Neocloud Brief · neocloud stocks · what is a neocloud · neocloud ETF
Neocloud stocks: the public companies renting out the AI buildout
A neocloud sells what AI actually needs — high-end GPUs by the cluster. The category topped $25B in 2025 revenue and is forecast to approach $400B by 2031. Here’s the whole public universe, sorted into the three layers that matter.
01The 30-second read
Fast read
A neocloud is a cloud provider built almost entirely around renting high-end GPUs — GPU-as-a-Service — for AI training, inference, and agentic workloads. Small against the hyperscalers, but growing at a pace the cloud market has never seen: $9B in Q4 2025 alone (+223% YoY), >$25B for full-year 2025, forecast to approach $400B by 2031.
- Cloud layer — CoreWeave and Nebius: compute, software, and multi-year contracts
- Campus / power layer — Applied Digital, plus former miners pivoting to HPC: IREN, Hut 8, TeraWulf, Cipher Mining
- Supply side — the silicon and memory underneath, including the DRAM memory ETF
The through-line: demand is real, but every name on the list spends far ahead of revenue. Contracts are signed years out; cash, power, and delivery arrive later. That gap — not demand — is what the category trades on.
02What “neocloud” actually means
A neocloud is a cloud provider that specializes in accelerated compute rather than general-purpose cloud. The distinction matters in two directions:
Versus hyperscalers
- AWS, Azure, and Google Cloud sell breadth — storage, databases, networking, hundreds of services
- Neoclouds sell scarcity: high-end GPUs, fast interconnect, and the software to run AI workloads efficiently — often at prices that undercut hyperscaler GPU rates
Versus data-center REITs
- A REIT rents square footage and power
- A neocloud’s customer is buying compute: the accelerated instance, the cluster, the training run
The category is best defined by which of four bottlenecks a company controls — and a company that controls one bottleneck should not be valued like one that controls several.
| Bottleneck | What it means |
|---|---|
| Power | Interconnection queues, megawatts, gas or nuclear behind-the-meter |
| Delivery | Converting contracted megawatts into live data centers on schedule |
| GPU supply | Allocation from Nvidia — and increasingly AMD |
| Software | Orchestration, scheduling, utilization, developer tooling |
03The public universe: three layers
Layer 1 — Cloud layer (closest to a true neocloud)
| Ticker | What it sells | Anchor facts |
|---|---|---|
| CRWV | CoreWeave — GPU cloud + software, sold on multi-year contracts | Q1 2026 revenue $2.078B; backlog $99.4B (Mar 31, 2026); >1 GW active power, >3.5 GW contracted |
| NBIS | Nebius — full-stack AI cloud (GPU + platform software) | Nvidia partnership supporting >5 GW of Nvidia systems; up to 1.2 GW power/land in Pennsylvania; $17.4B Microsoft + $3B Meta deals |
Layer 2 — Campus / power layer (power-site conversion trades)
| Ticker | The pivot | Anchor facts |
|---|---|---|
| CORZ | Core Scientific — miner → AI colocation | ~1.1 GW contracted and $24B+ base contracted revenue; $14B+ AMD deal (530 MW → up to 2.5 GW); 437 MW billable |
| GLXY | Galaxy Digital — Helios campus, West Texas | 526 MW IT / ~800 MW gross contracted to CoreWeave on 15-year leases; Phase I delivered Jul 2026; >$1B average annual revenue |
| IREN | IREN Ltd — miner → AI cloud | >$4B contracted ARR; $2.8B in new multi-year cloud services (Jul 2026); Microsoft-anchored |
| APLD | Applied Digital — HPC campus developer | $36B total contracted long-term lease value |
| HUT | Hut 8 — miner → AI data centers | $9.8B 15-yr Beacon Point lease (1 GW campus fully commercialized, 704 MW contracted); options to ~$25.1B; $7B River Bend lease with Anthropic/Fluidstack |
| WULF | TeraWulf — miner → HPC hosting | $12.8B in AI contracts |
| CIFR | Cipher Mining — miner → HPC | $9.3B HPC pipeline; AWS leases, Fluidstack, Google AI hosting; 1 GW West Texas; 2.5 GW gas-backed power |
| BTDR | Bitdeer — miner → HPC | Earlier-stage pivot |
Layer 3 — Supply side
| Ticker | What it is | Why it matters here |
|---|---|---|
| DRAM | Roundhill Memory ETF | The HBM/memory basket underneath every GPU cluster — the bottleneck supplying the bottleneck |
| NVDA | Nvidia | Sells the GPUs, invests in the customers, backstops capacity (see the risks below) |
Private names to watch (IPO or secondary candidates): Crusoe, Lambda, Together AI, Vast.ai, RunPod, Groq, Nscale, Civo, Vultr, Fluidstack.
Prices, mid-Sep 2026 (as reported): CRWV $81.08 · NBIS $222.13 · IREN $45.99 · APLD $25.20 · WULF $16.11 · CIFR $16.79 · DRAM $59.11. These move fast — snapshots, not quotes.
04The bull case
1 · The demand curve isn’t in question
- Neocloud revenue hit $9B in Q4 2025 alone, +223% YoY; >$25B FY2025
- Forecasts cluster around ~$400B by 2031; AI labs are signing capacity years ahead
2 · Contract visibility is unusually high
- CRWV’s $99.4B backlog, APLD’s $36B contracted lease value, HUT’s $9.8B leases, IREN’s $4B+ ARR
- Long-dated, signed obligations from creditworthy counterparties — rare for growth names
3 · Power is the moat — and the miners own it
- Former miners bring energized sites, interconnection rights, and power contracts
- Converting a 1 GW campus to AI hosting can beat greenfield on time and cost — the re-rating reason
4 · Validation from the top of the stack
- Nvidia partner deals (Nebius, CoreWeave) and hyperscaler leases (AWS–Cipher, Microsoft–IREN, Anthropic–Hut 8)
- External confirmation that the compute is genuinely needed
05The bear case
1 · Capex outruns revenue, everywhere
- CRWV: ~$14–16B 2026 capex vs a $12–14B revenue guide; Nebius: $20–25B capex on a $3.0–3.4B guide
- A 1,024-GPU H100 cluster needs roughly 75% utilization just to cover costs — small misses compound
2 · Funded with debt, dilution — and vendor money
- Debt loads, ABS facilities, and equity issuance are the norm
- Circularity: Nvidia holds equity in customers (its CoreWeave stake was ~1.21% at IPO) while supplying and backstopping them — revenue, debt, and investment trace to the same few counterparties
3 · Concentration cuts both ways
- The biggest contracts come from a handful of hyperscalers and AI labs
- One renegotiation, delay, or shift to self-build hits a name hard
4 · The market has already wobbled
- Two 2026 selloffs: efficient Chinese open-source models (“less compute” fears) and a WSJ off-balance-sheet analysis that knocked Nebius 14% in a morning
- Watch hyperscaler capex guidance and any Meta move into compute rental
A further wrinkle: the campus/power names are landlords and developers with lease-conversion risk. A Compass Point note argued APLD, WULF, and CIFR were trading below the value of their signed contracts — which can read as either opportunity or skepticism about delivery.
06The numbers
Market-size estimates for this category do not agree. That disagreement is itself informative — the category is young and definitions vary.
| Source | Figure | Horizon |
|---|---|---|
| Synergy Research | >$25B FY2025; ~$400B | by 2031 |
| Signisys | ~$20B (2026) → ~$180B | by 2030 |
| TBRC / GII | $26.87B (2025) → $42.17B (2026) | 56.9% CAGR |
| GPUaaS (broader market) | $49.84B | by 2032, 36% CAGR |
| Name | Contracted value | Power / scale |
|---|---|---|
| CRWV | $99.4B backlog (Q1 2026) | >1 GW active; >3.5 GW contracted; targeting 1.7 GW by end-2026 |
| NBIS | $17.4B Microsoft + $3B Meta | Up to 1.2 GW (PA); >5 GW Nvidia systems; 800 MW–1 GW connected |
| IREN | >$4B ARR; $2.8B new services | Microsoft-anchored |
| APLD | $36B contracted lease value | HPC campuses |
| HUT | $9.8B + $7B leases (options to ~$25.1B) | 704 MW contracted; 8.5 GW pipeline |
| WULF | $12.8B | HPC hosting |
| CIFR | $9.3B pipeline | 1 GW West Texas; 2.5 GW gas-backed |
07What could change the story
Bull triggers
- Hyperscaler capex guidance raised, not cut — the demand ceiling keeps lifting
- Contracted megawatts convert to live megawatts on schedule
- Funding stays cheap: more ABS deals and pre-paid contracts at reasonable rates
- Utilization holds at or above the ~75% cost-coverage line
Bear triggers
- Hyperscaler capex cuts — the fastest way to re-rate the whole list down
- Delivery slips (Nebius’s Vineland site is the live example) push revenue right
- Funding terms deteriorate, or dilution accelerates to plug gaps
- More efficient models reduce compute intensity per unit of AI progress
- A major counterparty renegotiates or shifts to self-build (watch Meta)
08FAQ
What is a neocloud stock?
A publicly traded company that monetizes AI compute demand by renting high-end GPUs and related infrastructure — as cloud capacity, AI-factory campuses, or HPC colocation.
What are the pure-play neocloud stocks?
The cloud layer is CoreWeave (CRWV) and Nebius (NBIS). Applied Digital (APLD) is the closest campus-layer pure-play, alongside a cohort of former miners pivoting into HPC: IREN, Hut 8 (HUT), TeraWulf (WULF), and Cipher Mining (CIFR).
Which neocloud stock is the best?
It depends which layer you want: cloud software margins (CRWV, NBIS), power-site conversion optionality (IREN, APLD, HUT, WULF, CIFR), or the supply side (DRAM). This page is research, not a recommendation — the deep pages cover each name’s bull and bear case.
Are neocloud stocks profitable?
Mostly no. The category is capital-intensive by design: revenue and backlog grow quickly, but capex, debt service, and depreciation run ahead of it. GAAP profitability is the exception here, not the rule.
What’s the biggest risk?
Capital intensity plus circularity: heavy debt and dilution, delivery risk on contracted megawatts, customer concentration, and a financing web where supplier, customer, and investor can be the same few companies.
Is there a neocloud ETF?
Yes — since August 6, 2026. The Roundhill Neocloud ETF (NCLD) is the first pure-play neocloud ETF, holding ~21 positions with roughly 58% of assets in Nebius and CoreWeave. The DRAM memory ETF gives supply-side exposure and broader data-center funds carry partial exposure, but NCLD is the closest thing to the basket. See neocloud ETF.
Sources & method
- Market size: Synergy Research Group (neocloud revenue $9B in Q4 2025, +223% YoY; >$25B FY2025; ~$400B by 2031); Signisys (~$20B 2026 → ~$180B 2030); TBRC/GII Research ($26.87B 2025 → $42.17B 2026); GPUaaS market projection as reported
- Contracts and scale: CoreWeave Q1 2026 results (revenue $2.078B, backlog $99.4B, >1 GW active, >3.5 GW contracted); Nebius/Nvidia partnership and Pennsylvania site announcements (2026); IREN $2.8B cloud services (Jul 2026); Applied Digital $36B contracted lease value; Hut 8 $9.8B and $7B lease announcements; TeraWulf $12.8B; Cipher Mining $9.3B pipeline (CoinDesk, 24/7 Wall St., company releases)
- Risk framing: Nvidia backstop/circularity reporting (Forbes, Capacity, Economic Times Data Centers, io-fund); WSJ off-balance-sheet analysis and its effect on Nebius (as reported by Exoswan); Compass Point note on APLD/WULF/CIFR trading below contract value (CoinDesk, Jul 9, 2026); 1,024-GPU H100 cost-coverage estimate (Spheron, Jul 2026)
- Prices: reported quotes from 24/7 Wall St. and Pluang (Sep 17, 2026) plus the Neocloud Brief snapshot (Sep 18, 2026). Approximate, point-in-time.