Every public ticker that builds, ships, powers or sells AI compute. Click any node in the map below to open its quote, AI score and chart on Fintwit.
🔴 = capacity-constrained (single-source, 18+ month lead time, or sold out into 2027+). Percentages are dollar-share approximations within each stage. 133 US-listed tickers mapped across 8 stages.
Hover any node for ticker, stage, and constraint reason. Multi-listing rows (e.g. AAPL/AMZN/MSFT in-house Arm) open a picker. Foreign-only listings (TPE:, KRX:, HKG:) are shown for context but are not clickable on Fintwit. Updated quarterly.
AI stocks by stage
The AI compute supply chain has eight discrete stages. Each stage contains a handful of dominant public companies — most of which are bottlenecks, not commodities. The list below names every US- tradable ticker in each stage with the company description and stage-share weight, so you can quickly find AI infrastructure stocks, AI semiconductor stocks, hyperscaler stocks, data-center REIT stocks, power-generation AI stocks, and more.
1. Atoms
Electronic-grade is tight; commodity poly is not.
Gallium & Germanium
🔴 constrainedChina export controls Aug 2023.
InP substrates
🔴 constrainedPhotonics laser bottleneck.
Magnets for cooling fans, motors.
Interconnect, busbars, transformers.
High-purity quartz
🔴 constrainedSpruce Pine, NC nearly sole-sourced.
Photoresist & specialty chemicals
🔴 constrainedEUV photoresist near-monopoly.
Industrial / specialty gases
Ne, Xe, NF3, WF6.
Si wafers (300mm)
🔴 constrained300mm wafer slots tight.
2. Equipment
EUV lithography
🔴 constrainedSingle-source EUV; High-NA limited.
Process control / metrology
Onto on advanced packaging.
Test equipment (ATE)
🔴 constrainedHBM test slots tight.
Adv. packaging tools
🔴 constrainedThe CoWoS bottleneck.
3. Foundries
TSMC (N3, N2, CoWoS)
🔴 constrainedSole-source for AI accelerators.
Ramping; not yet at AI scale.
Restricted from leading nodes.
4. Memory
SK Hynix (HBM3E, HBM4)
🔴 constrainedSole HBM3E supplier to Nvidia for periods.
Samsung Memory
🔴 constrainedCatching up on HBM3E qualification.
Micron (HBM3E, HBM4)
🔴 constrainedSold out 2025/2026.
5. Packaging
Single tightest bottleneck in AI.
Arizona facility for Apple/AMD.
600584·002156·002185
6% of stageChina OSAT trio.
ABF substrates
🔴 constrainedCapacity tight for AI sockets.
6. Chip design
Accelerators
Nvidia (Hopper, Blackwell, Rubin)
🔴 constrainedAllocation-rationed.
AMD (MI300, MI325, MI350)
Broadcom (Google TPU, Meta MTIA)
🔴 constrainedTPU v5/v6 design partner.
Marvell (Amazon Trainium ASIC)
Tenstorrent / Cerebras / Groq / SambaNova
CPU
Gaining share fast in AI servers.
Powers Grace, Graviton, Cobalt.
Being acquired by SoftBank.
Networking
Nvidia (NVLink, Spectrum-X)
🔴 constrainedInfiniBand + scale-up.
Broadcom (Tomahawk, Jericho, Bailly)
🔴 constrainedEthernet AI fabric king.
PCIe / CXL / Ethernet retimers.
Photonics
Coherent (transceivers, SiPh)
🔴 constrainedNVIDIA $2B March 2026.
Lumentum (transceivers, EML)
🔴 constrainedNVIDIA $2B March 2026.
Marvell (optical DSPs, Polariton)
300502·300308
14% of stageAyar / Lightmatter / POET
(private/POET)
1% of stageIn-package optics, photonic compute.
Power ICs
Monolithic Power
🔴 constrainedLost some Blackwell sockets to Infineon.
Won Nvidia Blackwell Vcore.
Vertical-power-delivery niche.
7. System integration
Server / ODM
Largest GB200 rack assembler.
TPE:3231·6669
12% of stageDell (PowerEdge XE, AI factories)
Pegatron / Compal / Asus / MiTAC
4938·2324·2357·3706
6% of stageCooling
Boyd / Cooler Master / Asetek / CoolIT / JetCool / Accelsius
(mostly private)
20% of stageTPE:3017·3324
15% of stageThermal modules for GB200.
Connectors
NVIDIA socket connectors.
8. Data center
Hyperscalers
Microsoft (Azure / OpenAI workload)
Alphabet (Google Cloud + DeepMind)
Tencent / Alibaba / ByteDance / Baidu
China demand is separate supply chain.
REITs
Digital Realty
🔴 constrainedSwitch / CyrusOne / QTS / Compass
All taken private.
Vantage / Stack / EdgeConneX
Aligned / DataBank / Cologix
Power gen
Constellation
🔴 constrainedNuclear baseload.
Comanche Peak nuclear + gas.
Talen Energy
🔴 constrainedSusquehanna nuclear.
Renewable + Duane Arnold restart.
Regulated utilities (DUK·D·SO·ETR·DTE·XEL·EVRG)
Bloom Energy
🔴 constrainedOracle, AEP, Brookfield deals.
Oklo / NuScale / BWXT (SMRs)
Future.
GE Vernova (turbines, transformers)
🔴 constrained80GW gas turbine backlog into 2029.
EPC
9-yr data center backlog.
GE Vernova (transformers, HVDC)
🔴 constrainedSchneider Electric
🔴 constrainedSiemens Energy
🔴 constrainedHitachi Energy
🔴 constrainedHVDC.
Powell Industries
🔴 constrainedSwitchgear.
Hammond Power Solutions
🔴 constrainedDry-type transformers.
Argan (Gemma EPC)
🔴 constrainedGas-plant EPC.
Generac / Cummins / Caterpillar (gensets)
Backup.
Frequently asked questions
What are the best AI stocks to buy right now?
The most popular AI infrastructure names are NVIDIA (NVDA) for accelerators, TSMC (TSM) for foundry capacity, Broadcom (AVGO) for custom ASICs and networking, AMD for CPUs and MI-series accelerators, Micron (MU) and SK Hynix for HBM memory, ASML for EUV lithography, and the hyperscalers (MSFT, GOOGL, AMZN, META, ORCL). Power and cooling exposure runs through Vertiv (VRT), Eaton (ETN), GE Vernova (GEV), Constellation Energy (CEG) and Vistra (VST). The map above shows how every one of them connects to the rest of the stack and which ones are capacity-constrained.
What is the AI compute supply chain?
The AI compute supply chain is the end-to-end path that turns raw materials into a running AI workload: raw materials → cap- equipment makers → foundries → memory → advanced packaging → chip designers → system integrators → data centers. A bottleneck at any stage limits how fast the industry can scale. Today the tightest stages are EUV lithography (ASML), CoWoS advanced packaging (TSMC, BESI), HBM memory (SK Hynix, Micron) and grid power (CEG, VST, GEV).
How often is this list updated?
Quarterly. Stage-share weights, capacity-constraint flags, and ticker assignments are refreshed each quarter based on the companies’ latest filings, earnings transcripts, and verified industry sources.