On-Demand Compute

Snapshot

What is this Market Category?

Why Now?

• AI workload intensity crossed a capital and power threshold. AI infrastructure requires unusually high rack-level power density—JLL cites workloads exceeding 100 kW per rack—making accelerator deployment a specialized data-center problem rather than a routine server-refresh project. [11]
• Accelerator scarcity created a serviceable market for specialized capacity brokers. Customers that cannot obtain or deploy sufficient GPUs can rent them from providers that aggregate hardware, data-center capacity, networking, and cluster operations. ABI Research identifies approximately $42 billion of neocloud GPUaaS revenue in 2025, indicating that the market has moved beyond experimental demand. [5]
• AI infrastructure spending accelerated sharply. IDC reported that organizations’ spending on compute and storage hardware infrastructure for AI deployments reached $82 billion in the second quarter of 2025, a 166% year-over-year increase. [8]
• Capital markets learned to finance compute infrastructure as an asset class. The category is increasingly funded through a blend of venture equity, asset-backed debt, credit facilities, and public-market capital. Crusoe secured a $750 million Brookfield credit facility, while CoreWeave accessed public equity through a $1.5 billion IPO. [1]
• Hyperscaler demand validated specialized providers while leaving room for challengers. Microsoft’s reported $19 billion arrangement with Nebius strengthened investor confidence in specialized AI infrastructure, while Nvidia investments in CoreWeave, Lambda, Nebius, and Nscale signaled that the chip leader sees strategic value in expanding the provider ecosystem around its accelerators. [2][5]

What’s Happening?

CAGR and TAM

• ABI Research, “Profiling Seven Leading Neocloud Companies,” 2025 estimates that GPU-as-a-service revenue from neocloud providers will rise from $42 billion in 2025 to more than $250 billion by 2030 . This implies roughly a 42% compound annual growth rate , calculated from the reported endpoints, and uses a provider-focused GPUaaS framing rather than the entire cloud infrastructure market. [5]
• JLL, “The Rise of Neocloud in the AI Landscape,” 2025 describes the neocloud segment as experiencing an 82% five-year revenue CAGR and identifies approximately 190 operators . JLL’s figure appears broader and more growth-intensive than ABI’s GPUaaS forecast, likely reflecting a wider segment definition and a smaller or differently constructed base. [11]
• IDC, “Worldwide Artificial Intelligence Infrastructure Forecast,” 2025 covers AI servers and AI storage for 2024 and forecasts the market through 2029, while IDC separately forecasts the compute segment of the semiconductor market to grow 36% in 2025 to $349 billion , with a 12% five-year CAGR through 2030 . These are upstream infrastructure and semiconductor measures, not directly comparable to neocloud GPUaaS revenue. [4][13]
• NextMSC, “GPU As a Service Market,” 2026 estimates the global GPU-as-a-service market at $8.60 billion in 2025 , growing to $106.04 billion by 2035 at a 28.5% CAGR , and reports that on-demand usage represented approximately 36% of the 2025 market . This lower base illustrates how sharply estimates vary depending on whether a report counts only a narrower GPUaaS segment or includes specialized neocloud revenue and associated infrastructure services. [9]

Category creation events

• CoreWeave’s March 2025 IPO gave the specialized GPU-cloud model a public-market reference point and materially increased visibility for the neocloud category. Forbes later described the company as having a market capitalization exceeding $60 billion, while JLL identified the IPO as a major momentum event for GPU clouds. [2][11]
• Nebius’s financing and Microsoft relationship helped reposition specialized AI clouds as strategic suppliers rather than simply alternative compute resellers. Nebius announced $700 million of financing in December 2024, and ABI Research cites a $19 billion Microsoft deal as an investor-confidence signal. [5][7]
• The shift from venture financing to infrastructure debt marked a category maturation event. CoreWeave’s financing included large debt facilities, Crusoe obtained a $750 million Brookfield credit facility, and the broader neocloud group was reported to have raised more than $32 billion in debt financing alongside approximately $10 billion in equity. [1][2]

Capital concentration

• CoreWeave remains the clearest capital concentration point. Financial and industry coverage describes the company as having raised more than $12 billion and later using public markets after its IPO, with debt providers including major asset managers and banks. [3][7]
• Crusoe’s Series D raised $600 million in December 2024 , led by Founders Fund, with participation from Fidelity, Long Journey Ventures, Mubadala, Nvidia, Ribbit Capital, and Valor Equity Partners. [7]
• Lambda’s Series D raised $480 million in February 2025 , led by Andra Capital and SGW, with participation from Nvidia and others; the company subsequently combined equity funding with asset-backed financing to expand its GPU cloud. [1][5][7]
• Nebius raised $700 million in December 2024 from investors including Accel, Nvidia, and Orbis Investments, and later secured additional financing cited by JLL as reaching $1 billion in 2025. [7][11]
• Debt is becoming as strategically important as equity. A 2025 industry mapping report cited Crusoe’s $750 million Brookfield credit facility, CoreWeave’s $1.5 billion IPO, and other hardware-backed financing as evidence that the category is being financed more like infrastructure than conventional software. [1]

Market Incumbents

• Amazon Web Services — Offers GPU and AI accelerator instances through a hyperscale cloud footprint and competes on breadth, reliability, and enterprise procurement access. [4][15]
• Microsoft Azure — Provides AI-optimized infrastructure and large-scale accelerator capacity, including strategic arrangements with specialized providers such as Nebius. [5]
• Google Cloud — Supplies GPU and proprietary TPU capacity for AI training and inference through a global cloud platform. [4][15]
• Oracle Cloud Infrastructure — Competes in AI infrastructure with large accelerator deployments and a willingness to finance expansion through substantial debt commitments. [2]
• NVIDIA — Supplies the dominant accelerator platform and invests in multiple GPU-cloud providers, giving it influence across both hardware and on-demand capacity supply. [2][5]
• IBM Cloud — Provides enterprise cloud infrastructure and AI compute for customers that prioritize existing procurement, security, and hybrid-cloud relationships. [4]
• CoreWeave — Publicly listed specialized GPU cloud whose IPO made it the category’s most visible pure-play incumbent. [2][11]

CoreWeave

Amazon Web Services

NVIDIA

Market Challengers

• Nebius — Publicly listed AI infrastructure company building a specialized GPU cloud and securing strategic financing and customer relationships. [5][7][11]
• Lambda — GPU cloud and AI infrastructure provider that raised a $480 million Series D and is expanding through equity and asset-backed financing. [1][5][7]
• Crusoe — AI infrastructure scale-up combining GPU cloud capacity with purpose-built AI factories and large project financing. [1][7]
• Nscale — Specialized AI cloud provider identified among Nvidia-backed neocloud companies. [2]
• Vultr — Independent cloud provider competing for developer and enterprise accelerator workloads through distributed infrastructure and usage-based access.
• Together AI — AI infrastructure platform combining GPU access with model training, inference, and developer APIs.
• RunPod — Developer-oriented GPU cloud focused on flexible access to rented accelerators and distributed supply.

Lambda

Crusoe

Nebius

Market Innovators

• Modal — Developer-focused serverless compute platform that abstracts accelerator provisioning for AI workloads.
• Baseten — Managed inference infrastructure company focused on deploying and serving models with rented accelerator capacity.
• Replicate — Model-serving platform that makes AI inference available through APIs backed by on-demand compute.
• Anyscale — Distributed AI compute and orchestration platform built around Ray for training and inference workloads.
• Lepton AI — AI-native cloud platform focused on simplifying model deployment and inference on accelerator infrastructure.
• Gensyn — Decentralized compute network pursuing a different supply model for machine-learning workloads.
• Shadeform — GPU-cloud aggregation layer that helps customers discover and access capacity across providers.

Industry Coverage and Market Data

Market Reports

• Profiling Seven Leading Neocloud Companies, 2025 — ABI Research — Estimates neocloud GPUaaS revenue at $42 billion in 2025 and more than $250 billion by 2030, while profiling providers including Lambda and Nebius. [5]
• Worldwide Artificial Intelligence Infrastructure Forecast, 2025 — IDC — Forecasts AI server and AI storage infrastructure for 2025–2029 using an infrastructure-market framing rather than a narrow GPU-rental definition. [4]
• Artificial Intelligence Infrastructure Spending to Reach $758Bn USD, 2025 — IDC — Reports $82 billion of AI compute and storage infrastructure spending in Q2 2025, up 166% year over year, and frames the market around organizational infrastructure investment. [8]
• Investment in and Adoption of AI Infrastructure Drives Increase in Semiconductor Market, 2025 — IDC — Forecasts the compute segment of semiconductors to reach $349 billion in 2025, with a 12% five-year CAGR through 2030. [13]
• GPU As a Service Market Size to Reach USD 106.04 Billion by 2035, 2026 — NextMSC — Uses a narrower GPUaaS market definition, estimating $8.60 billion in 2025 and a 28.5% CAGR through 2035; it identifies on-demand usage as the largest 2025 contract segment. [9]
• The Rise of Neocloud in the AI Landscape, 2025 — JLL — Describes approximately 190 neocloud operators and an 82% five-year revenue CAGR, emphasizing the data-center and power-density dimension of the category. [11]

Industry Articles

• Inside The Neocloud Economy: What’s Next For GPU-As-A-Service? — Forbes / R. Scott Raynovich — Examines the public-market impact of CoreWeave’s IPO, Nvidia’s investments, and the shift from equity to debt financing. [2]
• The Rise of Neocloud in the AI Landscape — JLL — Frames neocloud as a specialized data-center and infrastructure segment with unusually high rack power density and roughly 190 operators. [11]
• Profiling Seven Leading Neocloud Companies — ABI Research — Profiles specialized AI-cloud providers and quantifies the GPUaaS opportunity through 2030. [5]
• Mapping the Neocloud Landscape, 2025 — Industry landscape report — Maps equity, debt, and infrastructure financing across CoreWeave, Lambda, Crusoe, and other providers. [1]
• AI Cloud: Which Startup Is Ahead? — New Market Pitch — Compares AI-cloud providers using revenue, backlog, interest expense, and losses, including CoreWeave’s reported 2025 financial profile. [6]

Financial News Sources

• Lambda Scores $480 Million to Grow AI Services — Futurium — Reports Lambda’s $480 million Series D, CoreWeave’s secondary financing, Crusoe’s $600 million Series D, and Nebius’s $700 million financing. [7]
• Inside The Neocloud Economy: What’s Next For GPU-As-A-Service? — Forbes — Reports CoreWeave’s IPO-driven market visibility, more than $60 billion market capitalization, and the sector’s move toward debt financing. [2]
• Mapping the Neocloud Landscape, 2025 — Industry financing analysis — Reports CoreWeave’s $1.5 billion IPO, Crusoe’s $750 million Brookfield facility, and Lambda’s $480 million Series D. [1]
• AI Cloud: Which Startup Is Ahead? — New Market Pitch — Reports CoreWeave’s $5.13 billion 2025 revenue, $66.8 billion backlog, $1.23 billion interest expense, and $1.17 billion net loss. [6]
• Investment in and Adoption of AI Infrastructure Drives Increase in Semiconductor Market — IDC — Provides the semiconductor-market context for the compute expansion underlying on-demand infrastructure demand. [13]

Frontier and Open Questions

• Will specialized neoclouds remain independent, or will hyperscalers absorb their economics? Hyperscalers have the enterprise contracts and global operating footprint, while CoreWeave, Lambda, Crusoe, and Nebius are demonstrating that specialized providers can finance and deploy capacity quickly; the resolution will depend on whether accelerator scarcity persists and whether customers value specialization enough to maintain multiple suppliers. [2][5]
• Does on-demand compute become primarily usage-based, or primarily committed and contract-backed? The market is described as on-demand, but large infrastructure investments increasingly depend on reserved capacity, customer contracts, and debt financing; CoreWeave, Crusoe, and Lambda are the clearest operators testing this model. [1][2][5]
• Can debt-financed GPU infrastructure produce durable margins? CoreWeave’s reported $5.13 billion revenue was accompanied by $1.23 billion of interest expense and a $1.17 billion net loss, raising the question of whether utilization and pricing will outrun hardware depreciation and financing costs. [6]
• Will inference expand the category more than training did? Training requires large clusters but is episodic, while inference can create persistent demand across many applications; managed-inference innovators such as Baseten, Modal, Replicate, and Lepton AI are positioned to determine whether compute is sold as raw capacity or as a higher-level serving platform.
• Should the category include AI factories and data-center development? Crusoe and other providers blur the boundary between cloud service and physical infrastructure by financing and operating purpose-built AI factories, while JLL’s treatment emphasizes power density and data-center capacity as central to neocloud economics. [1][11]
• Will decentralized or aggregated supply challenge vertically integrated GPU clouds? Gensyn and Shadeform represent alternative theses: distributed coordination and multi-provider aggregation could lower dependence on any single operator, but the category’s enterprise buyers may continue to prefer guaranteed capacity, consistent networking, and contractual service levels.

Adjacent Concepts and Categories

• GPU as a Service — The narrowest adjacent category, covering rented accelerator capacity by the hour, reservation, or contract.
• AI Infrastructure — The broader infrastructure layer encompassing servers, storage, networking, accelerators, and data-center capacity.
• AI Factories — Purpose-built facilities designed around high-density accelerator deployment and AI workload production.
• AI Inference Platforms — Managed services that turn rented accelerator capacity into production model-serving infrastructure.
• Hyperscale Cloud — The incumbent cloud layer against which specialized GPU clouds compete and frequently interoperate.
• Bare-Metal Cloud — Direct access to physical servers, often important for high-performance training and predictable networking.
• Accelerator Supply Chains — The semiconductor, server, networking, power, and financing dependencies that constrain available compute.
• Cluster Scheduling and Orchestration — The software layer that allocates scarce GPUs, manages distributed workloads, and improves utilization.

Sources