§ 10 PUBLIC COMPANIES TRACKER
The listed names in the AI compute stack.
172 publicly traded companies where AI compute is a material revenue driver, plus 13 frontier AI labs and 20 AI-native startups at the bottom. Organized into 17 layers — from the EUV scanner at the top to the labs and product startups that consume it all.
Editorial registry, not investment advice. Company names link to homepages; tickers link to Yahoo Finance. Curated quarterly. Last audited May 2026.
Current pressure points in the listed map.
A curated tape from the updated registry notes. Each chip is pulled from the public-company data below.
-
Accelerators & logic design NVIDIA NVDA - Q1 FY2027 Data Center revenue $75.2 B
- Q2 revenue guide $91.0 B +/- 2%
-
Accelerators & logic design Broadcom AVGO - Q1 FY2026 AI revenue $8.4 B, +106% YoY
- Q2 AI semiconductor revenue expected $10.7 B
-
HBM memory SK Hynix 000660.KS - ≈50% of HBM supply
-
Power, cooling & datacenter infra Vertiv Holdings VRT - FY2026 net sales guide $13.5 B-$14.0 B
- adjusted FCF $2.1 B-$2.3 B
-
Power, cooling & datacenter infra nVent Electric NVT - Record sales/orders
- $2.6 B backlog
- FY2026 organic growth guide 21%-23%
-
Hyperscalers, Chinese cloud & NeoClouds Microsoft MSFT - FY26 Q3 capex $31.9 B
- calendar-2026 capex expected around $190 B
- OpenAI’s largest cloud customer
-
Hyperscalers, Chinese cloud & NeoClouds Alphabet GOOGL - FY2026 capex guidance $180 B-$190 B
- Q1 2026 capex $35.7 B
-
Hyperscalers, Chinese cloud & NeoClouds Baidu BIDU - Q1 2026 AI Cloud Infra revenue RMB8.8B, +79% YoY
- GPU Cloud revenue +184% YoY
-
Hyperscalers, Chinese cloud & NeoClouds CoreWeave CRWV - Q1 FY2026 backlog $99.4 B
- FY2026 guide $12 B-$13 B revenue and $31 B-$35 B capex
-
Hyperscalers, Chinese cloud & NeoClouds Nebius NBIS - Q1 2026 revenue $399 M, +684% YoY
-
Server OEMs & ODMs Dell Technologies DELL - FY2026 AI-optimized server revenue $24.683 B
- FY2027 backlog opened at $43 B
-
Server OEMs & ODMs Wiwynn 6669.TW - AI-related products were more than half of FY2025 revenue
- Q1 2026 revenue NT$276.5 B, +62% YoY
EUV SCANNERS 1 LISTED
EUV scanners.
One vendor — ASML — ships the EUV scanners that print every leading-edge chip. The drive laser is private (Trumpf), the projection optics are private (Zeiss SMT, owned inside the Zeiss foundation), the source-collector mirrors are private. ASML is the only public way to own the EUV ceiling directly.
Open chapterWAFER FABS 8 LISTED
Wafer fabs.
TSMC sits at roughly nine-tenths of leading-edge logic capacity (defined as sub-7 nm-class production). Samsung and Intel are the only credible second sources at the bleeding edge; SMIC is China’s DUV-bound attempt. Mature-node fabs (UMC, GF, Hua Hong, Tower) matter for power, analog, and PMICs around AI systems — not for the FLOPS-defining die itself. Japan’s Rapidus is the consortium attempt to build a sovereign 2 nm node.
Open chapterTSMC
N2 / N3 fabs and CoWoS advanced packaging. Every NVIDIA, AMD, Apple, and Google accelerator starts here.
Samsung Electronics
Foundry (SF2 in qualification), the only end-to-end semis conglomerate. See HBM section for memory exposure.
Intel
18A in HVM, 14A in pilot. The US leading-edge backstop bet. Also ships Gaudi 3 / Falcon Shores accelerators (see Accelerators section).
GlobalFoundries
Mature-node specialty foundry — RF, power management, automotive. Not an AI logic story.
UMC
Mature-node Taiwanese foundry, second-largest in TW.
SMIC
China’s leading-edge attempt, DUV-only after export controls. The Huawei Ascend node.
Hua Hong Semiconductor
Mature-node Chinese foundry. Specialty processes.
Tower Semiconductor
Analog/RF and silicon-photonics specialty foundry; $1.3 B contracted 2027 SiPho revenue and $920 M SiPho/SiGe investment plan.
SILICON WAFERS & FAB MATERIALS 13 LISTED
Silicon wafers & fab materials.
Bare 300 mm silicon is a Shin-Etsu / SUMCO duopoly (each roughly a third of the market), with GlobalWafers and Siltronic the next tier. Specialty fluids and gases (Linde, Air Products, Air Liquide, Entegris) supply every fab on Earth. Photoresists are a Japanese near-monopoly (Tokyo Ohka Kogyo, JSR — now private under JIC, Shin-Etsu). MKS supplies vacuum and photonics subsystems for every tool. None of these names move on AI capex headlines, but every fab build needs them in the ground first.
Shin-Etsu Chemical
300 mm silicon wafers + photoresist materials. One half of the Shin-Etsu / SUMCO bare-wafer duopoly.
SUMCO
300 mm silicon wafers — the other half of the duopoly.
GlobalWafers
Third-largest 300 mm silicon wafer maker, Taiwan-listed. Operates fabs across the US, Asia, and Europe.
Siltronic
European-listed silicon-wafer maker. Fourth globally; focused on specialty and 300 mm.
Linde
Industrial gases — every fab’s electronic-grade nitrogen, helium, and specialty etchants.
Air Products & Chemicals
The other big bulk and specialty gases supplier into fabs.
Air Liquide
European peer to Linde and APD. Strong electronic-gases franchise across Asian fabs.
Entegris
Wafer-handling, fluid handling, and ultra-pure filtration for every leading-edge fab. The pick-and-shovel name with broadest tool-set coverage.
MKS Instruments
Vacuum subsystems, lasers, and photonics integrated into every fab tool. Atotech subsidiary covers specialty plating chemistry.
Element Solutions
Specialty chemicals for electronics — plating, solder, packaging interconnect. Heavy advanced-packaging exposure.
Tokyo Ohka Kogyo (TOK)
EUV and ArF photoresists. With JSR (now private), Shin-Etsu, and Sumitomo, completes the Japanese photoresist near-monopoly.
Resonac
Formerly Showa Denko Materials. CMP slurries, encapsulation, copper-foil substrates for advanced packaging.
Sumitomo Chemical
Specialty chemicals including photoresists and high-purity materials for semiconductors.
WAFER-FAB EQUIPMENT 21 LISTED
Wafer-fab equipment.
The Big Five — AMAT, LRCX, KLAC, Tokyo Electron, ASML — gate volume economics at every leading-edge fab. SCREEN is the genuine #2 in coater/developer alongside TEL. Test (Advantest, Teradyne, Cohu, Aehr) and inspection / metrology (KLA, Onto, Camtek, PDF Solutions) ride the AI test-intensity wave. Disco supplies the dicing, grinding, and polishing tools every advanced package walks through. The long tail (Axcelis ion implant, EBARA CMP, Ultra Clean / Ichor subassemblies, FormFactor probes) shows up everywhere.
Applied Materials
Broadest deposition / etch / CMP toolset. Sells across every node and every fab.
Lam Research
Etch and deposition leader, with deep franchises in 3D NAND and leading-edge logic patterning.
KLA
Process control and inspection. Yield management is their moat at every leading-edge node.
Tokyo Electron
Coater/developer co-leader with SCREEN. Strong etch and CVD. Tied at the hip with TSMC and Samsung.
SCREEN Holdings
Coater/developer and wet-process equipment. The actual second leg in coat/develop alongside TEL.
Disco Corporation
Dicing, grinding, and polishing tools. Every advanced package walks through Disco tools at some step.
Advantest
High-end ATE for AI accelerators and HBM. The clearest AI test-intensity play.
Teradyne
ATE — Advantest’s competitor across SoC and memory test, plus robotics.
Onto Innovation
Metrology and inspection — heavy advanced-packaging exposure.
Camtek
Inspection of HBM stacks and advanced packaging substrates.
FormFactor
Probe cards — the test sockets every accelerator wafer touches before dicing.
King Yuan Electronics
Outsourced test services — wafer and final test capacity. Sits between OSAT and ATE.
BE Semiconductor
Hybrid-bonding tools for fine-pitch 3D stacking. The bonder ramp is for HBM-next and reticle-stitched logic stacks, not HBM4 itself.
Veeco Instruments
Laser annealing, ion-beam deposition, and InP laser tools; more than $250 M of InP laser manufacturing orders, with Axcelis merger pending China SAMR.
Axcelis Technologies
Ion implantation specialist. Strong silicon-carbide (SiC) and mature-node exposure.
Ultra Clean Holdings
Subassemblies, modules, and precision cleaning services for the WFE OEMs. Direct AI-capex pull-through.
Ichor Holdings
Fluid delivery subsystems for fab tools. Sells into LRCX, AMAT, and KLAC; small-cap WFE pure-play.
PDF Solutions
Yield-analytics SaaS for fabs. Data-platform franchise with TSMC, Samsung, and the leading IDMs.
Aehr Test Systems
Wafer-level burn-in and reliability test. Niche on heterogeneous integration and high-power devices.
EBARA Corporation
CMP polishers and dry vacuum pumps — every fab. Quiet leverage to the leading-edge buildout.
Cohu
Test handlers, interfaces, and inspection systems adjacent to the ATE supply chain.
ADVANCED PACKAGING & SUBSTRATES 11 LISTED
Advanced packaging & substrates.
CoWoS capacity gated accelerator shipments through 2024 and early 2025 — TSMC’s expansions have eased it, but capacity is still allocated, not bought. ASE, Amkor, and JCET dominate OSAT; Kulicke & Soffa and ASMPT supply the bonding tools. The often-overlooked layer is the ABF substrate — Ibiden, Unimicron, AT&S, and Nan Ya PCB physically carry every NVIDIA accelerator package.
Open chapterASE Technology
Largest OSAT globally. Back-end packaging, test, and growing advanced-packaging capacity.
Amkor Technology
Second-largest OSAT. Arizona advanced-packaging fab opens 2027 with Apple as anchor.
JCET Group
Third-largest OSAT globally. The China-aligned packaging option; absorbed STATS ChipPAC.
Kulicke & Soffa
Wire-bonding equipment leader; advancing into thermo-compression and hybrid-bonding adjacencies.
ASMPT
Bonding-tool maker with a strong advanced-packaging franchise; competes with BESI on hybrid-bonding roadmaps.
Ibiden
ABF-substrate maker. Every NVIDIA accelerator package sits on an Ibiden substrate.
Unimicron
ABF-substrate maker — the other large supplier to AMD, Intel, and the broader accelerator chain.
AT&S
Austrian ABF-substrate maker — third leg alongside Ibiden and Unimicron. NVIDIA and Intel customer.
Nan Ya PCB
IC substrates and high-end PCBs for accelerators and networking silicon.
Chipbond Technology
Bumping, COG, and final-test specialist. The OSAT layer for driver ICs and dense packaging.
Powertech Technology
Memory back-end and test specialist — DRAM and HBM stack test capacity.
HBM MEMORY 3 LISTED
HBM memory.
Three suppliers globally: SK Hynix (≈50% share, NVIDIA’s primary), Samsung (qualifying HBM4 after a rough 2024), and Micron (Idaho HQ, ramping in Hiroshima). Pricing was up multiples through 2025 as smartphone DRAM bid for the same wafers. Memory is now ~30% of hyperscaler capex.
Open chapterSK Hynix
HBM3e and HBM4 leader; NVIDIA’s primary external memory supplier.
Samsung Electronics
HBM3e and HBM4 for AI memory; 1Q26 DS results were driven by HBM4 mass-product sales, with first HBM4E samples planned for 2Q26.
Micron Technology
US-domiciled DRAM and HBM. Capacity ramping in Hiroshima with Idaho HBM lines in build-out.
FLASH MEMORY & STORAGE 8 LISTED
Flash memory & storage.
The other half of the memory story, and the most parabolic public trade of the past year. NAND ASPs moved 2× off the 2023 bottom as supply discipline (capacity cuts across the SanDisk-Kioxia JV, SK Hynix, Samsung, Micron) collided with AI-driven demand: training-data lakes, KV-cache offload during long-context inference (Kioxia AiSAQ, SanDisk HBF — High Bandwidth Flash), and the structural shift toward all-flash enterprise storage. SanDisk’s Feb 2025 spin from Western Digital created the cleanest US-listed pure-play NAND name; Kioxia’s Dec 2024 Tokyo IPO did the same in Asia. HDD makers (Seagate, WDC) ride the same wave through hyperscaler nearline drives. The private AI-storage cohort — Vast Data, Weka, DDN, Hammerspace — is where the most interesting IPO pipeline sits.
SanDisk
Pure-play NAND flash and enterprise SSDs. Spun out of Western Digital in February 2025. HBF (High Bandwidth Flash) is the AI-inference product wedge — flash arranged like HBM for capacity-bound inference workloads.
Kioxia Holdings
Japanese NAND manufacturer (formerly Toshiba Memory); 50/50 BiCS Flash JV partner with SanDisk. AiSAQ — vector-database-on-flash for AI inference — is the differentiation wedge. IPO’d on TSE Dec 2024 after Bain Capital ownership.
Western Digital
Pure-play HDD operator after the SanDisk spin. Hyperscaler nearline drives at HAMR-class densities for training-data archives. The other half of the HDD duopoly with Seagate.
Seagate Technology
Pure-play HDD. Mozaic 3+ HAMR platform at 30 TB+ density. Sold out on hyperscaler nearline through 2026; the cleanest "AI training-data storage" trade on the market.
Pure Storage
Enterprise all-flash arrays. FlashBlade //EXA is the AI-training and inference-vector-database product line. Meta hyperscaler design-win is the marquee.
NetApp
Enterprise hybrid storage with deep all-flash and AI-data-pipeline exposure. ONTAP AI partnerships across NVIDIA, Lenovo, and the hyperscalers.
Silicon Motion
SSD controllers — the brains of nearly every consumer and enterprise NVMe SSD. The quiet pick-and-shovel beneficiary of the NAND price recovery.
Phison Electronics
Taiwanese SSD controller and NAND module specialist. Silicon Motion competitor with a deep customer book across consumer, industrial, and enterprise SSDs.
ACCELERATORS & LOGIC DESIGN 10 LISTED
Accelerators & logic design.
NVIDIA owns the accelerator customer. AMD is the only credible merchant GPU alternative at scale. Intel ships Gaudi 3 / Falcon Shores into the same market but its accelerator revenue is dwarfed by the foundry restructuring story (kept under Wafer fabs above). Broadcom and Marvell sell custom silicon to hyperscalers (TPU, Trainium, Maia, MTIA). Lattice and Rambus are the public-IP plays (FPGAs, memory interfaces). Cerebras is the public wafer-scale play; Groq, Etched, Lightmatter, MatX, Furiosa, Hailo, Tenstorrent, and SambaNova are the credible private bets.
Open chapterNVIDIA
Designs the AI accelerator everyone else buys. Doesn’t own a fab. Owns the customer.
AMD
MI400 / Helios series — the only credible merchant GPU alternative at hyperscale.
Broadcom
Hyperscaler ASIC partner (Google TPU, Meta MTIA) + datacenter networking silicon.
Marvell Technology
Custom ASICs (AWS Trainium, NeuronLink interconnect), high-speed optical PHYs, and DSPs.
Cerebras Systems
Wafer-scale accelerators. Inference cloud is the commercial wedge. Newly public in May 2026.
Qualcomm
On-device AI (Snapdragon NPUs) + AI200 / AI250 datacenter inference accelerators.
Arm Holdings
CPU IP underneath nearly every hyperscaler custom silicon (Graviton, Trainium, Maia, MTIA) plus NVIDIA Grace.
MediaTek
Mobile and edge AI; co-developing AI-PC Arm SoCs with NVIDIA (GB10 / DGX Spark).
Lattice Semiconductor
Small-form-factor FPGAs for edge AI, server BMC, and security adjacencies. The pure-play small-FPGA name.
Rambus
Memory-interface IP (HBM, DDR5, CXL controllers) and security IP. Royalty leverage to every leading-edge accelerator design.
EDA SOFTWARE 6 LISTED
EDA software.
Synopsys and Cadence are the duopoly on bleeding-edge digital signoff — nearly every advanced-node SoC routes through one or both. Synopsys integrated Ansys’s simulation tools after the 2025 close, deepening the multi-physics moat. Siemens (via Mentor) is the third leg, strongest in PCB, analog-mixed-signal, and multi-physics. Keysight covers RF + test-bench EDA. China is building its own stack — Empyrean and VeriSilicon are the public proxies. Highly recurring revenue, high margins, and the leading indicator is design-start volume, not capex.
Synopsys
EDA leader — every advanced-node digital chip taped out goes through Synopsys somewhere in the flow. Integrated Ansys’s multi-physics simulation stack after the 2025 close.
Cadence Design Systems
EDA co-leader — verification, signoff, and system-level design. Heavy AI-design and Cadence.AI exposure.
Siemens AG
Owns Mentor Graphics — third in digital EDA, strongest in PCB, analog-mixed-signal, and multi-physics.
Keysight Technologies
RF, electromagnetic, and high-speed digital design and test. EDA + test-bench bridge for AI networking silicon.
VeriSilicon
Chinese chip-design-services and IP house — the closest thing to a Synopsys/Cadence equivalent inside the China stack.
Empyrean Technology
Largest domestic Chinese EDA vendor. Sanctions-driven import substitution is the structural tailwind.
NETWORKING, INTERCONNECT & OPTICAL 14 LISTED
Networking, interconnect & optical.
Arista is the merchant-silicon datacenter network leader; Cisco is the incumbent; Ciena is the DCI long-haul play. Astera, Credo, MACOM, and Semtech sell the retimers and interconnect ICs that move data inside racks and between buildings. Lumentum and Coherent supply the lasers, EMLs, and VCSELs the transceiver vendors integrate. Innolight, Eoptolink, and Accelink are the China-listed volume leaders in finished optical modules. Celestica and Fabrinet do the assembly. The pre-IPO interconnect cohort — Enfabrica (AI fabric), Ayar Labs (co-packaged optics), Eliyan (chiplet interconnect) — is where the real architecture innovation is happening.
Open chapterArista Networks
Merchant-silicon datacenter switching leader. Riding the back-end AI fabric buildout.
Cisco Systems
Enterprise networking incumbent. Splunk + Silicon One are the AI angles.
Astera Labs
PCIe / CXL retimers and scale-up fabric silicon. NVLink Fusion ecosystem partner.
Credo Technology
Active electrical cables (AECs) and SerDes IP — the copper interconnect inside rack-scale systems.
Lumentum
Optical components — EMLs, DMLs, and VCSELs — that transceiver vendors integrate. 800G + 1.6T pipeline.
Coherent Corp
Optical components, lasers, and transceiver assembly. Today’s entity combines II-VI, Finisar, and the historical Coherent laser franchise.
MACOM Technology
High-speed analog ICs and optical components for telecom and datacenter.
Semtech
Tri-Edge / CopperEdge active copper interconnect; LoRa wireless. AI-fabric exposure via active cables.
Ciena
Long-haul and DCI (data-center interconnect) optical transport. Hyperscaler multi-region traffic pull-through.
Fabrinet
Optical and precision-optics contract manufacturer. Assembles NVIDIA Quantum InfiniBand modules.
Celestica
EMS contract manufacturer. The HPS (Hardware Platform Solutions) segment is heavily exposed to hyperscaler AI hardware.
Innolight
China’s leading optical transceiver vendor — 800G and 1.6T volume leader in finished modules.
Eoptolink Technology
Chinese optical transceiver vendor; one of the top three by ship volume globally.
Accelink Technologies
Chinese optical-components vendor with telecom and datacenter exposure.
POWER, COOLING & DATACENTER INFRA 17 LISTED
Power, cooling & datacenter infra.
The 2025–30 bottleneck. Inside the building, Vertiv, Eaton, Schneider, nVent, and Hubbell build the power and cooling chain; Modine, Daikin, and Munters supply the thermal management; Generac and the genset OEMs ship the backup power. EPC contractors (Quanta Services, MasTec, EMCOR) build the substations and pads outside. The turbine OEMs (GE Vernova, Mitsubishi Heavy, Siemens Energy) and standby-genset makers (Cummins, Caterpillar) are sold out through 2028.
Open chapterVertiv Holdings
Power and cooling chain inside the datacenter. CDU / busway / UPS / chillers. The most direct AI-cooling play.
Eaton
Electrical distribution — switchgear, busway, transformers. Heavy datacenter exposure.
Schneider Electric
Power distribution and prefab modular datacenter shells. Deep hyperscaler relationships.
nVent Electric
Electrical enclosures, thermal management, and liquid-cooling infrastructure for AI racks.
Hubbell
Electrical components — utility transmission gear plus the datacenter distribution backbone.
Modine Manufacturing
Thermal management — datacenter CDUs, chillers, and air-cooling at AI scale.
Generac Holdings
Standby power generation — datacenter gensets and behind-the-meter prime power adjacencies.
Daikin Industries
Global HVAC leader; AI-grade datacenter chillers and refrigerant-cycle solutions.
Munters Group
Datacenter air-management specialist — Swedish-listed pure-play on AI-cooling capex.
Quanta Services
EPC contractor — substations, transmission lines, datacenter site power. The grid build-out trade.
MasTec
Infrastructure construction — utility and grid-facing backlog plus pipelines and communications.
EMCOR Group
Mechanical and electrical contractor — datacenter HVAC and power systems.
GE Vernova
Gas turbines and power-generation equipment. Backlog sold out through 2028 after the GE spin.
Mitsubishi Heavy Industries
Gas turbines and nuclear systems. Asia’s gas-turbine supply.
Siemens Energy
Gas turbines, grid equipment, and transmission. The European power-build trade.
Cummins
Backup generators at every datacenter — and increasingly behind-the-meter prime power.
Caterpillar
Large standby and prime-power generators for datacenters and BTM gas plants.
POWER GENERATION & FUEL 19 LISTED
Power generation & fuel.
Nuclear is the cleanest hyperscaler PPA story: Constellation is restarting Three Mile Island for a Microsoft PPA (target COD ~2028), Talen sells Susquehanna power to AWS, Vistra bought Energy Harbor for the same playbook. Vogtle owners (Southern via SO) and IPPs (NRG) are the conventional plays. SMR developers (Oklo, NuScale, BWXT) are the longer-dated trades, with Holtec, TerraPower, X-energy, and Kairos still private. Centrus enriches the HALEU fuel; Cameco / Brookfield own Westinghouse 49/51. Uranium spot is tracked through Sprott; the miners are Cameco, NexGen, Energy Fuels, Uranium Energy. Bloom Energy is the public bet on behind-the-meter fuel cells. Cheniere and Energy Transfer feed the gas turbines.
Constellation Energy
Largest US nuclear operator. Restarting Three Mile Island (Crane Clean Energy Center) for a Microsoft PPA, target COD ~2028.
Vistra
Independent power producer — nuclear (post-Energy Harbor), gas, and retail. Heavy Texas datacenter exposure.
Talen Energy
Susquehanna nuclear plant + AWS Cumulus campus PPA. The original behind-the-meter nuclear-AI structure.
NextEra Energy
Largest US renewable developer plus Florida regulated utility. The diversified renewables-for-AI bet.
NRG Energy
Texas-heavy IPP with gas-fired generation and retail electricity. Direct hyperscaler-PPA exposure.
BWX Technologies
Naval nuclear reactors plus commercial SMR components.
Oklo
Small modular reactor developer. Aurora powerhouse design. Sam Altman–backed.
NuScale Power
Pressurized-water SMR developer. The first NRC-certified SMR design, restructuring after UAMPS cancellation.
Centrus Energy
Uranium enrichment — only US producer of HALEU fuel that SMRs and advanced reactors need.
Bloom Energy
Solid-oxide fuel cells for behind-the-meter datacenter power. Major partnership with Equinix; AEP and others bidding.
Cameco
Uranium miner. 49% co-owner of Westinghouse alongside Brookfield. The longest-dated nuclear trade.
Brookfield Asset Management
Owns 51% of Westinghouse (Cameco the other 49%) plus the largest private renewables and AI-infra portfolios.
Cheniere Energy
Largest US LNG exporter. Domestic gas pull-through into AI gas-turbine demand is the indirect AI play.
Energy Transfer
Natural-gas pipelines feeding the gas-turbine peakers and BTM plants behind datacenter sites.
First Solar
US solar-module manufacturer. The volume bet for AI renewables PPAs that need to be in-region.
Energy Fuels
US uranium miner plus rare-earth processing. Domestic-fuel-cycle leverage.
Uranium Energy Corp
US uranium ISR operator/developer; Burke Hollow commenced production in April 2026, processed through Hobson CPP licensed for up to 4M lbs U3O8/year.
NexGen Energy
Canadian uranium developer (Rook I / Arrow). One of the largest undeveloped high-grade deposits on Earth.
Sprott Physical Uranium Trust
Physical-uranium-backed trust. The cleanest pure-play on the spot price.
DATACENTER REITS 9 LISTED
Datacenter REITs.
The landlords. Equinix and Digital Realty hold the legacy colocation footprint and are converting fast to liquid-cooled, AI-grade sites. Iron Mountain is the sleeper because of its powered-shell pipeline. GDS and VNET are the China-listed datacenter operators. Asia-Pacific exposure is Goodman Group (AU), Keppel DC REIT and Mapletree Industrial (SG). DigitalBridge is the publicly-traded digital-infra holdco. The biggest names in the category — Vantage, Aligned, Stack, EdgeConneX, Compass Datacenters, plus pre-2022 take-privates CyrusOne, QTS, Switch — are all owned by infra funds.
Equinix
Largest colocation REIT — interconnect-dense urban sites. Converting metro footprint to liquid-cooled, AI-grade.
Digital Realty
Wholesale + colocation REIT — bigger megawatt footprint, faster AI conversion across mega-campuses.
Iron Mountain
Records-management REIT with a sleeper datacenter and powered-shell pipeline.
GDS Holdings
China’s largest carrier-neutral datacenter operator. SEA expansion via DayOne subsidiary; major AI tenant footprint.
VNET Group
Chinese carrier-neutral datacenter operator; 889 MW wholesale capacity in service, 623 MW utilized, and 452 MW under construction.
DigitalBridge Group
Digital-infrastructure asset manager — owns Vantage, DataBank, and a long roster of private DC operators.
Goodman Group
Australian industrial REIT pivoting hard into hyperscale datacenter development. Largest AU-listed AI-infra exposure.
Keppel DC REIT
Singapore-listed pure-play datacenter REIT. Asia-Pacific colocation footprint.
Mapletree Industrial Trust
Singapore industrial REIT with growing US datacenter portfolio (via Mapletree Investments).
HYPERSCALERS, CHINESE CLOUD & NEOCLOUDS 19 LISTED
Hyperscalers, Chinese cloud & NeoClouds.
The US hyperscalers (Microsoft, Amazon, Google, Meta, Oracle) are spending a combined ~$600 B/yr on AI infrastructure. The Chinese cloud trio (Alibaba, Tencent, Baidu) are the parallel AI infrastructure operators — different chips, different models, same buildout pace. SoftBank is the Stargate participant + Arm owner + OpenAI investor. NeoClouds (CoreWeave, Nebius, IREN, Applied Digital, Cipher, Hut 8, Core Scientific, TeraWulf, Cleanspark) and former-Bitcoin operators are the levered GPU-compute trades for investors who can’t hold MSFT/META at the size they’d like. Lambda, Crusoe, and Together AI lead the private cohort.
Open chapterMicrosoft
Azure + the OpenAI partnership (through 2030) + Foundry. The most aggressive hyperscaler AI capex profile.
Alphabet
In-house TPU since 2015, the largest scale-up domain on Earth. DeepMind + Gemini frontier models, see AI labs section.
Amazon
AWS Trainium custom silicon + Anthropic partnership ($8 B+). The other hyperscale-AI flywheel.
Meta Platforms
MTIA accelerators and ~2 GW Hyperion campus. Owns FAIR + Llama, the largest open-weight family.
Oracle
Stargate consortium with OpenAI, SoftBank, and MGX. OCI AI clusters at multi-GW scale.
Alibaba
Alibaba Cloud + Qwen models (see AI labs section). The largest Chinese AI infrastructure operator by capex.
Tencent
Tencent Cloud + Hunyuan models. The other half of the Chinese hyperscaler duopoly.
Baidu
Baidu AI Cloud + Ernie models + Kunlun custom silicon. The Chinese AI-first cloud bet.
SoftBank Group
Owns Arm (75%+), is the lead Asian Stargate participant, and a major OpenAI investor. The public proxy for Masa’s AI infrastructure bet.
CoreWeave
NeoCloud — leveraged GPU rental at hyperscale. Microsoft and OpenAI customer concentration. IPO’d Mar 2025.
Nebius
Post-Yandex AI cloud with a new owned Pennsylvania AI factory securing up to 1.2 GW of power/land.
IREN
NASDAQ-listed operator (Australia-founded) — crypto-mining capacity pivoting to HPC / AI cloud. West Texas + Canada footprint.
Applied Digital
Datacenter developer + GPU-as-a-service; contracted capacity exceeded 1 GW after the Polaris Forge 3 hyperscaler lease.
Cipher Mining
Bitcoin miner with announced HPC pivot. Black Pearl site has a Fluidstack / Google-backed AI buildout.
Hut 8
Bitcoin miner, post US Bitcoin Corp merger. Smaller HPC pivot relative to IREN and Cipher.
Core Scientific
Bitcoin-mining infrastructure converting to HPC. CoreWeave is the anchor tenant on the new sites.
TeraWulf
Bitcoin-and-HPC operator; 60 MW critical IT HPC capacity energized for Core42 at Lake Mariner, with 168 MW Abernathy JV targeted for Q4 2026.
CleanSpark
Bitcoin miner with public HPC ambitions and a deep Georgia / Wyoming power footprint.
Galaxy Digital
Crypto financial services + the Helios West Texas HPC campus (CoreWeave anchor tenant on the AI side).
SERVER OEMS & ODMS 13 LISTED
Server OEMs & ODMs.
Where accelerators turn into deployable systems. Dell and HPE dominate enterprise; Super Micro is the AI-native specialist. Lenovo is the global enterprise leader most US investors underweight. The Taiwanese ODMs (Foxconn, Quanta, Wiwynn, Inventec, Pegatron, Gigabyte, MiTAC, ASRock, Acer) increasingly bypass the OEMs and build NVL72-class racks direct for hyperscalers.
Dell Technologies
Enterprise server leader, largest non-ODM AI server vendor. PowerEdge XE9680 / XE9712 + reference designs.
Hewlett Packard Enterprise
Cray HPC + Juniper acquisition. Enterprise-AI positioning, HPE GreenLake AI Factory.
Super Micro Computer
AI-native server vendor — fastest to ship NVIDIA reference designs.
Lenovo
Global enterprise server leader with the broadest AI server portfolio outside the US. Heavy hyperscaler exposure too.
Foxconn (Hon Hai)
Rack-scale AI server assembly (NVL72 integrator) and the largest contract manufacturer on Earth.
Quanta Computer
ODM rack assembly — the other half of NVL72 system production. Heavy MSFT / Meta exposure.
Wiwynn
Hyperscaler-focused ODM (Wistron spin-out). Designs and builds for Meta, MSFT, AWS.
Inventec
Server ODM with growing AI rack-scale business, particularly for hyperscalers and tier-2 cloud builders.
Pegatron
Contract manufacturer expanding from consumer electronics into AI server ODM work.
Gigabyte Technology
Taiwanese AI server vendor — fast at NVIDIA reference design adoption, growing hyperscaler share.
MiTAC Holdings
Server ODM (MiTAC Computing, TYAN). Hyperscaler-direct builds plus the Intel server platform partnership.
ASRock
ASRock Rack — AI server boards and rack-scale systems for tier-2 cloud and on-prem AI builders.
Acer
PC and server vendor with growing AI server unit (Acer ATS) and Altos enterprise line.
FRONTIER AI LABS 13 LABS
Frontier AI labs.
The labs that consume the rest of the stack. Almost all are private — public exposure is mostly indirect through Microsoft (OpenAI), Amazon (Anthropic), Alphabet (DeepMind), Meta (Llama / FAIR), and the Chinese cloud trio. Model families listed below; version numbers are omitted because they age in weeks. Names link out to each lab’s homepage.
GPT · Codex
Largest commercial compute buyer on Earth. Stargate consortium with Oracle, SoftBank, and MGX. Microsoft-allied through 2030.
Claude Opus · Sonnet · Haiku · Claude Code
Consistently top of OpenRouter usage on Sonnet. Compute-constrained on every dimension. Amazon ($8 B) and Google ($3 B+) backed.
Gemini Pro · Gemini Flash · Gemini Nano
In-house TPU stack since 2015, largest scale-up domain on Earth. Gemini Flash is the OpenRouter cheap-tier workhorse.
Grok · Grok Code
Acquired by SpaceX in 2026; now a wholly-owned subsidiary. Memphis Colossus 2 cluster + integration with X for distribution. No public exposure since the parent is private.
Llama · Llama Guard · Movie Gen
Largest open-weight family. ~2 GW Hyperion campus is the in-house AI infrastructure flex.
Mistral · Codestral · Pixtral
European foundation-model lab; on-prem and sovereign-AI angle. Nvidia, Microsoft, and a16z backed.
Command · Embed · Rerank
Enterprise-focused; Oracle and Fujitsu reseller partnerships. Sovereign-AI deployments outside the US.
DeepSeek · DeepSeek-Coder
Quant-fund-backed (High-Flyer). The R-series reasoning models broke open the cost curve in early 2025; sustained OpenRouter top-5 since.
Qwen · Qwen-VL · Qwen-Coder · Qwen-Max
Largest Chinese open-weight family. The export-favorite for non-US deployments.
GLM · ChatGLM
Tsinghua-spun, Beijing-based. The Chinese state-aligned foundation lab and a 2024 export-control target.
Kimi · Kimi-VL
Long-context pioneer (200k+ tokens). Alibaba-backed; the Kimi consumer chat app is one of the most-used Chinese assistants.
MiniMax · Hailuo · Talkie
Video generation and emotional TTS are the wedges; the consumer Talkie app is large outside China too.
Yi · Yi-VL
Kai-Fu Lee’s lab. Open-weight Yi family; aggressively priced on enterprise API.
AI-NATIVE STARTUPS 20 STARTUPS
AI-native startups.
The companies built on top of the labs — the products users actually touch. Coding tools, app builders, AI search, voice and video, vector infra, model hubs. All private, all growing faster than any prior software cohort. Names link out to each homepage. Valuations are 2025–26 reporting; treat as directional, not gospel.
AI-native code editor. The default IDE for AI-assisted software engineering; fastest-growing developer product in history.
Cloud IDE + AI agents that build and ship full apps. The "prompt to running software" bet.
Devin — the autonomous software engineer. The agent-replaces-developer thesis at hyper-growth.
Frontier coding agent with ultra-long-context models. The "agent that actually reads your whole codebase" pitch.
Generate full-stack apps from a prompt. Fastest path from $0 to $100 M ARR in software history.
AI-native web development from StackBlitz. Direct Lovable competitor for the prompt-to-app market.
AI-native search engine plus the Comet browser. The default Google alternative for technical queries.
Model hub + open-source AI infrastructure. The default discovery and distribution layer for every open-weight release.
Managed vector database — the default RAG and retrieval-memory backend for AI applications at scale.
Vector and full-text search built on object storage — roughly 10× cheaper than traditional vector databases. Cursor, Anthropic, Notion, Cognition, Linear, Harvey, and Ramp run on it; Cursor alone searches 1T+ code chunks.
Enterprise AI search and assistants. Connects to every SaaS app a company uses.
AI for big-law. Foundation-model-powered research, drafting, and review deployed at top firms.
Compounding-intelligence platform for private credit and PE — screen, diligence, underwrite, and monitor deals on top of a firm’s own institutional memory.
Bret Taylor’s AI customer-experience platform. Replaces call-center agents at enterprise scale.
AI voice synthesis and cloning. The default TTS for podcasts, audiobooks, and game studios.
AI music generation from text prompts. The largest consumer AI-music platform.
AI video generation + creative tools. Gen-4 is the production-grade text-to-video model.
Consumer-first text-to-video. The fastest-growing AI-video product on the App Store.
AI-generated corporate video — text-to-spokesperson avatars. The L&D and training-video category leader.
AI avatars + multilingual video dubbing. Strong creator-economy and SMB marketing traction.