Report
AI Infrastructure Stack
Where the momentum actually is — ten layers, seventy stocks, twelve months of price action.
How to read this
Ten layers of the AI buildout — from the power plants at the bottom to the apps at the top. Seven representative public-market names per layer, ranked by trailing 12-month price return. A green pill means positive; a red pill means negative; the bigger the pill, the larger the move.
What this is not: not investment advice, not a forecast, not a recommendation. A snapshot of where capital has rotated over the trailing year. Use it to understand which layers are working and which are not — then do your own work on individual names.
The stack at a glance
Layers ranked by median 12-month return. Bar length = positive count (out of 7).
Power & Energy
What powers the AI data centers. Nuclear, gas, renewables, fuel cells.
| # | Ticker | Name | Role in the stack | 1Y |
|---|---|---|---|---|
| 1 | BE★ | Bloom Energy | Solid-oxide fuel cells for DC backup and on-site baseload | +1259% |
| 2 | GEV | GE Vernova | Gas turbines, transmission, grid equipment | +127% |
| 3 | CCJ | Cameco | Uranium fuel supply chain | +102% |
| 4 | TLN | Talen Energy | Nuclear baseload (AWS Susquehanna PPA) | +31% |
| 5 | NEE | NextEra Energy | Largest US renewables developer + grid utility | +22% |
| 6 | CEG | Constellation Energy | Nuclear fleet (Microsoft Three Mile Island restart) | −11% |
| 7 | VST | Vistra | Independent power producer with nuclear fleet | −14% |
AI Compute Chips
The silicon doing the work.
| # | Ticker | Name | Role in the stack | 1Y |
|---|---|---|---|---|
| 1 | INTC★ | Intel | Gaudi accelerators + foundry comeback narrative | +418% |
| 2 | AMD | AMD | MI300 / MI350 GPUs + EPYC CPUs | +261% |
| 3 | MRVL | Marvell | Custom AI silicon + DSPs for hyperscalers | +182% |
| 4 | AVGO | Broadcom | Custom XPUs + Tomahawk switching silicon | +78% |
| 5 | ARM | Arm Holdings | IP licensing across edge + data center | +69% |
| 6 | NVDA | NVIDIA | AI GPU monopoly | +63% |
| 7 | QCOM | Qualcomm | Edge AI + on-device inference | +27% |
Semi Equipment & Foundry
Who makes the silicon. The bottleneck of the bottleneck.
| # | Ticker | Name | Role in the stack | 1Y |
|---|---|---|---|---|
| 1 | MU★ | Micron | HBM memory stacked beside AI accelerators | +608% |
| 2 | LRCX | Lam Research | Etch + deposition equipment | +226% |
| 3 | AMAT | Applied Materials | Broadest wafer-fab equipment portfolio | +145% |
| 4 | KLAC | KLA | Process control + defect inspection | +121% |
| 5 | TSM | TSMC | Foundry of record for NVDA, AMD, AAPL, AVGO | +103% |
| 6 | ASML | ASML | EUV lithography monopoly | +95% |
| 7 | ENTG | Entegris | Specialty materials + consumables to fabs | +64% |
Networking & Interconnect
How chips talk to each other inside the data center. Ethernet is overtaking InfiniBand.
| # | Ticker | Name | Role in the stack | 1Y |
|---|---|---|---|---|
| 1 | CIEN★ | Ciena | Optical transport for AI back-end fabric | +569% |
| 2 | CRDO | Credo Technology | Active Electrical Cables + retimers | +173% |
| 3 | HPE | HP Enterprise | Aruba + Juniper (DC networking +383% post-deal) | +87% |
| 4 | CSCO | Cisco | DC networking, AI orders ramping | +81% |
| 5 | ANET | Arista Networks | 800G / 1.6T Ethernet switching | +46% |
| 6 | APH | Amphenol | High-speed interconnects + assemblies | +38% |
| 7 | CALX | Calix | Access networking (broadband / fiber-to-home) | −14% |
Optical / Photonics
Long-haul bandwidth — the AI bottleneck nobody outside the industry talks about.
| # | Ticker | Name | Role in the stack | 1Y |
|---|---|---|---|---|
| 1 | LITE★ | Lumentum | 800G / 1.6T pluggable transceivers | +1043% |
| 2 | AAOI | Applied Optoelectronics | 400G / 800G optics, datacom-heavy mix | +800% |
| 3 | COHR | Coherent | Optical components, lasers, industrial photonics | +345% |
| 4 | GLW | Corning | Optical fiber + bend-resistant cable for DCs | +265% |
| 5 | FN | Fabrinet | Contract optical manufacturing | +198% |
| 6 | MTSI | MACOM Technology | RF + optical semiconductors | +193% |
| 7 | IPGP | IPG Photonics | Industrial fiber lasers | +73% |
Data Center Power & Cooling
Physical plant — racks, UPS, chillers, immersion and liquid cooling. Boring infrastructure, big rerate.
| # | Ticker | Name | Role in the stack | 1Y |
|---|---|---|---|---|
| 1 | POWL★ | Powell Industries | Electrical equipment for DC build-out | +330% |
| 2 | VRT | Vertiv Holdings | Cooling + power distribution + rack systems | +203% |
| 3 | NVT | nVent Electric | Liquid cooling enclosures + electrical connection | +138% |
| 4 | MOD | Modine Manufacturing | DC cooling specialist (50-70% segment growth) | +135% |
| 5 | AAON | AAON | HVAC for data centers + commercial | +22% |
| 6 | ETN | Eaton Corporation | Electrical infrastructure for DCs | +13% |
| 7 | CARR | Carrier Global | HVAC + cooling (broader portfolio dilutes AI bid) | −19% |
Data Centers & Colocation
The physical buildings. REITs and AI-cloud pure-plays.
| # | Ticker | Name | Role in the stack | 1Y |
|---|---|---|---|---|
| 1 | APLD★ | Applied Digital | Pure-play AI hosting | +437% |
| 2 | GDS | GDS Holdings | China DC hyperscaler | +49% |
| 3 | DBRG | DigitalBridge | DC investment platform (private + public DC assets) | +35% |
| 4 | IRM | Iron Mountain | Records storage pivoting to DC + colocation | +22% |
| 5 | EQIX | Equinix | Interconnection-rich global colo | +19% |
| 6 | CRWV | CoreWeave | Pure-play AI cloud (post-IPO) | +15% |
| 7 | DLR | Digital Realty | Hyperscale DC REIT | +9% |
Servers & Storage
Where the AI training data lives. HDD / SSD pure-plays have been the surprise winners.
| # | Ticker | Name | Role in the stack | 1Y |
|---|---|---|---|---|
| 1 | WDC★ | Western Digital | Pure-play nearline HDD post-flash split | +799% |
| 2 | STX | Seagate | HAMR drives at 36TB / disk capacity | +573% |
| 3 | DELL | Dell Technologies | AI servers + storage + edge | +106% |
| 4 | PSTG | Pure Storage | All-flash arrays for AI training workloads | +35% |
| 5 | NTAP | NetApp | Hybrid-cloud storage | +19% |
| 6 | SMCI | Super Micro Computer | AI server systems (accounting overhang lifted slowly) | −32% |
| 7 | NTNX | Nutanix | Hyperconverged infrastructure | −42% |
Cloud Hyperscalers
The clouds running everyone's AI workloads. The capex spend funding most of this stack.
| # | Ticker | Name | Role in the stack | 1Y |
|---|---|---|---|---|
| 1 | GOOGL★ | Alphabet | GCP + DeepMind / Gemini | +133% |
| 2 | BIDU | Baidu | China hyperscaler + Ernie | +54% |
| 3 | AMZN | Amazon | AWS + Anthropic equity stake | +26% |
| 4 | ORCL | Oracle | OCI + Stargate footprint expansion | +14% |
| 5 | BABA | Alibaba | Aliyun + Qwen / Tongyi models | +11% |
| 6 | META | Meta Platforms | Largest single-customer GPU buyer + Llama | −6% |
| 7 | MSFT | Microsoft | Azure + OpenAI stake + Copilot | −9% |
AI Software & Apps
Apps sitting on top of the stack. The weakest layer — investors have rotated out of app-layer SaaS.
| # | Ticker | Name | Role in the stack | 1Y |
|---|---|---|---|---|
| 1 | DDOG★ | Datadog | Observability for AI workloads | +85% |
| 2 | CRWD | CrowdStrike | AI-driven endpoint security | +39% |
| 3 | PANW | Palo Alto Networks | AI security platform | +24% |
| 4 | PLTR | Palantir | AIP / Foundry | +7% |
| 5 | CRM | Salesforce | Agentforce + Data Cloud | −38% |
| 6 | ADBE | Adobe | Creative AI (Firefly) | −39% |
| 7 | NOW | ServiceNow | Now Assist | −50% |
Cross-layer takeaways
The four uniformly positive layers (7 / 7)
- Semi Equipment & Foundry — every name positive, median +121%
- Optical / Photonics — every name positive, median +266%
- AI Compute Chips — every name positive, median +78%
- Data Centers & Colocation — every name positive, median +35%
The pattern: the deeper you go into physical infrastructure, the more uniformly positive the layer. Capital has rotated toward the silicon, the equipment that makes the silicon, and the optical components that connect the silicon — and away from the apps sitting on top of it all.
Mid-tier (5–6 of 7 positive)
- Networking & Interconnect (6 / 7) — only Calix negative, and that's because its business is consumer broadband, not data center
- DC Power & Cooling (6 / 7) — only Carrier negative, because its HVAC portfolio is too diluted to benefit cleanly from the AI bid
- Servers & Storage (5 / 7) — drag from SMCI (accounting overhang) and NTNX (hyperconverged stagnating); meanwhile WDC and STX have both done over 5×
- Cloud Hyperscalers (5 / 7) — capex burden weighing on MSFT and META despite the AI build
- Power & Energy (5 / 7) — extreme dispersion: BE +1259% (fuel cells) vs CEG and VST negative on "AI demand already priced in"
The weak layer
AI Software & Apps (4 / 7). The only layer where most names are below the line. The three biggest decliners in the entire 70-stock report all sit here: NOW −50%, ADBE −39%, CRM −38%. Investors are paying for picks-and-shovels infrastructure, not for application-layer SaaS — at least not yet.
Biggest single-stock winner
Bloom Energy +1259%
Solid-oxide fuel cells, layer 1 (Power & Energy).
Biggest single-stock loser
ServiceNow −50%
AI app-layer fatigue, layer 10 (Software & Apps).
Data source. All 12-month returns are price returns from Finviz "Perf Year" field, retrieved 2026-05-19. Equal-weighted within each layer. No dividends, no transaction costs, no taxes.
Methodology. Tickers were selected as the most representative public-market plays per layer; some names plausibly fit multiple layers (e.g. SMCI as both compute and server integrator) — they are assigned to their dominant business. Lists are not exhaustive; capped at seven names per layer for legibility.
Not investment advice. This is a snapshot of where capital has moved over the trailing twelve months across the AI infrastructure stack. It is not a forecast. It is not a recommendation. It is not financial, tax, or legal advice. Past returns do not predict future returns. Do your own due diligence before acting on any individual name.