GPU-as-a-Service

GPU-as-a-Service (GPUaaS) is a cloud computing model in which a specialized provider, often called a neocloud, rents large GPU clusters for AI training and inference, bundling the data centers, power, and orchestration software around them. CoreWeave, Nebius, and IREN are the leading US-listed providers, and CoreWeave’s revenue backlog reached $99.4 billion in Q1 2026.

What is GPU-as-a-Service?

GPU-as-a-Service inverts the classic cloud bargain. Instead of buying and depreciating its own accelerators, an AI lab or enterprise rents access to clusters of thousands of GPUs under contracts that range from on-demand hours to multi-year, multi-gigawatt commitments. The provider does everything below the model: it secures grid power, builds or leases the data center, buys the GPUs, wires them into clusters with high-speed networking, and operates the scheduling and orchestration software. Because the entire business is one product, these providers are called neoclouds, in contrast to general-purpose hyperscalers. The model’s economics are visible in its leaders’ disclosures: CoreWeave grew Q1 2026 revenue 112% year over year to $2.08 billion and ended the quarter with a $99.4 billion revenue backlog and more than 3.5 gigawatts of contracted power (CoreWeave, May 7, 2026), while Nebius grew 684% to $399 million (Nebius, May 13, 2026). IREN’s co-founder and co-CEO Daniel Roberts states the underlying scarcity directly:

“The world is structurally short compute, and the bottleneck is delivered data center and GPU capacity.”

— Daniel Roberts, co-founder and co-CEO, IREN (IREN, May 7, 2026)

The most striking feature of the 2025-2026 market is that hyperscalers themselves became the biggest GPUaaS customers. Microsoft contracted about $17.4 billion of dedicated capacity from Nebius through 2031 (Nebius 6-K, Sep 8, 2025) and signed a $9.7 billion, five-year GPU cloud contract with IREN (IREN, Nov 3, 2025), renting capacity it could not build fast enough itself.

How is GPU-as-a-Service used in thematic investing?

The term defines the cloud layer of the US AI Data Center concept: the companies whose entire revenue is AI data-center capacity sold as a service. For investors, GPUaaS providers are the most direct, and most leveraged, expression of AI infrastructure demand. Their contracts make the build-out measurable, from CoreWeave’s roughly $22.4 billion of OpenAI agreements to IREN’s $3.1 billion of contracted annualized revenue (CoreWeave, May 7, 2026; IREN, May 7, 2026), but the model is capital-hungry and customer-concentrated, which is why neocloud stocks carry both the fastest growth and the heaviest balance-sheet risk in the theme.

Related terms & concepts

FAQ

What is the difference between a neocloud and a hyperscaler?

A hyperscaler (AWS, Microsoft Azure, Google Cloud) runs a general-purpose cloud with hundreds of services; a neocloud is built around one product, renting GPU clusters for AI, with the data centers and power around them. The two now interlock: Microsoft has contracted about $17.4 billion of capacity from Nebius (Nebius 6-K, Sep 8, 2025) and $9.7 billion from IREN (IREN, Nov 3, 2025), effectively outsourcing part of its AI build-out to neoclouds.

Which companies offer GPU-as-a-Service?

The largest US-listed pure plays are CoreWeave (CRWV), whose Q1 2026 revenue grew 112% to $2.08 billion with a $99.4 billion backlog (CoreWeave, May 7, 2026), Nebius Group (NBIS), which grew 684% to $399 million (Nebius, May 13, 2026), and IREN, which targets $3.7 billion of annualized revenue by the end of 2026 (IREN, May 7, 2026). Hyperscalers also sell GPU capacity, but as one service among many.

Sources & references

  1. CoreWeave Reports Strong First Quarter 2026 Results · CoreWeave, Inc., 2026-05-07
  2. IREN Business Update and Q3 FY26 Results · IREN Limited, 2026-05-07
  3. Nebius reports first quarter 2026 financial results · Nebius Group N.V., 2026-05-13
  4. Nebius announces AI infrastructure agreement with Microsoft (Form 6-K, Exhibit 99.1) · Nebius Group N.V. / SEC EDGAR, 2025-09-08
  5. IREN Secures $9.7bn AI Cloud Contract with Microsoft · IREN Limited, 2025-11-03