What is a neocloud? The investor's guide to neocloud stocks — the public GPU-cloud operators ($CRWV, $NBIS, $IREN, $APLD), the bitcoin miners pivoting to AI, the shared NVIDIA/Vertiv/TSMC supply chain, how to invest, and the risks.
More Macroplane supply-chain guides · Baskets · Investment theses · Macro trends
A neocloud is a specialized AI cloud that buys NVIDIA GPUs, builds them into data centers, and rents that accelerated compute to AI labs and enterprises by the hour. Unlike AWS, Azure, or Google Cloud, a neocloud runs one product — GPUs at scale — and lives or dies on how cheaply it can get chips, power, and capital. The investable names are CoreWeave ($CRWV), Nebius ($NBIS), IREN ($IREN), Applied Digital ($APLD), and a cohort of bitcoin miners pivoting their megawatts to AI.
👉 See the whole group at once: the AI Cloud / Neoclouds basket tracks every public neocloud and HPC operator with live price, market cap and recent moves.
The term came out of the GPU shortage. When demand for NVIDIA ($NVDA) accelerators ran years ahead of supply, a class of operators appeared whose entire business was getting GPUs into racks and renting them out — no productivity suite, no general-purpose compute, no enterprise sales motion built over twenty years. Just compute.
That focus is the difference from a hyperscaler. AWS, Azure and Google Cloud sell thousands of services and treat GPUs as one line item. A neocloud sells GPU-hours and optimizes the whole stack around them: liquid-cooled racks, InfiniBand fabric, and power contracts sized for clusters that draw tens of megawatts. The trade-off is concentration. A hyperscaler can absorb a slow quarter in one product; a neocloud's revenue, debt, and survival all ride on the same GPU demand curve.
Three things separate a real neocloud from a reseller:
The catalogue splits into three buckets. Lumping them together is the most common mistake, because their risk profiles are not the same.
Pure-play neoclouds — built from the start to rent GPU compute:
Miners pivoting to AI — bitcoin miners that already had power, land, and substations, and are redirecting megawatts from hashing to GPU hosting:
The builders and ODMs — not neoclouds themselves, but the hardware layer the operators can't function without:
What a real neocloud footprint looks like on the ground: Nebius's operating and announced data centers across Europe and the US, each one a multi-megawatt site sized for GPU clusters.
The combined index tells the story of the demand wave better than any single name. The equal-weight AI Cloud / Neoclouds basket is up over 500% across two years, led by $NBIS:
This is where a neocloud stock is really a concentrated bet on the chain behind it. Open any operator on Macroplane and the same upstream appears, because they all build from the same parts list. CoreWeave's graph is the clearest example:
Nebius buys from the same shelf — $NVDA, $AMD, $INTC, $TSM, $MU, $SMCI, $ANET, plus $VRT for power — and adds its own twist of building rather than leasing. The chips themselves all trace back to one fab: TSMC ($TSM). That is the real chokepoint behind every neocloud on the list. The full picture is in The AI Supply Chain Map, and the power constraint — these sites are gated by megawatts — is the data-center power story.
There is no clean way to buy "the neocloud index" off the shelf — no pure-play neocloud ETF exists yet, and broad AI funds dilute the exposure with megacaps. The three realistic approaches:
| Approach | What you get | Best for |
|---|---|---|
| A single neocloud stock ($CRWV, $NBIS) | Direct, high-beta exposure to one operator's contracts and execution | Conviction in a specific name and tolerance for single-stock volatility |
| The AI Cloud / Neoclouds basket | The whole operator group, equal-weight, in one view | Playing the theme without picking the winner |
| A broad AI-infrastructure ETF | Neoclouds plus $NVDA, hyperscalers, and the rest of the chain | Lower volatility, much diluted neocloud exposure |
The picks-and-shovels angle also works: instead of the operators, own the suppliers every operator depends on — $NVDA, $VRT, $ANET, $TSM — which spread the demand across the whole cohort rather than betting on one balance sheet.
A neocloud is a specialized AI cloud provider that builds and operates its own NVIDIA GPU data centers and rents that accelerated compute to AI labs and enterprises. It runs one product — GPUs at scale — rather than the broad service catalogue of a general-purpose hyperscaler like AWS or Azure. The term was popularized to describe operators that emerged specifically to serve the GPU shortage.
The public pure-plays are CoreWeave ($CRWV), Nebius ($NBIS), and Applied Digital ($APLD). A cohort of bitcoin miners is pivoting into the same business — IREN ($IREN), Hut 8 ($HUT), TeraWulf ($WULF), Cipher Mining ($CIFR), and Bitdeer ($BTDR). The hardware builders behind them include Celestica ($CLS), Super Micro ($SMCI), and Penguin Solutions ($PENG). All of them sit in the AI Cloud / Neoclouds basket. Private neoclouds like Lambda and Crusoe have no ticker.
A hyperscaler (AWS, Azure, Google Cloud) sells thousands of services and treats GPUs as one of many. A neocloud sells GPU compute and almost nothing else, optimizing its racks, networking, cooling and power deals around accelerated workloads. The upside is focus and speed; the downside is concentration — a neocloud's whole business rides on GPU demand.
Yes to both. CoreWeave ($CRWV) is the largest US pure-play neocloud, running entirely NVIDIA GPUs. Nebius ($NBIS) is the European-rooted, vertically integrated neocloud that builds its own data centers rather than leasing them. They are the two defining names in the category.
Not a pure one as of 2026. Broad AI-infrastructure and data-center ETFs hold some neocloud names alongside $NVDA and the hyperscalers, but they dilute the exposure heavily. To track the operators directly, the AI Cloud / Neoclouds basket is the closest equal-weight view.
That depends on your view of GPU-cloud economics and your tolerance for volatility. The bull case is contracted backlogs (anchor deals with Meta and Microsoft reported in the tens of billions) funding fast data-center ramps. The bear case is customer concentration, debt-funded capex, GPU allocation risk, and chip depreciation. This is not financial advice — review the latest filings and size for the swings.
A neocloud is a specialized AI cloud provider that builds and operates its own NVIDIA GPU data centers and rents that accelerated compute to AI labs and enterprises. It runs one product — GPUs at scale — rather than the broad service catalogue of a general-purpose hyperscaler like AWS or Azure. The term was popularized to describe operators that emerged specifically to serve the GPU shortage.
The public pure-plays are CoreWeave ($CRWV), Nebius ($NBIS), and Applied Digital ($APLD). A cohort of bitcoin miners is pivoting into the same business — IREN ($IREN), Hut 8 ($HUT), TeraWulf ($WULF), Cipher Mining ($CIFR), and Bitdeer ($BTDR). The hardware builders behind them include Celestica ($CLS), Super Micro ($SMCI), and Penguin Solutions ($PENG). All of them sit in the AI Cloud / Neoclouds basket. Private neoclouds like Lambda and Crusoe have no ticker.
A hyperscaler (AWS, Azure, Google Cloud) sells thousands of services and treats GPUs as one of many. A neocloud sells GPU compute and almost nothing else, optimizing its racks, networking, cooling and power deals around accelerated workloads. The upside is focus and speed; the downside is concentration — a neocloud's whole business rides on GPU demand.
Yes to both. CoreWeave ($CRWV) is the largest US pure-play neocloud, running entirely NVIDIA GPUs. Nebius ($NBIS) is the European-rooted, vertically integrated neocloud that builds its own data centers rather than leasing them. They are the two defining names in the category.
Not a pure one as of 2026. Broad AI-infrastructure and data-center ETFs hold some neocloud names alongside $NVDA and the hyperscalers, but they dilute the exposure heavily. To track the operators directly, the AI Cloud / Neoclouds basket is the closest equal-weight view.
That depends on your view of GPU-cloud economics and your tolerance for volatility. The bull case is contracted backlogs (anchor deals with Meta and Microsoft reported in the tens of billions) funding fast data-center ramps. The bear case is customer concentration, debt-funded capex, GPU allocation risk, and chip depreciation. This is not financial advice — review the latest filings and size for the swings.