Specialized AI Clouds Belong on the Shortlist for Multi-Year AI Commitments
In our latest thought leadership brief, Specialized AI Clouds Belong on the Shortlist for Multi-Year AI Commitments, completed in partnership with CoreWeave, Futurum Research examines why enterprises signing multi-year AI infrastructure contracts should evaluate specialized AI clouds alongside…
Cite this

Enterprises are signing multi-year AI infrastructure contracts before they have a clear picture of their options. The largest US cloud providers plan to raise capital expenditure by nearly 80% in 2026, power is now as scarce as capital, and whichever provider secures an enterprise’s GPU capacity today is likely to host its AI workloads for years. Yet most enterprises have barely evaluated the specialized AI clouds competing with hyperscalers for that business. In Futurum’s September 2026 AI Product Series survey, 8% to 19% of respondents did not know enough about a given specialized AI cloud to say whether they use it, compared with 1% to 2% for the largest hyperscalers.
Hourly GPU rates also understate the real differences between providers. A Signal65 total cost of ownership (TCO) and tokenomics study of a 5,000-GPU NVIDIA Blackwell deployment over three years found that CoreWeave cost less than three anonymized hyperscalers in six of seven configurations, and the gap widened once inference throughput was counted. On NVIDIA GB200 NVL72, a 22% three-year cost advantage became a 65% advantage in cost per million tokens.
In our latest thought leadership brief, Specialized AI Clouds Belong on the Shortlist for Multi-Year AI Commitments, completed in partnership with CoreWeave, Futurum Research examines why enterprises should price multi-year AI commitments on cost per token rather than hourly GPU rates, and why specialized AI clouds now belong on every shortlist for GPU-intensive work, alongside the hyperscalers rather than in their place.
In this brief, you will learn:
- Why scarce capacity and undecided buyers raise the stakes of today’s AI infrastructure contracts
- How cost per token reveals differences between providers that hourly GPU rates hide
- Where CoreWeave fits best: large-scale training and high-volume inference, where GPU costs dominate the bill
- How to run an evaluation that holds every provider to the same workload model and cost-per-token test
If you are interested in learning more, be sure to download your copy of Specialized AI Clouds Belong on the Shortlist for Multi-Year AI Commitments today.
Subscribe to The Weekly
The week's market signals, one email, no filler. Unsubscribe any time.
More from Nick Patience
Claude Overtakes GPT in AI Model Adoption as Market Heads to $5.3 Trillion
Nick Patience, VP & Practice Lead, AI Platforms at Futurum, shares new data showing Claude overtaking GPT in AI model adoption, 68.7% to 62.4%, as the AI market heads to $5,301.9B by CY2030.
Google Returns to the Frontier With Gemini 4 Argon
Futurum’s Nick Patience and Fernando Montenegro share their insights on Gemini 4 Argon, Google’s new frontier model, and what its defender-first release means for enterprises and security vendors.
OpenAI Moves Up the Stack and Competes With the Platforms It Powers
Futurum Research’s Mitch Ashley, Nick Patience, and Vikram Rothnam examine how OpenAI used DevDay 2026 to move up the stack, claiming the work surface, the agent runtime, and the pricing that meters both, and what it leaves the CIO to govern.