Companies

AI Infrastructure Companies

221 active companies tracked across 8 stack layers.

L1 · Energy & DC Infra

Aligned

Hyperscale-tier colocation operator with 548 MW capacity, focused on AI-ready data centers.

Aligned Data Centers

Liquid-cooled hyperscale build prime. Adiabatic + immersion deployments at scale. Macquarie-backed. Energy-efficiency narrative differentiator vs traditional REIT colo.

Asetek

AIO liquid cooling on the consumer side; datacenter pivot from 2024 with In-Rack CDU products. Recent enterprise-AI strategy revival vs prior consumer focus.

Bloom Energy

Solid-oxide fuel-cell power systems. AI-datacenter angle: on-site behind-the-meter generation that bypasses 5-7yr grid interconnect queues. Named in 2025-2026 hyperscaler PPA announcements.

Caterpillar

On-site power generation — gas/diesel gensets, prime + backup power for datacenters.

Compass Datacenters

Build-to-suit hyperscaler datacenters. Brookfield + Ontario Teachers backed. Single-tenant 50-200 MW campuses; tracked in any "who is building for hyperscaler X" question.

Constellation Energy

Largest US nuclear operator; Three Mile Island (Crane) restart for Microsoft (~835MW).

CoolIT Systems

Direct-to-chip liquid cooling specialist. OEM relationships with the big server vendors (Dell, HPE, Lenovo, Supermicro). Acquired by HMT in 2024; remains operationally independent.

CoreSite

Major U.S. colocation operator, acquired by Equinix in 2021 and folded into its global platform.

Cummins

Power generation — gensets, gas engines, prime/backup power for datacenters.

CyrusOne

Colocation and interconnection specialist, now part of Digital Realty Trust after 2021 acquisition.

Digital Realty

Largest colo REIT by footprint. PlatformDIGITAL fabric ties multi-tenant + hyperscaler power-shell builds. Pre-leased pipeline a tracked watch-point for AI demand visibility.

Digital Realty Trust

Global leader in data center REITs with 300+ facilities, serving hyperscale and enterprise customers across all major regions.

Eaton

Electrical management conglomerate. Datacenter segment provides switchgear, UPS, busway, and large-scale power distribution. AI cluster MW-scale build-out is a named tailwind in earnings.

Equinix

Interconnection-dense colo prime. xScale joint venture for hyperscale build. Higher revenue-per-MW vs Digital Realty due to interconnect mix.

GE energy spin-off; gas turbines + grid + electrification. AI-datacenter angle: substation + grid interconnect timeline (gas turbines are the fastest dispatchable peak source for hyperscaler power requests).

Generac

Backup power systems. Datacenter segment for diesel + natural-gas standby generators. AI-cluster MW-class backup at 99.999% availability requirements is a growth tailwind.

Iceotope

Chassis-level precision liquid cooling (PLC) — sits between air and immersion. Major Schneider Electric strategic partnership. Edge / regulated-data use cases where full immersion is impractical.

Iron Mountain

Records-management pivot to datacenter REIT. Frankfurt / Virginia / Phoenix datacenter campuses. Hyperscaler-class pipeline disclosed; underweight vs DLR / EQIX but growing fastest.

Mitsubishi Heavy Industries

Gas turbines (Mitsubishi Power); order book sold out into ~2028.

Modine Manufacturing

Thermal management OEM. Datacenter segment for CDU (coolant distribution unit) + chillers. Pivoted from auto thermal to datacenter as the priority business; renamed segment to Climate Solutions.

Munters Group

Climate solutions; datacenter business grew significantly with AI cluster heat-density tailwind. Air-side handlers + direct-to-chip liquid systems. Strong European hyperscaler exposure.

NuScale Power

SMR developer — NuScale Power Module (light-water SMR), first NRC-certified SMR design.

nVent Electric

Connection + protection products. Datacenter segment includes liquid-cooling infrastructure (CDUs, manifolds) following the EriksenLab + Trachte acquisitions. Direct-to-chip and immersion both addressed.

Oklo

SMR developer — Aurora powerhouse (fast-fission microreactor); data-center PPAs.

Powell Industries

Custom-engineered electrical equipment — medium-voltage switchgear and power control systems for data center and industrial power distribution.

QTS

Colocation and hyperscaler campus builder, taken private by Blackstone.

Schneider Electric

Global energy + automation player. APC by Schneider is the legacy datacenter UPS brand; EcoStruxure is the modern stack. Direct competitor to Vertiv across the rack-to-utility power chain.

Siemens Energy

Heavy-duty gas turbines for datacenter power; record order backlog.

Stack Infrastructure

DigitalBridge-owned hyperscale builder; APAC + EMEA + Americas footprint. Pre-power-secured land bank the operational moat.

Switch

High-performance colocation data center operator with a mixed real estate/tech model.

Talen Energy

Independent power producer; Susquehanna nuclear PPA with Amazon (~1.92GW).

TerraPower

Gates-backed SMR developer — Natrium sodium-cooled fast reactor + molten-salt storage.

Vantage Data Centers

Global colocation operator with 314 MW capacity, serving hyperscalers and enterprises across Europe and North America.

Vertiv

Power + thermal infrastructure for hyperscale + colocation data centers. UPS / switchgear / busway / rack PDU / liquid cooling. The most-mentioned name on hyperscaler datacenter capex commentary.

Vistra

Power generator with nuclear fleet (Comanche Peak); AI-datacenter power deals.

X-energy

SMR / TRISO-fuel developer — Xe-100 HTGR; Amazon-backed; Dow Seadrift project.

L2 · Optical & Networking

Amphenol Corporation

Top-tier AEC connectors/assemblies (FCI-brand QSFP-DD/OSFP active cables) for AI GPU clusters.

Arista Networks

Cloud-networking incumbent. AI-spine + leaf switch deployments at the top hyperscalers. EOS software the named differentiator vs Cisco / Juniper. NVIDIA Spectrum-X partnership.

Astera Labs

Connectivity silicon for AI infrastructure — PCIe/CXL retimers and smart cable modules for GPU scale-up fabrics.

Ayar Labs

Optical I/O for chip-to-chip + scale-up. TeraPHY + SuperNova product line. Intel + AMD + GlobalFoundries strategic partners. Co-packaged optics alternative for scale-up vs Ethernet scale-out.

Celestial AI

IP and systems company specializing in optical interconnects for AI scale-up, coupling photonic devices into interposers.

Ciena

Optical networking systems — coherent DCI and transport gear that connects data centers at scale.

Cisco

Networking incumbent. Silicon One ASIC family the AI-fabric play vs Broadcom Tomahawk. Splunk acquisition closed 2024 added observability. Strong enterprise-AI on-prem narrative.

Coherent Corp.

Photonics, lasers, and optical materials — indium-phosphide lasers and components for 800G/1.6T transceivers and CPO.

Cornelis Networks

Omni-Path 200 / 400 successor (spun out of Intel 2020). InfiniBand-class fabric without NVIDIA Mellanox lock-in. Strong DOE / national-lab deployment base.

Credo Technology

Active electrical cables (AECs) + retimer + serdes ICs. AEC adoption inside racks the cheaper alternative to optical for sub-7m runs. 800G / 1.6T product cycle.

Effect Photonics

Player in on-chip optical interconnect market for AI infrastructure.

Enfabrica

Accelerated Compute Fabric (ACF) — NIC-meets-switch silicon. Pre-revenue with funded customer trials at major AI labs. AI-fabric-NIC alternative to NVIDIA BlueField.

Eoptolink Technology

Optical transceiver manufacturer — high-volume 800G/1.6T modules for AI cluster interconnect.

Innolight Technology

High-volume optical transceiver maker — 800G/1.6T pluggable modules for AI data center interconnect.

Juniper Networks

Networking incumbent under pending HPE acquisition (regulatory review ongoing 2026). AI-Native Networking + Mist AIOps positioning. Apstra Intent-Based for AI fabric.

Lightmatter

Provides optical interconnect solutions and photonic engines for AI data centers, pushing chiplet-based architectures.

Lumentum Holdings Inc.

Photonics components — lasers and optical components for datacom transceivers and co-packaged optics.

MACOM Technology Solutions Holdings, Inc.

Analog & mixed-signal semiconductors — laser drivers, TIAs, and high-speed components for optical and RF networking.

Microchip Technology

META-DX2C retimer-based reference designs for 800G AECs supporting 112G PAM4 signals.

Molex

High-speed 112G/224G AEC assemblies as lower-cost alternative to passive copper for AI capacity upgrades.

Nubis

Offers optical engine chiplets for single-wavelength connections (acquired by Ciena in Oct 2025).

OpenLight

Player in on-chip optical interconnect market for AI infrastructure.

Ranovus

Develops interoperable optical chiplet technology and laser manufacturing, focusing on Ethernet standard interoperability for CPO.

Scintil

Provider of optical interconnect solutions for next-wave CPO scaling.

TE Connectivity

100G/lane AECs powered by Marvell DSPs and 1.6T AEC solutions for AI workloads.

Xscape Photonics

Develops optical engines for AI scale-up networks as part of the CPO scaling wave.

L3 · HBM & Packaging

Absolics

Supplier of glass substrates for advanced packaging that boost routing density and reliability of leading-edge AI chips.

Amkor Technology

US-based OSAT. Arizona advanced-packaging facility under CHIPS Act funding; positioned for US-sourced HBM + chiplet packaging — geopolitical hedge against TSMC concentration.

ASE Group

Leading pure-play OSAT with critical 2.5D/3D packaging capabilities for AI accelerators, GPUs, and FPGAs.

ASE Technology

Worlds largest OSAT (outsourced semiconductor assembly + test). FOCoS-Bridge packaging is the merchant alternative to TSMC CoWoS — capacity ramp watched for cost-of-AI relief.

Deca Technologies

Technology specialist in fan-out wafer-level packaging (FOWLP) building US-based open innovation platforms for AI and HPC.

JCET

Top global OSAT specializing in advanced packaging for high-performance modules and AI chips.

Micron Technology

HBM third entrant (~20% share). US-strategic supplier; CHIPS Act direct beneficiary. HBM3E 12-Hi stack the current product anchor; HBM4 roadmap announced for 2026.

Rambus

CXL Memory Initiative delivering controller chips for CXL expansion, pooling, and fabric switching

Samsung Electronics

HBM second-source (~30% share). NVIDIA qualification delays on HBM3E remain a tracked watch-point. AMD MI300 / Intel Gaudi confirmed customer mix.

SK Hynix

HBM market leader (~50% share). HBM3E volume production lead; HBM4 sample shipments. Primary supplier into NVIDIA H100/H200/B200 + B100 platforms.

SPIL (Siliconware Precision)

Established OSAT that opened a new AI-focused advanced packaging plant in Taichung in 2025.

TSMC

CoWoS advanced packaging — the constrained step for AI accelerator + HBM integration. CoWoS-S / -L / -R / SoIC capacity expansion is the most-tracked supply-side number in 2026 AI infrastructure.

Wolley Inc.

CXL memory expansion controller startup enabling 32–64GB/s bandwidth for AI workstation LLM runtime

L4 · Chips

Alphawave Semi

High-speed connectivity IP — SerDes, chiplets, and custom silicon for AI / datacenter interconnect.

Altium

PCB design software vendor (Altium Designer), now a subsidiary of Renesas Electronics

AMD

MI300 / MI325 / MI350 Instinct accelerator family — the dominant merchant alternative to NVIDIA in AI training silicon. ROCm software stack maturity is the primary watch-point.

Ansys

Simulation and reliability software for chip and system design; acquired by Synopsys in 2024

Applied Materials

Wafer-fab equipment — deposition, etch, CMP; broadest WFE portfolio.

Arm Holdings

CPU architecture and IP licensing — the instruction set and core designs behind data center CPUs (Grace, Graviton) and edge AI.

ASML

EUV / High-NA lithography — the sole EUV supplier gating leading-edge AI-chip fabrication.

Blaize

Graph Processing Unit (GPU) architecture for edge AI inference

BrainChip

Akida neuromorphic processor for ultra-low-power edge inference

Broadcom Inc.

Custom AI silicon and networking — co-designs hyperscaler accelerator ASICs (e.g. Google TPU) and supplies Ethernet switch and SerDes silicon for AI fabrics.

Cadence Design Systems

EDA leader — digital + analog implementation, simulation, and verification tools spanning the chip-design flow.

Cerebras

Chips

Cerebras Systems

Wafer-scale Engine — single-chip alternative to multi-chip GPU clusters. Filed for IPO in 2024 (deferred). Real production deployments in scientific / pharma + select hyperscaler inference.

Etched

Transformer-only inference ASIC (Sohu) — bets the whole die on the transformer architecture for far higher throughput-per-dollar.

Furiosa AI

South Korean NPU with power-efficient architecture for inference

Graphcore

IPU (Intelligence Processing Unit) for enterprise AI inference

Groq

LPU inference-only accelerator; deterministic-latency architecture. GroqCloud commercial inference service. Sample point for inference-specialised non-NVIDIA silicon path.

Infineon Technologies

Power semiconductors (SiC/GaN/Si) for datacenter power delivery.

Intel

Tofino Ethernet ASIC IP (from Barefoot 2019 acquisition, then announced to wind-down 2023 but P4 ecosystem persists). Gaudi accelerator family + Ethernet-native networking pitch.

Intel Foundry

Intel's contract foundry courting AI-chip customers as a TSMC second source.

KLA

Process control + yield management (inspection/metrology) for advanced nodes.

Lam Research

Etch + deposition equipment, critical for HBM/3D and advanced logic.

Marvell Technology, Inc.

Custom compute and connectivity silicon — designs hyperscaler AI ASICs (e.g. AWS Trainium) plus optical DSPs and SerDes.

MathWorks

Provides HDL Coder and deep learning toolboxes for FPGA/ASIC design workflows

Navitas Semiconductor

GaN + SiC power ICs for high-density datacenter power (NVIDIA 800V partner).

NeuReality

NR1-P AI inferencing accelerator chip for data centers

Novarm

Maker of DipTrace, a niche PCB design tool and schematic editor

NVIDIA

The dominant AI accelerator vendor — GPUs (H100/B200/GB200), NVLink, and the CUDA software stack that anchors AI training and inference.

onsemi

SiC + power semiconductors for automotive and datacenter power.

Rebellions

Korean AI inference chip startup. Atom chip + REBEL roadmap. Merged with Sapeon (SK Telecom AI chip subsidiary) in 2024. Sovereign-Korea-AI angle on inference silicon.

SambaNova Systems

Reconfigurable Dataflow Architecture (RDU) — appliance + cloud model. Strong enterprise + government deployments. Pivoted toward inference-of-large-models from training focus.

Samsung Foundry

Logic foundry absorbing AI-chip overflow demand as TSMC saturates.

Semtech

PMDs (physical medium devices), copper + optical short-reach SerDes. AI cluster intra-rack copper interconnect product line. Sienna acquisition gave LoRa IoT.

Siemens EDA

EDA (ex-Mentor Graphics) — Questa/UVM verification, Tessent DFT, Calibre physical verification.

Synopsys

EDA leader — synthesis, place-and-route, verification, and the largest commercial silicon IP portfolio.

Tenstorrent

Wormhole + Blackhole AI chips on RISC-V cores. Open-source software stack (TT-Metalium). Jim Keller-led architecture; named in multiple sovereign-AI government compute initiatives.

Tokyo Electron

Coater/developer, etch, deposition WFE; #2 by revenue.

Wolfspeed

Silicon-carbide (SiC) substrates + power devices for high-voltage power.

Zuken

Japan-based global PCB/MCM/HIC design software vendor focused on electronic design

L5 · Training & Inference

Anyscale

Commercial Ray; distributed training + inference platform.

Applied Digital

HPC datacenter operator transitioning to AI cloud SKU; CoreWeave ND lease.

Baseten

Training & Inference

Celestica

EMS / ODM whitebox networking supplier. Hyperscaler 400G / 800G switch builds — direct sole-source disclosed in earnings. AI-driven networking revenue the named growth segment 2024-2026.

Cloudflare

Workers AI = edge inference across 300+ PoPs; AI Gateway.

CoreWeave

AI-infrastructure neocloud peer; the public-comp reference for SpaceX/xAI compute-as-a-service positioning ahead of the IPO.

Crusoe

Training & Inference

Crusoe

Stranded-energy GPU cloud; 200MW Abilene TX site for Oracle/OpenAI Stargate.

Databricks

Lakehouse + Mosaic AI; enterprise customer for end-to-end model train/serve governance. Largest private-market valuation in the data-cloud-for-AI category. Pre-IPO; financials limited.

Dell Technologies

AI server + rack-scale systems (PowerEdge XE / GB200 NVL72 integrator).

Fireworks AI

Inference platform optimized for open-weight LLMs and multi-modal.

Flex

Contract manufacturing — AI server + datacenter power/rack integration.

Gretel

Synthetic-data API for LLM fine-tuning. Acquired by NVIDIA March 2025 (~$320M).

HPE

Slingshot fabric (from Cray HPE acquisition) — Ethernet-based HPC fabric for exascale supercomputers. AI-supercomputer wins (El Capitan, Frontier) the named credibility anchor.

Jabil

Contract manufacturing / rack-scale AI hardware integration.

Labelbox

Self-serve labeling platform + Alignerr expert network for LLM evals.

Lambda

GPU cloud + on-prem workstations; major H100/H200 rental capacity.

Meta Platforms

Largest single Ethernet-for-AI buyer; Ultra Ethernet Consortium driving member. AI Research SuperCluster + MTIA accelerator deployments use Ethernet fabric not InfiniBand.

MOSTLY AI

Synthetic tabular data for finance/healthcare where real data is PII-locked.

Nebius Group

Yandex spinout. NVIDIA-backed funding 2024. Finland + Kansas City GPU clusters.

Parallel Domain

Synthetic-data for AV/embodied AI. Toyota, Woven, Waabi customers.

Replicate

API-first inference for community-hosted models.

Scale AI

Largest RLHF/data-labeling vendor. Meta $14.3B investment Sept 2025.

Super Micro Computer

AI server and rack systems — high-density GPU servers and liquid-cooled, rack-scale systems for AI clusters.

Surge AI

Premium-labeling competitor to Scale; PhD-level RLHF for Anthropic/OpenAI.

Together AI

Inference-as-a-service for open-weight models (Llama, DeepSeek, Qwen).

Vultr

Independent GPU cloud (H100/MI300X); 32 PoPs; AMD Instinct partnership.

L6 · Foundation Models

1X Technologies

Humanoid robots (NEO) + embodied-AI models.

AI21 Labs

Jamba SSM/Transformer hybrid open-weights family; enterprise RAG focus.

Alibaba (Qwen)

Qwen2.5 / Qwen3 open-weights series — leading non-Western open-weights frontier.

Anthropic

Foundation-model + Claude product layer; enterprise team selling to compliance-sensitive verticals. Direct comparable to OpenAI for enterprise governance + measurement use cases.

Apptronik

Apollo humanoid robot platform.

Cohere

Enterprise-focused foundation-model lab; Command R+, Embed, Rerank.

DeepSeek

DeepSeek-V3 / R1 open-weights reasoning models; redefined cost-of-frontier Jan 2025.

Figure AI

Humanoid robots + Helix vision-language-action (VLA) model.

Google DeepMind

Frontier model lab — Gemini family; Alphabet's consolidated frontier research unit.

Mistral AI

European open-weights + frontier-API lab; Mistral Large 2, Codestral, Mixtral.

Moonshot AI (Kimi)

Kimi long-context frontier-API lab; CN-market closed-API competitor to GPT/Claude.

OpenAI

Foundation-model + product layer; enterprise tier sells to F500 with governance / compliance gates. Reference customer for measuring enterprise AI adoption KPIs.

Physical Intelligence

VLA foundation-model lab (π0, π0.5); $2.4B valuation Nov 2024.

Skild AI

General-purpose robotics brain / cross-embodiment foundation model; $300M Series A.

Stability AI

Stable Diffusion 3 / SD3.5 / Stable Video; canonical open-weights image+video lab.

Unitree Robotics

Quadruped + humanoid robots (G1 / H1).

xAI

Owner of the Colossus supercomputer (Musk-affiliated); trains Grok on the 100K+ GPU Memphis cluster physically hosted/operated in the SpaceX orbit.

L7 · Middleware

Arize AI

ML/LLM observability + evaluation platform (Phoenix open-source tracing).

Braintrust

LLM eval + observability platform for AI product teams.

Browserbase

Headless-browser infrastructure that lets agents drive browsers at scale.

Chroma

Developer-first OSS vector DB; default in LangChain/LlamaIndex tutorials.

CrewAI

Multi-agent orchestration framework for role-based agent crews.

Daytona

Secure, ephemeral dev environments used as agent sandboxes.

deepset

Haystack open-source LLM/agent framework + deepset Cloud.

E2B

Open-source code-execution sandbox for AI agents; 1B+ sandboxes.

Helicone

LLM observability + gateway — logging, caching, cost tracking.

HiddenLayer

Detects and defends against attacks on machine-learning models.

Lakera

Real-time guardrails against prompt injection and LLM/agent attacks.

LangChain

Agent/LLM orchestration framework (LangChain, LangGraph) + LangSmith eval/observability.

Langfuse

Open-source LLM engineering platform — tracing, evals, prompt management.

Letta

Stateful agent-memory framework (formerly MemGPT) with OS-style memory tiers.

LiteLLM

Open-source LLM gateway/proxy — unified API + spend tracking across 100+ models.

LlamaIndex

RAG/data-framework leader; document parsing + indexing for enterprise LLM apps.

Mem0

Open-source memory layer for AI agents; exclusive memory provider for the AWS Agent SDK.

Modal

Serverless compute for AI/ML; ~$4.65B valuation.

OpenRouter

Unified LLM API gateway / model marketplace routing across providers.

Palo Alto Networks

Security vendor building the AI-control-plane (Prisma AIRS); absorbed Protect AI + Portkey.

Pinecone

Managed vector DB category leader. Notion AI / Shopify / Gong customers.

Portkey

LLM gateway — routing, caching, guardrails, FinOps for production AI.

Prompt Security

Runtime protection and governance for enterprise GenAI and agents.

Protect AI

AI/ML security platform acquired by Palo Alto Networks.

Qdrant

Rust-based OSS vector DB; X (Twitter), Bayer, Disney customers.

Runloop

Managed runtimes/sandboxes for coding agents.

Unstructured

PDF/HTML/PPTX → LLM-ready chunks; ETL for unstructured data.

Weaviate

Open-source vector DB w/ managed cloud; hybrid search + multi-modal.

Zep

Temporal knowledge-graph memory for agents (Graphiti).

Zilliz

Commercial company behind Milvus offering Zilliz Cloud, a fully-managed vector database with the proprietary Cardinal search engine.