a proprietary, metadata-driven, multitenant application stack (Superpods, instances, orgs) on its Hyperforce infrastructure spanning its own data centers and leading public clouds, with a single logical multitenant database per instance and stateless app servers layered above it.[2][4][7][10] Around this core, it operates a hybrid data and streaming platform (Ajna Kafka, Hadoop/HDFS, monitoring stacks) and integrates external warehouses like Snowflake for Data Cloud and analytics workloads.[8][17]
AI-hardware · On-Prem · Native (own / self-hosted weights)
Foundational models: Samba-1 (own) — Native (own / self-hosted weights) (primary); Llama — Native (own / self-hosted weights)
Stack: object
Compliance: SOC 2
Own RDU AI chips plus its Samba-1 model and hosted open models for enterprise/government.
IoT · AWS · Off the LLM map (in-house non-LLM ML)
Foundational models: in-house non-LLM ML only
Stack: own time-series, PostgreSQL
Compliance: SOC 2, ISO 27001
Connected-operations (fleet/IoT); large own ML on sensor/video telemetry; off the foundational-LLM map at the core.
Enterprise · Hybrid · Native (own / self-hosted weights)
Foundational models: SAP ABAP LLM (own) — Native (own / self-hosted weights) (primary); Claude — Cloud-hosted (Bedrock/Vertex/Azure); GPT — Cloud-hosted (Bedrock/Vertex/Azure); Mistral / Cohere (sovereign) — Cloud-hosted (Bedrock/Vertex/Azure)
Stack: HANA
Compliance: SOC 2, ISO 27001
Joule is multi-model: SAP’s own ABAP LLM plus a Generative AI Hub fronting OpenAI, Gemini, Anthropic (Claude powers Joule agents), Llama, and sovereign Mistral/Cohere on SAP cloud.
AI/Data · AWS · Native (own / self-hosted weights)
Foundational models: Defense Llama (own, Llama-3 fine-tuned) — Native (own / self-hosted weights) (primary); frontier models (eval/RLHF) — Cloud-hosted (Bedrock/Vertex/Azure)
Stack: PostgreSQL, MongoDB
Compliance: SOC 2, FedRAMP, ISO 27001
Beyond data-labeling, Scale trains its own Defense Llama (fine-tuned Llama 3) for U.S. national security on its Donovan federal platform; evaluates frontier models for the DoD CDAO.
Cybersecurity · AWS · Native (own / self-hosted weights)
Foundational models: Ultraviolet (own) — Native (own / self-hosted weights) (primary); Claude — Direct API; GPT — Direct API
Stack: own, Snowflake
Compliance: SOC 2, FedRAMP
Purple AI combines SentinelOne’s own Ultraviolet models with Anthropic Claude + OpenAI GPT over its Singularity data lake; exposes a Purple MCP server for agentic SOC investigation.
DevTools · GCP · Direct API
Foundational models: GPT — Direct API (primary); Claude — Direct API
Stack: PostgreSQL, ClickHouse
Compliance: SOC 2
Error-monitoring; Autofix/Seer uses OpenAI + Anthropic over stack traces.
Enterprise · Multi-Cloud · Native (own / self-hosted weights)
Foundational models: Apriel 2.0 (Apriel Nemotron open model family) — Native (own / self-hosted weights) (primary); Now LLM / ServiceNow reasoning models (enterprise workflow-focused) — Native (own / self-hosted weights); NVIDIA NeMo microservices (reasoning, data processing, guardrailed workflows) — Direct API; StarCoder2 (3B code model variant trained by ServiceNow) — Native (own / self-hosted weights); StarCoder2 (7B model) — Direct API; StarCoder2 (15B model on NVIDIA accelerated infrastructure) — Direct API
Stack: RaptorDB (proprietary engine built on PostgreSQL)[7], PostgreSQL (underlying basis for RaptorDB)[7], MariaDB (legacy / still referenced as backend DB)[1], MySQL-compatible relational database (historical / variant of MySQL)[1][11]
Compliance: FedRAMP High (via AWS GovCloud and Azure Government sovereign deployments)[8], DoD IL5 (via AWS GovCloud for US defense workloads)[8], EU/UK/Saudi/India/Australia sovereign cloud posture (regulator-driven RFP compliance)[8]
ServiceNow runs a proprietary SaaS platform across hyperscaler regions on AWS, Azure, and GCP, plus sovereign-cloud deployments, with a custom database layer (RaptorDB on PostgreSQL) replacing MariaDB/MySQL for most instances.[7][8] Its architecture is tightly integrated with NVIDIA NeMo microservices and GPU infrastructure for training and inference of its Apriel/Nemotron family of open models, while exposing workflow, observability, and discovery services over these multi-cloud foundations.[2][4][5][8]
Vertical SaaS · Azure · Cloud-hosted (Bedrock/Vertex/Azure)
Foundational models: GPT — Cloud-hosted (Bedrock/Vertex/Azure) (primary)
Stack: SQL Server
Compliance: SOC 2, PCI DSS
Trades/field-services SaaS on Azure; AI assistants via Azure OpenAI over job/CRM data.
Defense · Hybrid · Native (own / self-hosted weights)
Foundational models: Hivemind (own autonomy) — Native (own / self-hosted weights) (primary)
Stack: own
Compliance: FedRAMP
Builds Hivemind, its own autonomy foundation for GPS/comms-denied flight.
E-Commerce · GCP · Direct API
Foundational models: GPT-4-class (Shopify Magic/Sidekick and related features) — Direct API (primary)
Stack: MySQL (heavily sharded, pod-based), Redis, Memcached, Kafka, Google Cloud Storage, Confluent Schema Registry
Compliance: PCI DSS, SOC 2, ISO 27001, GDPR
Shopify runs a modular monolith (primarily Ruby on Rails) deployed as multi-tenant pods, each with its own MySQL, Redis, and memcached, on Kubernetes clusters running on Google Kubernetes Engine, with Kafka-centric data pipelines for CDC, analytics, and event distribution.[1][10][11][14] Traffic is handled via a global edge (Anycast + nginx/OpenResty) feeding stateless app workers into these pods, enabling regional failover, blast-radius isolation, and large flash-sale scaling.[2][3][11][16]
Industrial · Azure · Native (own / self-hosted weights)
Foundational models: own ML — Native (own / self-hosted weights) (primary); GPT — Cloud-hosted (Bedrock/Vertex/Azure)
Stack: own, SQL Server
Compliance: ISO 27001
Industrial Copilot built with Microsoft on Azure OpenAI over its own engineering/automation ML.
CX AI · AWS · Direct API
Foundational models: Claude — Direct API (primary)
Stack: PostgreSQL, vector
Compliance: SOC 2
Conversational-AI agents (Bret Taylor) built multi-model but Anthropic-forward.
Collaboration · Multi-Cloud · Cloud-hosted (Bedrock/Vertex/Azure)
Foundational models: Claude — Cloud-hosted (Bedrock/Vertex/Azure) (primary); Anthropic Claude (managed via Amazon SageMaker, legacy phase) — Cloud-hosted (Bedrock/Vertex/Azure); PaLM / Gemini family (large language models for Slack AI features) — Cloud-hosted (Bedrock/Vertex/Azure)
Stack: MySQL (Vitess-managed, sharded keyspaces)[10][18], Redis (caching, real-time pub/sub)[8][15], Kafka (event streaming / persistence pipeline)[6][15], PostgreSQL (some services / ancillary data)[6], Elasticsearch or similar search index for messages/search[8][20], Snowflake (analytics / BI)[13], Hive + Presto on S3 (data lake / internal analytics)[13]
Compliance: SOC 2 (implied by enterprise SaaS posture and Slack AI privacy/security claims)[9], ISO 27001 (commonly referenced in Slack security/compliance materials, inferred)[9], Data residency controls via region- and AZ-scoped Vitess clusters and cellular architecture[10][11]
Slack’s core production stack runs primarily on AWS with a cellular architecture across availability zones, Vitess‑sharded MySQL storage, stateful real‑time messaging services, and global edge regions, while its AI serving stack has evolved into a multi‑cloud setup spanning AWS Bedrock and Google Cloud Vertex AI.[3][5][10][11][4] Slack AI is integrated as an application layer atop this core messaging infrastructure, with strict data‑segregation and privacy controls rather than being a foundational replacement of Slack’s existing backend.[9][4]
SaaS · AWS · Direct API
Foundational models: GPT — Direct API (primary)
Stack: SQL Server
Compliance: SOC 2
Work-management platform; AI formulas/summaries on OpenAI over sheet data.
Social · GCP · Direct API
Foundational models: GPT — Direct API (primary); Gemini — Cloud-hosted (Bedrock/Vertex/Azure)
Stack: Spanner, BigQuery
Compliance: SOC 2
My AI assistant runs on OpenAI + Google; heavy in-house vision ML elsewhere.
Data Platform · Multi-Cloud · Native (own / self-hosted weights)
Foundational models: Snowflake Cortex native models (embedding, classification, search, etc.) — Native (own / self-hosted weights) (primary); Open-source LLMs (e.g., Llama, Mistral family) used in customer Snowpark Container Services / Model Serving — Native (own / self-hosted weights); Partner LLMs (e.g., Anthropic Claude, OpenAI GPT family) accessible via Snowflake Cortex integrations — Cloud-hosted (Bedrock/Vertex/Azure)
Stack: Proprietary multi-cluster shared-data columnar engine on cloud object storage (Amazon S3, Azure Blob Storage, Google Cloud Storage)[1][16][19], Service-oriented cloud services layer for metadata, transactions, optimization, authentication, and coordination[1][16][19], Massively parallel processing virtual warehouses as independent compute clusters[14][16][19]
Compliance: Role-based access control and fine-grained security policies in cloud services layer[1][16], Encryption of data at rest and in transit on public cloud object storage (S3/Blob/GCS)[1][16], Secure data sharing and governance features integrated into the services layer[1][11][16]
Snowflake runs a proprietary three-layer, multi-cluster shared-data architecture on AWS, Azure, and GCP, separating compressed columnar storage on cloud object stores from MPP compute warehouses and a distributed cloud services control plane.[1][16][19] Internally it is a service-oriented system with independently scalable storage, compute, and metadata/transaction services built for OLAP workloads.[16][19]
Cybersecurity · GCP · Native (own / self-hosted weights)
Foundational models: DeepCode AI (own) — Native (own / self-hosted weights) (primary)
Stack: PostgreSQL, MongoDB
Compliance: SOC 2, ISO 27001, GDPR
Developer-security platform on GCP/AWS scanning code and dependencies; DeepCode AI combines symbolic analysis with ML for fix suggestions. Backed by a curated open-source vulnerability database.
FinTech · AWS · Direct API
Foundational models: GPT — Direct API (primary)
Stack: PostgreSQL
Compliance: SOC 2, PCI DSS
Digital finance; in-house risk ML plus OpenAI for member support/assistant.
Conglomerate (Japan) · Hybrid · Native (own / self-hosted weights)
Foundational models: own ML — Native (own / self-hosted weights) (primary); partner LLM — Cloud-hosted (Bedrock/Vertex/Azure)
Stack: own
Compliance: ISO 27001
Telecom/investment group building its own Japanese LLMs (Sarashina) plus partner models; major AI-infra investor.
Aerospace · Hybrid · Off the LLM map (in-house non-LLM ML)
Foundational models: in-house non-LLM ML only
Stack: own
Compliance: ITAR
Rockets + Starlink; massive in-house autonomy/vision/control ML; off the foundational-LLM map at the core.
Legaltech · AWS · Direct API
Foundational models: GPT — Direct API (primary)
Stack: PostgreSQL, vector
Compliance: SOC 2
AI contract drafting in Word, built on OpenAI with legal-tuned retrieval.
Streaming Media · GCP · Direct API
Foundational models: GPT-4-class (AI DJ, personalization tooling) — Direct API (primary)
Stack: Apache Cassandra, Google BigQuery, PostgreSQL, Google Cloud Bigtable, Google Cloud Storage, Redis, HDFS
Compliance: SOC 2, ISO 27001, GDPR, CCPA
Spotify runs a large-scale microservices architecture on Google Cloud Platform, with thousands of backend services and streaming/batch data pipelines built around Pub/Sub, Dataflow/Beam, and BigQuery for recommendation and analytics workloads.[5][9][10] Audio is stored in cloud object storage and delivered via multiple CDNs, while Cassandra and other stores serve user metadata, playlists, and real-time features.[3][5][7]
Website Builder · Hybrid · Direct API
Foundational models: GPT-4-class — Direct API (primary)
Stack: MySQL, MongoDB, Cassandra
Compliance: SOC 2, PCI DSS, GDPR
Website/commerce builder historically on its own data centers, increasingly cloud-hybrid; polyglot persistence (MySQL/MongoDB/Cassandra). AI site-generation and copy use OpenAI.
Automotive (EU) · Multi-Cloud · Native (own / self-hosted weights)
Foundational models: Mistral — Cloud-hosted (Bedrock/Vertex/Azure) (primary); own ML — Native (own / self-hosted weights)
Stack: own
Compliance: ISO 27001, GDPR
Automaker (Jeep/Peugeot/Fiat) embedding Mistral in-vehicle plus its own ADAS ML.
FinTech · AWS · Off the LLM map (in-house non-LLM ML)
Foundational models: in-house non-LLM ML only
Stack: MongoDB, custom sharded database layer (DocDB)
Compliance: PCI DSS, SOC 1, SOC 2, ISO 27001
Stripe’s publicly documented infrastructure shows heavy AWS usage for observability and operations, plus a sharded/tiered storage design that Stripe built to scale data access and migrations. Public sources do not confirm a core production reliance on a foundational LLM; the available evidence points to a proprietary payments platform rather than an open model stack.
Media · AWS · Direct API
Foundational models: GPT — Direct API (primary)
Stack: PostgreSQL
Compliance: SOC 2
Publishing platform; AI writing/discovery features on OpenAI.
Healthcare AI · Multi-Cloud · Native (own / self-hosted weights)
Foundational models: Suki clinical (own) — Native (own / self-hosted weights) (primary); GPT — Cloud-hosted (Bedrock/Vertex/Azure)
Stack: vector
Compliance: HIPAA, SOC 2
Ambient clinical assistant blending its own medical models with partner LLMs.
Audio AI · AWS · Native (own / self-hosted weights)
Foundational models: Suno (own) — Native (own / self-hosted weights) (primary)
Stack: object
Compliance: GDPR
Trains + serves its own music-generation models.
DevTools · AWS · Middleware / wrapper
Foundational models: partner LLM — Middleware / wrapper (primary)
Stack: PostgreSQL (pgvector)
Compliance: SOC 2
Postgres backend-as-a-service; AI via pgvector + partner embeddings.
AI-coding · Multi-Cloud · Native (own / self-hosted weights)
Foundational models: Tabnine (own) — Native (own / self-hosted weights) (primary); Claude — Cloud-hosted (Bedrock/Vertex/Azure)
Stack: vector
Compliance: SOC 2
Privacy-first code assistant with its own models plus optional partner LLMs, deployable air-gapped.
Retail · GCP · Native (own / self-hosted weights)
Foundational models: Gemini — Cloud-hosted (Bedrock/Vertex/Azure) (primary); own ML — Native (own / self-hosted weights)
Stack: own
Compliance: SOC 2, PCI DSS
Retailer on Google Cloud; store-companion + guest AI on Gemini over its own merchandising ML.
Dev-infra · Multi-Cloud · Off the LLM map (in-house non-LLM ML)
Foundational models: in-house non-LLM ML only
Stack: Cassandra, PostgreSQL
Compliance: SOC 2
Durable-execution platform; model-agnostic orchestration, off the foundational-LLM map at the core.
Healthcare · GCP · Native (own / self-hosted weights)
Foundational models: Tempus genomic/clinical ML (own) — Native (own / self-hosted weights) (primary); GPT — Direct API
Stack: BigQuery, own
Compliance: HIPAA, SOC 2
Precision-medicine platform with its own multimodal genomic/clinical models plus LLMs for clinician copilots.
Internet · On-Prem · Native (own / self-hosted weights)
Foundational models: Hunyuan — Native (own / self-hosted weights) (primary)
Stack: TDSQL, TDSQL Boundless, Redis, Apache Kafka, OpenStack/TStack, TCE/TCS
Compliance: VPC integration, IP whitelisting, encryption, access controls, regional deployment, audit logging, conversation logging, content moderation, prompt injection defense
Tencent’s production stack is hybrid in practice, combining large-scale self-hosted infrastructure (notably OpenStack/TStack and Tencent Cloud-native/TCE/TCS) with major open-source infrastructure components such as Apache Kafka. Its database layer is centered on Tencent-built systems like TDSQL/TDSQL Boundless, with enterprise controls spanning regional deployment, access control, encryption, and auditing.
Automotive · Hybrid · Native (own / self-hosted weights)
Foundational models: Tesla FSD / autonomous driving "world model" — Native (own / self-hosted weights) (primary); Optimus robot control and perception models — Native (own / self-hosted weights)
Stack: AWS S3 (object storage / data lake)[10], LakeFS (data lake versioning)[10], Parquet (columnar file format for video / telemetry)[10], TFRecord (training data format)[10], Snowflake (cloud data warehouse)[10], BigQuery (cloud data warehouse)[10], Redis (real-time state / caching)[1], Kafka / Pulsar (streaming backbone)[1][10]
Compliance: TLS/SSL-encrypted APIs for vehicle-cloud communication[10], Tokenized vehicle identifiers for data ingestion[10], Phased OTA rollout and canary deployments for safety and reliability[1][4], Simulation-driven QA and regression testing before production updates[4]
Tesla runs a vertically integrated, edge-first architecture where custom FSD hardware in vehicles and robots performs real-time inference, while a hybrid cloud (including AWS and in-house GPU clusters) ingests telemetry into Kafka/Flink streams, S3/LakeFS lakes, and Snowflake/BigQuery warehouses for continuous model training and OTA updates.[1][7][8][10] Production for cars, energy products, and robots is tightly coupled to this AI and data infrastructure, effectively making Gigafactories and AI clusters part of a single closed-loop manufacturing and autonomy system.[2][6][7]
AdTech · AWS · Native (own / self-hosted weights)
Foundational models: Kokai ML (own) — Native (own / self-hosted weights) (primary)
Stack: own columnar, Aerospike
Compliance: SOC 2
Programmatic ad DSP; Kokai is its own large-scale bidding/forecasting ML over trillions of ad-auction events.
Analytics · Multi-Cloud · Native (own / self-hosted weights)
Foundational models: Spotter (own) — Native (own / self-hosted weights) (primary); GPT — Direct API
Stack: own, Falcon
Compliance: SOC 2, GDPR
Search/AI analytics; Spotter agent blends its own engine with OpenAI over the in-memory data store.
Restaurant SaaS · AWS · Direct API
Foundational models: GPT — Direct API (primary)
Stack: PostgreSQL, Kafka
Compliance: SOC 2, PCI DSS
Restaurant POS/platform; AI menu/ops assistants on OpenAI over transaction data.
AI Infra · Hybrid · Native (own / self-hosted weights)
Foundational models: Meta Llama (e.g., Llama 3 family) — Native (own / self-hosted weights) (primary); Mistral (e.g., Mixtral / Mistral Instruct variants) — Native (own / self-hosted weights); Qwen (Alibaba open-weight family) — Native (own / self-hosted weights); GLM (e.g., GLM 5.2) — Native (own / self-hosted weights)
Stack: Amazon S3 (Iceberg tables)[2], Apache Iceberg[2], MotherDuck (DuckDB-based analytics serving layer)[2], Kinesis[2], Kafka[2]
Compliance: PCI (via PCI-compliant infrastructure for regulated enterprise deployments)[12]
Together AI operates its own GPU cloud and managed Kubernetes clusters as an AI-native infrastructure layer, while also integrating with external cloud storage and streaming services for data and analytics workloads.[2][11][12] The production stack is centered on open-weight models deployed as containerized workloads on Together-managed GPU clusters, exposed via serverless inference APIs and voice/agent pipelines.[3][4][9][11][16]
Mapping · Azure · Native (own / self-hosted weights)
Foundational models: own ML — Native (own / self-hosted weights) (primary); GPT — Cloud-hosted (Bedrock/Vertex/Azure)
Stack: own
Compliance: ISO 27001, GDPR
Mapping/location; own routing/map ML plus a Microsoft-partnered conversational in-car assistant on Azure OpenAI.
Cybersecurity · AWS · Middleware / wrapper
Foundational models: partner LLM — Middleware / wrapper (primary)
Stack: PostgreSQL
Compliance: SOC 2
Security hyperautomation; routes partner LLMs through its SOC workflow engine.
Communications · AWS · Direct API
Foundational models: GPT-4-class — Direct API (primary)
Stack: MySQL, DynamoDB, Kafka
Compliance: SOC 2, HIPAA, PCI DSS, GDPR
AWS-native communications APIs (voice/SMS/email via SendGrid) over MySQL + DynamoDB with Kafka eventing. Owns the Segment CDP; CustomerAI layers LLMs over unified customer profiles.
Mobility · Hybrid · Off the LLM map (in-house non-LLM ML)
Foundational models: in-house non-LLM ML only
Stack: Schemaless (MySQL-based Docstore), Docstore (on MySQL), MySQL, Cassandra, Riak, Apache Pinot, HDFS, Hive, Presto, Apache Spark, Elasticsearch, Kafka (as streaming log/store)
Compliance: SOC 2 (inferred from large-scale hybrid cloud and enterprise focus), PCI-DSS (inferred from handling global payments and card data), GDPR (inferred from operating across EU with personal location data)
Uber runs thousands of microservices on a hybrid cloud with multiple active data centers, using Docker on Mesos/Aurora, Kafka + Flink/Spark + Pinot/Presto/HDFS for real-time and batch data, and in-house layers like Michelangelo AI as a model gateway on top of this foundation.[11][16][3][9][6] Core operational data lives in Schemaless/Docstore on MySQL plus Cassandra/Riak, with extensive self-hosted open-source infrastructure (Kafka, Flink, Pinot, Spark, Presto, HDFS, Elasticsearch) augmented by proprietary platforms such as Michelangelo and Zero Growth Stack.[11][3][9][13][6]
EdTech · AWS · Direct API
Foundational models: GPT — Direct API (primary)
Stack: MySQL
Compliance: SOC 2
Learning marketplace; AI assistant + course tools on OpenAI.
Gaming/3D · Multi-Cloud · Native (own / self-hosted weights)
Foundational models: own ML — Native (own / self-hosted weights) (primary); partner LLM — Cloud-hosted (Bedrock/Vertex/Azure)
Stack: own
Compliance: SOC 2
Real-time 3D engine; Muse AI blends partner LLMs with its own generative tooling for creators.
Compliance · AWS · Direct API
Foundational models: GPT — Direct API (primary)
Stack: PostgreSQL
Compliance: SOC 2, ISO 27001
Security-compliance automation; AI questionnaire/evidence features on OpenAI.
Healthcare · AWS · Cloud-hosted (Bedrock/Vertex/Azure)
Foundational models: partner LLM — Cloud-hosted (Bedrock/Vertex/Azure) (primary)
Stack: Oracle, own
Compliance: HIPAA, SOC 2, GxP
Life-sciences cloud; Veeva AI exposes partner LLMs over regulated CRM/clinical data with a governance layer.
DevTools · AWS · Direct API
Foundational models: GPT-4-class — Direct API (primary); Claude — Direct API
Stack: PostgreSQL (Neon), Redis, object storage
Compliance: SOC 2, ISO 27001, GDPR
Frontend cloud built atop AWS, exposing a global edge + serverless/Functions layer; stewards the open-source Next.js framework. v0 AI is built on OpenAI/Anthropic models.
Telecom · Hybrid · Native (own / self-hosted weights)
Foundational models: Gemini — Cloud-hosted (Bedrock/Vertex/Azure) (primary); own ML — Native (own / self-hosted weights)
Stack: own
Compliance: SOC 2
Telecom running its own network ML; customer/agent assistants on Google Cloud Gemini.
Retail · Hybrid · Off the LLM map (in-house non-LLM ML)
Foundational models: in-house non-LLM ML only
Stack: Apache Kafka, Spark Streaming, Storm, HDFS, NoSQL databases, Cassandra
Compliance: TLS/HTTPS, enterprise network segmentation, multi-cloud container orchestration controls, configuration management/drift control
Walmart’s production architecture is hybrid and multi-cloud: public-cloud workloads run across Azure and Google Cloud, while core infrastructure and brokers remain within Walmart-controlled private cloud environments.[1][2][7][10][15] The platform is heavily event-driven and containerized, with Kafka as a central backbone and WCNP/Kubernetes used for consumer workloads across clouds.[1][4]
Climate · GCP · Direct API
Foundational models: GPT — Direct API (primary)
Stack: PostgreSQL
Compliance: SOC 2
Carbon-accounting platform; AI features on OpenAI over emissions data.
E-Commerce · GCP · Native (own / self-hosted weights)
Foundational models: Gemini — Cloud-hosted (Bedrock/Vertex/Azure) (primary); own ML — Native (own / self-hosted weights)
Stack: own, MySQL
Compliance: SOC 2, PCI DSS
Home-goods retailer on Google Cloud; in-house recommendation/visual ML plus Gemini (Vertex) for shopping assistants.
Autonomous · GCP · Native (own / self-hosted weights)
Foundational models: Waymo Foundation Model for Autonomous Driving — Native (own / self-hosted weights) (primary)
Stack: Bigtable, Spanner, Google Cloud Storage, Proprietary in-vehicle data stores
Compliance: ISO 27001 (inferred via Alphabet/Google Cloud hosting)[4], SOC 2 (inferred via Alphabet/Google Cloud ecosystem)[4], Automotive safety and validation frameworks for AVs, including systematic safety case construction[1][8]
Waymo’s production driver uses a modular six-layer stack (sensor processing, perception, world modeling, prediction, planning, control) integrated with HD maps, simulation, and a foundation-model-based end-to-end driving architecture that fuses camera, lidar, and radar into a unified world representation.[1][2][8][9][10] Training, evaluation, and large-scale data handling run on Google/Alphabet’s internal and Google Cloud infrastructure, with closed-source models and tooling.[1][4][5][8][10]
Autonomous · Azure · Native (own / self-hosted weights)
Foundational models: AI Driver (end-to-end driving foundation model / 'world model') — Native (own / self-hosted weights) (primary); GAIA-3 (generative driving simulation foundation model) — Native (own / self-hosted weights); LINGO / language-capable driving explanation components (name varies across sources, internal LM-like module) — Native (own / self-hosted weights)
Stack: Azure Storage (Blob/Object)[7], Azure Databricks (Delta Lake/Parquet over object storage)[7], Azure Machine Learning experiment/model metadata stores[9][12], Internal time-series/telemetry stores for fleet data (unspecified, likely on Azure)[9][12]
Compliance: Azure built-in compliance (ISO 27001, SOC 1/2/3, GDPR support) inferred from exclusive Azure usage[7][12], Wayve safety framework for AV (Safety 2.0 paradigm, not a formal security cert but core to production governance)[8]
Wayve runs a proprietary end-to-end embodied driving stack on Microsoft Azure, using large vision-world models (AI Driver, GAIA-3) trained on petabyte-scale fleet and simulated data via Azure Kubernetes Service, Azure Machine Learning, Databricks, and GPU clusters, then deploys a compressed model on in-vehicle compute for production driving.[7][9][12] The architecture replaces modular perception/mapping/planning with a single neural policy and related foundation models (driving, simulation, scenario classification, language explanation), all controlled via internal MLOps pipelines on Azure.[1][4][9][12]
FinTech · AWS · Off the LLM map (in-house non-LLM ML)
Foundational models: in-house non-LLM ML only
Stack: PostgreSQL
Compliance: SOC 2
Automated investing on in-house quantitative ML; off the foundational-LLM map at the core.
AI Infra · Multi-Cloud · Middleware / wrapper
Foundational models: partner embeddings — Middleware / wrapper (primary)
Stack: vector
Compliance: SOC 2
Open-source vector database; the index, not a model, is the product.
Website Builder · AWS · Direct API
Foundational models: GPT-4-class — Direct API (primary); Claude — Direct API
Stack: MongoDB, MySQL, Redis
Compliance: SOC 2, GDPR
Visual web-development platform on AWS; MongoDB for the flexible CMS/site model with MySQL/Redis support. AI assists site generation and content via OpenAI/Anthropic.
MLOps · Multi-Cloud · Middleware / wrapper
Foundational models: partner LLM — Middleware / wrapper (primary)
Stack: object, ClickHouse
Compliance: SOC 2
ML experiment-tracking + LLMOps (Weave); model-agnostic, runs over customers chosen models.
QSR · GCP · Cloud-hosted (Bedrock/Vertex/Azure)
Foundational models: Gemini — Cloud-hosted (Bedrock/Vertex/Azure) (primary)
Stack: own
Compliance: SOC 2, PCI DSS
FreshAI drive-thru ordering built on Google Cloud Gemini over menu/voice data.
AI-coding · AWS · Direct API
Foundational models: Claude — Direct API (primary); GPT — Direct API
Stack: vector, PostgreSQL
Compliance: SOC 2
AI IDE (formerly Codeium) routing to Claude + GPT over its own retrieval/indexing engine.
FinTech · AWS · Off the LLM map (in-house non-LLM ML)
Foundational models: in-house non-LLM ML only
Stack: PostgreSQL
Compliance: SOC 2
In-house ML for FX/fraud; no core foundational-LLM dependency.
Website Builder · Hybrid · Direct API
Foundational models: GPT — Direct API (primary)
Stack: MySQL, own
Compliance: SOC 2, GDPR
Site builder; AI site/content generation on OpenAI.
Cybersecurity · Multi-Cloud · Off the LLM map (in-house non-LLM ML)
Foundational models: in-house non-LLM ML only
Stack: graph database (Wiz Security Graph), production databases for metadata storage (type not publicly disclosed), cloud-provider native encrypted storage for backend data
Compliance: ISO 27001, ISO 27017, ISO 27018, ISO 27701, PCI DSS v4.0.1, SOC 2 Type 2, SOC 3, CCPA, CSA STAR Level 1, FedRAMP High, GovRAMP, IRAP, TX-RAMP, TruSight
Wiz operates as a SaaS platform with an agentless, multi-cloud security architecture that ingests cloud, workload, identity, and data metadata into a unified Security Graph. Its production environment is described as immutable infrastructure managed through infrastructure-as-code, with isolated production databases and cloud-native encryption, but the exact underlying database vendors are not publicly disclosed.
HR · Hybrid · Native (own / self-hosted weights)
Foundational models: Workday Illuminate LLM (own ~800B) — Native (own / self-hosted weights) (primary)
Stack: MySQL, own
Compliance: SOC 2, ISO 27001
Runs Illuminate, its own ~800B-parameter LLM trained on HR/finance data from 10,500+ customers (800B transactions/yr) for privacy + enterprise fit; acquired Sana (2025) for custom AI agents.
Enterprise AI · AWS · Native (own / self-hosted weights)
Foundational models: Palmyra (Writer enterprise-grade LLM family, including task- and industry-specific variants) — Native (own / self-hosted weights) (primary); Self-hosted LLM (Writer-branded, deployable in customer cloud or on‑prem) — Native (own / self-hosted weights)
Stack: PostgreSQL, vector
Compliance: SOC 2 (inferred), GDPR (inferred)
Writer operates as a multi-tenant SaaS platform primarily hosted on AWS, with options for customers to deploy Writer’s self-hosted LLM on-premises or in their own cloud/VPC for stricter data residency and control.[9][16] The core stack combines enterprise-grade proprietary LLMs, a knowledge graph over customer data sources, and agent orchestration tooling (AI HQ / Agent Builder) exposed via web UI and APIs.[10][16][19]
AI/LLM · On-Prem · Native (own / self-hosted weights)
Foundational models: Grok — Native (own / self-hosted weights) (primary)
Stack:
Compliance:
The provided sources do not expose xAI’s actual production infrastructure, and most search results are generic XAI architecture guidance rather than xAI company disclosures. The only xAI-specific material in the results points to an internal product architecture (Home Mixer, Thunder, Phoenix, Grox) but does not establish cloud, database, or compliance stack details for xAI’s core production environment.[9][5]
Local/Internet · AWS · Native (own / self-hosted weights)
Foundational models: own ML — Native (own / self-hosted weights) (primary); GPT — Direct API
Stack: MySQL, Cassandra
Compliance: SOC 2
Local-discovery platform; own ranking/review ML plus OpenAI for its AI assistant.
AI Search · Multi-Cloud · Middleware / wrapper
Foundational models: partner LLMs (multi-model) — Middleware / wrapper (primary)
Stack: vector, Elasticsearch
Compliance: SOC 2
AI search/agent routing across multiple frontier + open models over its own web index.
Support · AWS · Direct API
Foundational models: GPT — Direct API (primary)
Stack: MySQL, Elasticsearch
Compliance: SOC 2, HIPAA
AWS-native support suite; AI agents + summarization on OpenAI.
Proptech · AWS · Native (own / self-hosted weights)
Foundational models: own ML (Zestimate) — Native (own / self-hosted weights) (primary); GPT — Direct API
Stack: own, SQL Server
Compliance: SOC 2
Real-estate marketplace; the Zestimate is large in-house valuation ML; natural-language search on OpenAI.
SaaS (India) · On-Prem · Native (own / self-hosted weights)
Foundational models: Zia / ZIA LLM (own) — Native (own / self-hosted weights) (primary)
Stack: own
Compliance: SOC 2, ISO 27001
Bootstrapped SaaS suite on its own Indian data centers; builds its own Zia AI models for privacy + cost control.
FoodTech (India) · AWS · Native (own / self-hosted weights)
Foundational models: own ML — Native (own / self-hosted weights) (primary); GPT — Direct API
Stack: own, PostgreSQL
Compliance: PCI DSS, SOC 2
Indian food-delivery; own logistics/recommendation ML plus OpenAI assistants.
Communications · Hybrid · Native (own / self-hosted weights)
Foundational models: Zoom SLM (own, federated) — Native (own / self-hosted weights) (primary); Claude — Direct API; GPT-4-class — Direct API
Stack: MySQL, MongoDB, Kafka
Compliance: SOC 2, HIPAA, FedRAMP, PCI DSS
Runs its own global data centers for real-time media, bursting to AWS/Oracle Cloud at peak. AI Companion uses a 'federated' approach blending its own models with Anthropic and OpenAI.
Cybersecurity · On-Prem · Off the LLM map (in-house non-LLM ML)
Foundational models: in-house non-LLM ML only
Stack: own
Compliance: SOC 2, FedRAMP
Own global proxy cloud + in-house ML; off the LLM map at the core.