Verified answer
Neocloud vs hyperscaler: what's the difference?
A neocloud is a GPU-first cloud whose product is rented AI compute, independent of AWS, Azure, Google Cloud and Oracle. A hyperscaler is a general-purpose cloud at global scale where AI compute is one product of many. DEPLOY tracks 32 verified neoclouds against that border; here is the full contrast.
Dimension by dimension
| Neocloud | Hyperscaler | |
|---|---|---|
| What it sells | AI compute itself: GPU / accelerator time, rented to third parties. | A broad platform (compute, storage, database, networking); AI is one product of many. |
| GPU-first? | Yes. The fleet and the data-center design exist for accelerated AI workloads. | No. General-purpose fleets; GPU capacity is a segment inside a much larger estate. |
| Independence | Independent of AWS, Azure, Google Cloud and Oracle. | Is one of them, or operates at that scale and scope. |
| Capacity model | Dedicated AI data-center capacity, often single-tenant clusters with high-bandwidth interconnect. | Massive multi-tenant regions serving every workload class. |
| Typical customer | AI labs, model builders and enterprises that need large, cost-efficient GPU clusters. | Every kind of IT buyer, from a hobbyist to a bank. |
| Examples | CoreWeave, Lambda, Crusoe, Nebius, and the fast-growing Asian operators. | AWS, Microsoft Azure, Google Cloud, Oracle Cloud. |
The test that draws the border
A neocloud IS
- GPU / AI-accelerator-first: compute is the product, not incidental to another business.
- Rents that compute to third parties (not solely for its own workloads).
- Independent of the hyperscalers (not AWS / Azure / Google Cloud / Oracle).
- Operates or contracts dedicated AI data-center capacity (a real physical footprint).
A neocloud is NOT
- Hyperscalers (general cloud at scale; AI is one product of many).
- Pure colocation / REITs (sell space + power, not compute).
- Enterprise self-builders that do not rent (xAI, Meta, Tesla for own use).
- AI labs that do not resell compute, and brokers/aggregators with no capacity of their own.
The full operator-class contrast set (neocloud, hyperscaler, colocation, enterprise self-build, sovereign) and the global roster are on the neocloud taxonomy page.
Common questions
Is CoreWeave a neocloud or a hyperscaler?
A neocloud. CoreWeave is the reference example: a GPU-first cloud whose product is rented AI compute, independent of the hyperscalers, running dedicated AI data-center capacity. Scale alone does not make a hyperscaler; selling a general-purpose platform does. See the full neocloud test.
Is Azure or AWS a neocloud?
No. Microsoft Azure, AWS, Google Cloud and Oracle Cloud are hyperscalers: general-purpose clouds at global scale where GPU/AI capacity is one product among many. A neocloud is defined partly by being independent of exactly these four.
Can a company be both a neocloud and a hyperscaler?
Not under this taxonomy. The neocloud definition requires independence from the hyperscalers, so the two classes are mutually exclusive at the top. A hyperscaler can offer neocloud-like GPU products, but it remains a hyperscaler because AI compute is one product of many on a general-purpose platform.
Why does the distinction matter?
Because it changes who you are buying from and what you are exposed to: a neocloud is a focused, independent GPU supplier, while a hyperscaler is a diversified platform. For mapping the AI-compute supply chain, conflating the two hides where dedicated GPU capacity actually sits. The physical campuses are in the data-center registry.