The AI Infra Ecosystem

The AI Infra Ecosystem Map is a living reference of 293 companies building the infrastructure layer of AI, mapped across Jensen Huang's five-layer cake — Application, Models, Infrastructure, Chips, and Energy. The middle layer — Infrastructure — is cut open and expanded in full across seven sublayers: cloud & compute, inference, agents, orchestration, MLOps, storage & networking, and data centers.

explore the map -> provider index -> compute ladder -> gpu census ->

293Companies

5Layers

66Open source

Leader

Major

Emerging

Startup

OSS Open source

Region

⌕

Methodology · how companies are chosen

This map expands the infrastructure layer of AI in full; the surrounding four layers (Application, Models, Chips, Energy) appear as context only. A company is listed when it ships a production product in one of the seven infrastructure sublayers with meaningful adoption, funding, or open-source traction. Tiers reflect market presence, not endorsement — Leader (category-defining), Major (established), Emerging (scaling), Startup (early-stage); OSS marks open-source-core projects. It is deliberately a living draft — additions and corrections are welcome.

Know a company that belongs here?

Suggestions are welcome — drop a note and we'll review.

Suggest a company →

Frequently asked questions

What is the AI Infra Ecosystem Map?

The AI Infra Ecosystem Map is a living reference of 293 companies building the infrastructure layer of AI. It organizes the market across the five layers of the AI stack — Application, Models, Infrastructure, Chips, and Energy — and expands the Infrastructure layer in full across seven sublayers: AI Cloud & Compute, Inference & Model APIs, Agent Infrastructure, Cluster Orchestration & Tenant Isolation, MLOps & Training, Storage & Networking, and Data Center & Hardware.

What is the AI infrastructure layer?

The AI infrastructure layer sits between AI models and the chips they run on. It includes GPU clouds and compute providers, inference and model-serving platforms, agent infrastructure (sandboxes, memory, tool use), Kubernetes and cluster orchestration, MLOps and training tooling, high-performance storage and networking, and the data centers, power, and cooling that everything runs on.

How are companies selected for the map?

A company is listed when it ships a production product in one of the seven infrastructure sublayers with meaningful adoption, funding, or open-source traction. Tiers reflect market presence, not endorsement: Leader (category-defining), Major (established), Emerging (scaling), and Startup (early-stage). Open-source-core projects carry an OSS badge.

How many companies are on the map?

The July 2026 edition maps 293 companies, of which 66 are open-source-core projects. The map is a living draft and is updated as the market evolves; suggestions and corrections are welcome via the suggestion form.

What are the five layers of the AI stack?

The five-layer model of the AI stack, popularized by NVIDIA CEO Jensen Huang, consists of: (1) Application — AI products, developer tools, and agents; (2) Models — frontier and open-weight labs; (3) Infrastructure — cloud, inference, orchestration, MLOps, storage, and data centers; (4) Chips — GPUs, accelerators, and memory; and (5) Energy — power, cooling, and data-center capacity.

Full company index text version · 293 companies · 66 OSS

The AI Infra Ecosystem Map is a curated, living reference of 293 companies building AI infrastructure, organized across the five layers of the AI stack — Application, Models, Infrastructure, Chips, and Energy — with the Infrastructure layer expanded across seven sublayers. The complete list below mirrors the interactive map above in plain text.

Infrastructure — expanded in full

AI Cloud & Compute · Hyperscalers, GPU-native clouds, sovereign clouds, and GPU marketplaces 84 companies

Hyperscalers

AWS — EC2 P5/Trn2, EKS, SageMaker. Broadest AI service portfolio (Leader)
Microsoft Azure — ND H100/H200, AKS, Azure OpenAI Service (Leader)
Google Cloud — A3/A4, GKE, TPU, Hypercomputer (Leader)
Oracle Cloud — BM.GPU bare-metal, OKE, RDMA-native (Major)
IBM Cloud — Enterprise AI cloud, H100 instances, hybrid cloud + on-prem (Major)
Alibaba Cloud — Chinese hyperscaler, Apsara AI, global GPU compute (Major)

GPU-Native AI Clouds

CoreWeave — K8s-native GPU cloud, NVIDIA Elite partner (Major)
Lambda — 1-Click Clusters, H100/H200, NVIDIA NCP (Major)
Together AI — GPU cloud + inference, strong OSS model support (Major)
Nebius — Full-stack AI cloud, K8s + Slurm, EU-HQ (Major)
Crusoe — Clean energy GPU cloud (Major)
Vultr — 32-region GPU cloud, bare metal (Major)
RunPod — Serverless GPU, community cloud (Emerging) · deep dive
Fluidstack — Distributed GPU cloud, multi-region (Emerging)
Shadeform — GPU aggregator across 15+ providers (Startup)
Paper Compute — GPU cloud, simple API and developer experience (Startup)
Runlayer — GPU cloud infrastructure for AI workloads (Startup)
HydraHost — GPU infrastructure platform, AI factory builder — EU-focused (Startup)
TensorWave — AMD MI300X-based GPU cloud, alternative to NVIDIA stack (Emerging) · deep dive
Cirrascale — Dedicated GPU clusters, deep-learning focused data center (Emerging)
Armada — Distributed GPU computing platform — cross-cluster workload scheduling (Startup)
FlexAI — AI compute orchestration — GPU cluster management, any cloud (Startup)
Bitdeer — GPU cloud and AI datacenter capacity — Nasdaq-listed, ex-mining infrastructure (Emerging)
Voltage Park — On-demand and reserved H100 GPU clusters (Emerging)
Gcore — Edge & GPU cloud, 140+ PoPs globally, EU-hosted H100 clusters (Emerging)
Firmus — Sustainable AI compute, liquid-cooled UK data centers (Startup) · deep dive
GMI Cloud — GPU cloud for AI workloads, Asia-Pacific infrastructure (Startup) · deep dive
Corvex — Security-first GPU cloud — confidential computing on NVIDIA clusters, model weights encrypted even from the host (Nasdaq: MOVE) (Emerging) · deep dive
Neysa — Managed GPU clusters and AI infrastructure platform (Startup)
STN — Bare metal GPU cloud provider (Startup)
Prime Intellect — Distributed AI training, open compute network (Startup)
Latitude.sh — Bare metal & GPU cloud, developer-friendly pricing (Startup)
DENVR Networks — High-density GPU data centers, HPC-focused cloud (Startup)
Verda — EU GPU cloud, formerly DataCrunch (Startup)
Atlas Cloud — UK-based GPU cloud for enterprise AI (Startup)
Hot Aisle — GPU colocation and managed AI compute (Startup)
Buzz HPC — HPC GPU cloud for research and enterprise (Startup)
Hetzner — German cloud, cost-effective H100 GPU instances (Emerging)
Hyperbolic — Open-access GPU cloud and inference API (Startup)
Deepinfra — GPU-backed LLM inference APIs, pay-per-token (Startup) · deep dive
Massed Compute — GPU cloud for AI training workloads, UK-based (Startup)
Exabits — Fractional GPU compute marketplace (Startup)
Sesterce — French GPU cloud, bare metal H100 clusters (Startup)
Highrise — GPU cloud for AI startups (Startup)
Arc Compute — GPU cloud infrastructure provider (Startup)
Sharon AI — GPU cloud provider (Startup)
FarmGPU — GPU cloud for AI workloads (Startup)
PaleBlueDot.AI — Sustainable GPU cloud (Startup)
Mithril — GPU cloud provider (Startup)
Firebird — GPU cloud infrastructure (Startup)
Moonlite — GPU cloud provider (Startup)
Megaspeed — GPU cloud provider (Startup)
BlueSky Ai — GPU cloud provider (Startup)
Radiant AI — Vertically integrated AI infrastructure platform — power, land, capital & compute unified; NVIDIA Cloud Partner with Blackwell/Rubin GPUs and Radiant AI Catalyst software stack (Major)

GPU Marketplaces

Vast.ai — GPU marketplace — 17K+ GPUs from 1,400+ providers in 500+ locations; owns no GPUs, per-second billing (Startup) · deep dive
Cudo Compute — Decentralized GPU marketplace, sustainable compute, UK-based (Startup)
Aethir — Decentralized GPU network for AI and cloud gaming (Startup)
Akash Network — Open decentralized cloud compute marketplace (Startup)
Clore.ai — GPU marketplace, crypto-incentivized rental (Startup)
Salad — Consumer GPU network for distributed AI inference (Startup)
GPU.net — GPU cloud network and marketplace (Startup)

Sovereign & Regional

Scaleway — EU sovereign AI cloud, GPU instances, K8s (Emerging)
OVHcloud — European cloud, data sovereignty (Major)
Nscale — European AI cloud, GPU infrastructure (Emerging)
DigitalOcean — Developer-focused cloud, GPU droplets (Major)
Hyperstack — GPU cloud, UK-based — EU-sovereign AI infrastructure (Emerging)
GMO GPU Cloud — Japanese GPU cloud, H100 clusters, APAC sovereign (Emerging)
Core42 — UAE national AI cloud, G42 + NVIDIA partnership (Emerging)
Humain — Saudi-backed AI cloud, NVIDIA partnership (LEAP 2025) (Startup)
FPT AI Factory — Vietnamese AI cloud, FPT Corporation backed (Startup)
E2E Cloud — Indian GPU cloud provider, competitive H100 pricing (Startup)
Yotta — Indian hyperscale data center and GPU cloud (Startup)
Sakura Internet — Japanese cloud provider, H100 GPU instances (Emerging)
NAVER Cloud — Korean cloud, HyperCLOVA AI platform (Emerging)
SK Telecom — Korean telecom, AI cloud and T3K data center services (Major)
VESSL AI — Korean ML platform and GPU cloud (Startup)
backend.ai — Korean GPU cluster management and cloud (Startup)
BytePlus — ByteDance international cloud platform, GPU instances (Startup)
RunSun Cloud — Chinese GPU cloud provider (Startup)
Qubrid — GPU and quantum cloud platform (Startup)
TELUS — Canadian telecom AI cloud services (Emerging)
Telenor — Nordic telecom GPU cloud infrastructure (Emerging)
Tatra SuperCompute — Slovak HPC supercomputer cloud (Startup)
Indosat — Indonesian telecom cloud and GPU services (Startup)

Inference & Model APIs · Managed inference providers, serverless model APIs, and inference-as-a-service platforms 12 companies

Inference API Providers

Together AI — GPU cloud + inference API, strong OSS model support (Major)
Fireworks AI — Inference platform, compound AI systems, fast token generation (Emerging)
Baseten — Inference platform for custom ML models (Emerging)
Modal — Serverless GPU, Python-native, fast cold start — inference and batch workloads (Emerging)
Replicate — Run OSS models via API, developer-friendly (Emerging)
FriendliAI — LLM inference optimization, high-throughput token generation (Emerging) · deep dive
SambaNova — Cloud inference platform on custom RDU chips — ultra-fast token gen (Emerging)
openrouter — LLM API aggregator — single API to 200+ models, smart routing (Startup)
Featherless — Serverless LLM inference, instant deployment, no cold start (Startup)
Fal — Generative-media inference platform — fast image, video and audio model APIs (Emerging)
Telnyx — Carrier-edge inference — OpenAI-compatible model APIs on 4,000+ owned GPUs colocated at 18 global telecom PoPs (Emerging) · deep dive
AI Fabrik — Edge inference delivery network — modular token-factory sites, 6k+ B300 GPUs (2026), <20ms latency to tier-1/2 cities; incubated inside Gruve (Startup)

Agent Infrastructure · Sandboxes, durable execution, memory, and tool infrastructure for running autonomous agents in production 22 companies

Sandboxes & Code Execution

Modal — Serverless GPU compute + sandboxes for agent code execution (Major)
Cloudflare — Workers, Durable Objects, and Sandboxes — edge runtime for agents (Major)
Vercel — AI SDK, Sandbox, and Fluid compute for agent backends (Major)
E2B — OSS Firecracker microVM sandboxes for AI-generated code (Emerging, OSS)
Daytona — Agent-native sandbox runtime — fast cold starts, snapshot and restore (Emerging)
Fly.io — Fast-boot microVMs — per-agent isolation on Fly Machines (Emerging)
CodeSandbox (Together AI) — Sandbox SDK for code execution, acquired by Together AI (Emerging)
Runloop — Devboxes purpose-built for coding agents (Startup)

Durable Execution & Orchestration

Temporal — Durable execution — the reliability backbone for long-running agents (Major)
Inngest — Durable functions + AgentKit for multi-step agent workflows (Emerging)
Restate — Lightweight durable execution engine from the Apache Flink creators (Startup, OSS)
DBOS — Durable execution as a library on Postgres — no extra infra (Startup, OSS)

Memory & State

Redis — Semantic caching, vector search, and session state for agents (Major)
Mem0 — OSS memory layer — long-term, entity, and session memory for agents (Startup, OSS)
Zep — Agent memory built on temporal knowledge graphs (Graphiti) (Startup, OSS)
Letta — Stateful agents from the MemGPT research team (Startup, OSS)

Tools, Browsers & Protocols

MCP (Model Context Protocol) — Open standard for agent-tool connections — adopted industry-wide (Major, OSS)
Browserbase — Headless browser infrastructure for web-using agents (Emerging)
Composio — Managed tool and integration layer — hundreds of app connectors (Startup)
Arcade — Tool calling with delegated user auth for agents (Startup)
Exa — Search engine API built for LLMs and agents (Startup)
Firecrawl — OSS web crawling and extraction for LLM and agent pipelines (Startup, OSS)

Cluster Orchestration & Tenant Isolation · Kubernetes platforms, tenant cluster orchestration, GPU scheduling, AI factory software 42 companies

Kubernetes Platforms

Kubernetes (CNCF) — De facto standard, DRA, GPU device plugins (Leader, OSS)
Red Hat OpenShift — Enterprise K8s, GPU Operator, security hardening (Major)
Amazon EKS — Managed K8s on AWS, widely adopted in enterprise (Major)
Google GKE — Managed K8s on GCP, Autopilot mode (Major)
Azure AKS — Managed K8s on Azure, GPU node pools (Major)
Rancher (SUSE) — Multi-cluster K8s management (Major)
RKE2 — SUSE Rancher K8s, security-hardened, FIPS-compliant (Emerging, OSS)
Kamaji — OSS hosted control planes (Startup, OSS)
k0rdent — Cluster lifecycle management, GitOps (Startup, OSS)
Rafay — K8s operations platform, enterprise (Emerging)
Kubeadm — Bootstrap tool for K8s clusters, OSS standard (Major, OSS)
Mirantis — Enterprise Kubernetes (k0rdent, Lens) for GPU cluster management (Emerging)

GitOps & Delivery

Argo CD — Declarative GitOps continuous delivery for Kubernetes (Major, OSS)
Flux — CNCF GitOps toolkit for Kubernetes (Major, OSS)
Helm — Kubernetes package manager, chart standard (Leader, OSS)
Crossplane — OSS control plane framework, provision infra from K8s (Emerging, OSS)
Harness — AI-native CI/CD and platform engineering, pipeline automation (Emerging)

Tenant Cluster Orchestration

vCluster — Full K8s API server per team on shared infra. OSS, GPU-native (Major, OSS)
vCluster Standalone — Deploy the control plane directly on bare metal — no existing K8s required (Major)
vNode — Kernel-native workload sandboxing, Linux user namespaces, no VM overhead (Emerging)

AI Factory Platforms

NVIDIA DSX — Unified AI factory: DSX OS, Sim, MaxLPS, Reference Designs (Major)
KAI Scheduler — NVIDIA OSS K8s scheduler, topology-aware GPU placement (Emerging, OSS)
NVIDIA AICR — AI Cluster Runtime — canonical NVIDIA-accelerated K8s (Emerging, OSS)
NVSentinel — K8s-native GPU health monitoring and fault remediation (Emerging, OSS)
NVIDIA Infra Controller — Bare-metal provisioning for multi-tenant GPU infra (DSX OS) (Emerging, OSS)

GPU Scheduling

Run:ai (NVIDIA) — GPU virtualization, fractional GPUs, quota mgmt (Major)
NVIDIA GPU Operator — Automates GPU driver/plugin lifecycle on K8s (Major, OSS)
Kueue — K8s-native job queueing and gang scheduling (Emerging, OSS)
Volcano (CNCF) — GPU-aware batch scheduling for Kubernetes (Emerging, OSS)
Karpenter — K8s node auto-provisioner, GPU-aware, spot-friendly (Emerging, OSS)

Security, Policy & Identity

OPA / Kyverno — Policy engine and admission controller for Kubernetes (Major, OSS)
Falco — OSS runtime security and threat detection for K8s (Major, OSS)
Trivy — OSS container and IaC vulnerability scanner (Major, OSS)
Keycloak — OSS identity and access management, SSO (Major, OSS)
Okta — Enterprise identity platform, workforce and customer IAM (Leader)
Auth0 — Developer-friendly identity platform (Okta) (Major)

Workload Orchestration

Slurm (SchedMD) — HPC job scheduler, research standard (Major)
Slinky (CNCF) — Bridge between Slurm and Kubernetes (Startup, OSS)
Ray (Anyscale) — Distributed compute for training and serving (Major, OSS)
SkyPilot — Multi-cloud job launcher, least-cost spot (Startup, OSS)
dstack — OSS ML orchestration for multi-cloud GPU (Startup, OSS)
Kubeflow — ML pipelines on Kubernetes (Emerging, OSS)

MLOps & Training · Experiment tracking, distributed training, inference engines and observability tooling 27 companies

Training & Experiment Tracking

Weights & Biases — Experiment tracking, evals, registry. MLOps standard (Major)
MLflow — OSS ML lifecycle — tracking, registry, projects (Major, OSS)
Comet — ML experiment tracking and model monitoring (Emerging)
Snorkel — Programmatic data labeling and training data creation (Emerging)
Determined AI — Distributed training platform (HPE) (Emerging)
Braintrust — AI eval, logging and prompt management — LLM observability platform (Emerging)
Lightning AI — PyTorch Lightning — training, fine-tuning and deployment platform (Emerging) · deep dive
Surge AI — RLHF data labeling and human feedback for frontier model training (Emerging)
DataRobot — End-to-end ML and generative AI lifecycle platform (Major)

Inference Engines & Serving Frameworks

NVIDIA Triton — Multi-framework inference server, TensorRT-LLM (Major, OSS)
vLLM — OSS LLM inference, PagedAttention (Major, OSS)
NVIDIA Dynamo — Distributed inference for multi-node LLM (DSX OS) (Emerging, OSS)
llm-d (CNCF) — K8s-native disaggregated LLM inference (Startup, OSS)
KServe — K8s-native model inference platform, serverless serving (Major, OSS)
Ray Serve — Scalable model serving built on Ray (Major, OSS)
Modular — MAX inference engine and compiler stack (Emerging)
TGI (HuggingFace) — Text Generation Inference for HF models (Major, OSS)
SGLang — Fast LLM inference, structured generation (Emerging, OSS)

Observability & Evaluation

Arize AI — ML observability, LLM tracing, evals (Emerging)
Datadog — GPU + LLM monitoring, full-stack observability (Major)
Dynatrace — Full-stack AI observability, LLM monitoring (Major)
Prometheus — OSS metrics and alerting, K8s monitoring standard (Leader, OSS)
Grafana — OSS metrics, logs, traces. K8s infra standard (Major, OSS)
DCGM — NVIDIA Data Center GPU Manager — GPU telemetry and health (Major, OSS)
Elastic — Search and observability for AI infra log pipelines (Major)
Fiddler — ML monitoring and explainability (Emerging)
Evidently AI — OSS ML monitoring, data drift, model health (Emerging, OSS)

Storage & Networking · Parallel storage, object storage, vector databases, high-speed interconnects 31 companies

AI-Optimized Storage

WEKA — Parallel file system with GPU-Direct Storage (Major)
Vast Data — Universal disaggregated all-flash, AI-native (Major)
DDN — AI400X, Lustre-based, deep HPC heritage (Major)
Pure Storage — FlashBlade for AI training (Major)
MinIO — S3-compatible OSS object storage (Emerging, OSS)
Ceph — OSS distributed storage — object, block, and file (Major, OSS)
Databricks — Lakehouse platform, Unity Catalog, MLflow (Major)
Snowflake — Data cloud with Cortex AI, vector search (Major)
Hammerspace — AI data platform — global namespace, NVIDIA reference design storage (Emerging)
NetApp — Enterprise storage, ONTAP AI, GPU-Direct NFS for training pipelines (Major)
Scality — OSS S3-compatible object storage (RING) for AI data lakes (Emerging, OSS)

Vector & Analytical Databases

Turbopuffer — Vector search optimized for cold storage at scale (Startup)
LanceDB — OSS vector DB on Lance columnar format (Startup, OSS)
ClickHouse — OLAP DB for ML observability and log analytics (Major, OSS)
SurrealDB — Multi-model OSS database, AI backends (Startup, OSS)
PlanetScale — Serverless MySQL, metadata and feature stores (Emerging)

Network Silicon & Switches

NVIDIA Mellanox — InfiniBand, ConnectX-7, Quantum-X800. Dominant HPC interconnect (Leader)
Broadcom — Tomahawk 6, Jericho3-AI, 1.6Tbps Ethernet AI fabric (Leader)
Arista Networks — Etherlink AI switches, AI/ML scale-out (Major)
Upscale AI — Purpose-built AI networking — SkyHammer™ ASIC (scale-up) + NVIDIA Spectrum-X Ethernet fabrics (scale-out), open standards (Emerging)
Cisco — Silicon One, AI-optimized Ethernet fabric (Major)
Akamai — Edge compute and CDN, AI inference delivery (Major)
DriveNets — Disaggregated network OS, cloud-scale routing (Emerging)

Network Automation & Multi-Tenancy

Netris — NAAM platform — fabric automation and hardware-enforced tenant isolation for GPU clouds (Emerging)
Hedgehog — Open network fabric for AI/ML clusters (Startup, OSS)
Aviz Networks — AI-native network observability and analytics (Startup)

Optical & Emerging Interconnect

Coherent — 800G/1.6T optical transceivers (Major)
Lightmatter — Photonic interconnect chips for AI (Startup)
Ayar Labs — Optical I/O chiplets for in-package photonics (Startup)
Credo — Active electrical cable (AEC) and SerDes silicon for AI cluster interconnect (Startup)
WhiteFiber — Open fiber and networking infrastructure for AI data centers (Startup)
Lumentum — Optical transceivers and photonics for AI datacenter interconnect (Major)

Data Center & Bare Metal · Colocation operators, cooling, power, server OEMs, bare metal provisioning 24 companies

Colocation & DC Operators

Equinix — 260+ DCs globally, xScale AI campuses (Leader)
Digital Realty — 300+ DCs, PlatformDIGITAL, hyperscale builds (Leader)
Vantage — Hyperscale campuses, Stargate partner (Major)
QTS (Blackstone) — Mega-campus builds, 1GW+ pipeline (Major)
Telehouse — European colocation operator — London + Paris AI hubs (Emerging)
Polarise — European AI datacenter operator — sovereign GPU campuses (Startup)

Cooling & Power

Vertiv — Liquid cooling, thermal mgmt, power infra for AI racks (Major)
Schneider Electric — DC power distribution, cooling, DCIM (Major)
CoolIT Systems — Direct liquid cooling for AI racks (Emerging)
Crusoe Energy — Stranded/flared gas converted to AI compute power (Major)
Bloom Energy — Solid oxide fuel cells for clean DC power (Emerging)
Radiant — Portable nuclear microreactors (Kaleidos) for datacenter power (Startup)
Alva Energy — Power infrastructure and energy supply for AI datacenters (Startup)

Server OEMs

NVIDIA DGX — DGX B200/H100, SuperPOD, turnkey AI factory systems (Leader)
HPE — ProLiant Gen12, Cray EX, AI-optimized servers (Major)
Dell — PowerEdge AI Factory, liquid-cooled GPU servers (Major)
Supermicro — GPU servers, rack-scale, fast SKU cycles (Major)
Lenovo — ThinkSystem AI servers, high-density GPU platforms, global OEM (Major)
Celestica — Electronics manufacturing services — AI server hardware, DCS group (Emerging)
2CRSi — French OEM — energy-efficient, immersion-cooled GPU servers (Emerging)

Bare Metal Provisioning

vMetal — Bare metal provisioning & lifecycle (Metal3/Ironic, PXE) (Emerging)
NVIDIA Infra Controller — Bare-metal provisioning for multi-tenant GPU infra (Emerging, OSS)
Canonical MAAS — Metal as a Service — automated server provisioning (Emerging)
Ironic (OpenStack) — OSS bare metal provisioning service (Emerging, OSS)

Context layers

Applications · AI products, developer tools, vertical software, agent frameworks 25 companies

AI-Native Developer Tools

GitHub Copilot — AI pair programmer, VS Code & JetBrains (Leader)
Cursor — AI-native code editor, deep codebase context (Major)
Windsurf — AI-native IDE by Codeium, agentic coding (Emerging)
Cognition — Devin — AI software engineering agent (Emerging)
Replit — AI-powered cloud IDE (Major)
JetBrains — IntelliJ, AI Assistant — developer IDE platform (Major)

Agent & LLM Frameworks

LangChain — LLM orchestration, LangGraph agents (Major, OSS)
LlamaIndex — Data framework for LLM apps, RAG (Major, OSS)
CrewAI — Multi-agent orchestration (Emerging, OSS)
AutoGen — Multi-agent conversational systems (Microsoft) (Emerging, OSS)

Vertical AI

Harvey — AI for legal (Emerging)
Sierra — AI customer service agents (Emerging)
Perplexity AI — AI answer engine (Major)
Abridge — AI clinical documentation (Emerging)
ElevenLabs — Voice AI — text-to-speech, voice cloning, real-time audio (Emerging)
H Company — Enterprise AI agents, Paris — $220M seed, frontier agent research (Emerging)
Lovable — AI app builder — design to production in minutes (Emerging)
Factory — AI software engineering platform — automated code and PRs (Emerging)
Glean — Enterprise AI search and knowledge platform (Emerging)

Analyst, Media & Events

SemiAnalysis — Deep-dive GPU/AI infra research (Major)
The New Stack — Cloud-native, K8s, platform engineering (Major)
TLDR AI — Daily AI newsletter (Emerging)
KubeCon / CNCF — Premier cloud-native conference & ecosystem (Leader, OSS)
AI Engineer — AI engineering practitioners conference (Emerging)
GTC (NVIDIA) — NVIDIA annual AI & accelerated computing conf. (Major)

Models · Foundation models, open-weight releases, model hubs 11 companies

Frontier Model Labs

OpenAI — GPT-5.x, o-series reasoning, Sora (Leader)
Anthropic — Claude Opus 4.x, Sonnet 4.6 (Leader)
Google DeepMind — Gemini 3.x, AlphaFold (Leader)
Meta AI — Llama 4 open-weight, Muse Spark (Leader, OSS)
DeepSeek — V4 / R-series — open-weight reasoning (Major, OSS)
Mistral AI — European frontier lab, open-weight (Major, OSS)
xAI — Grok 4, Colossus 2 (~555k GPU cluster) (Major)
Cohere — Command A, enterprise RAG-optimized (Major)

Model Hubs & Local Inference

Hugging Face — Model hub, Transformers, datasets (Leader, OSS)
NVIDIA NGC — Pre-trained models, curated GPU containers (Major)
Ollama — Run LLMs locally, simple CLI (Emerging, OSS)

Chips · GPUs, custom ASICs, HBM memory, server systems 19 companies

GPU & Accelerators

NVIDIA — H100/H200/B200/GB200, CUDA, ~80% share (Leader)
AMD — MI300X/MI325X, ROCm (Major)
Intel — Gaudi 3, Ethernet-native (Major)
Cerebras — Wafer-scale engine (Emerging)
Groq — LPU, ultra-low latency inference (Emerging)
Google TPU — TPU v6e/v8, JAX/Pathways (Major)
FuriosaAI — Korean AI NPU, RNGD chip — transformer-optimized inference silicon (Emerging)
Rebellions — Korean AI chip startup, ATOM/REBEL NPU — pre-IPO (Startup)
d-Matrix — Corsair inference accelerator — digital in-memory compute for fast, efficient LLM serving (Startup)
Etched — Sohu — transformer-only ASIC betting on a single architecture for extreme throughput (Startup)
NextSilicon — Maverick intelligent compute accelerator for HPC and AI workloads (Startup)
Fractile — In-memory compute inference hardware — UK startup targeting radical inference speedups (Startup)

HBM & Memory

SK Hynix — HBM3E leader, ~95% GPU memory share (Leader)
Samsung — HBM3E, CXL memory (Major)
Micron — HBM3E, CXL pooling (Major)

Server Systems

NVIDIA DGX — DGX B200/H100, SuperPOD (Leader)
HPE — ProLiant Gen12, Cray EX (Major)
Dell — PowerEdge AI Factory (Major)
Supermicro — GPU servers, rack-scale (Major)

Energy · Power, cooling, data centers, bare metal provisioning 10 companies

Data Center Operators

Equinix — 260+ DCs globally, xScale AI campuses (Leader)
Digital Realty — 300+ DCs, PlatformDIGITAL (Leader)
Vantage — Hyperscale campuses, Stargate partner (Major)
QTS (Blackstone) — Mega-campus builds, 1GW+ (Major)

Cooling & Power

Vertiv — Liquid cooling, power infra for AI racks (Major)
Schneider Electric — DC power distribution, cooling (Major)
Crusoe Energy — Stranded gas to AI compute power (Major)
Bloom Energy — Solid oxide fuel cells for DC power (Emerging)
IREN — Digital infrastructure + clean power — Canadian AI/HPC compute campus (Emerging)
Hammerhead AI — RL-driven power orchestration — turns stranded data center power into AI capacity (Startup)

Link copied to clipboard