Skip to content
Let's talk

Futurum Research

Research Report

Semiconductors2 min read

The Off Ramp From Per-Token Pricing

Research Director, Semiconductors, Supply Chain & Emerging Tech

The short answer

In our latest report, The Off Ramp From Per-Token Pricing, completed in partnership with QumulusAI, Futurum Research examines how organizations progress from token-metered experimentation to reserved, hybrid AI infrastructure as their AI applications mature, and how that progression helps them…

Futurum's Brendan Burke,

Cite this

Brendan Burke, The Futurum Group, "The Off Ramp From Per-Token Pricing," September 16, 2026. https://trial.futurumgroup.com/research-reports/the-off-ramp-from-per-token-pricing/

The Off Ramp From Per-Token Pricing

Agentic AI is multiplying token consumption per task by up to 100x, and that growth lands directly on the bill for organizations still running production inference on per-token serverless APIs. Per-token pricing is often the fastest path to experimentation and early production, but sustained, high-volume inference turns that same pricing model into a cost control problem that compounds exponentially rather than linearly as usage scales.

AI workload deployment has already gone hybrid: 41% of workloads run in public cloud, 36% in organizations’ own data centers, 13% in colocation, and 6% on bare metal or HPC providers. Reserved and owned compute together account for 66% of AI compute consumption, while on-demand sits at just 19%. Infrastructure tier selection, not workload design, is the primary lever organizations control to keep AI costs predictable as agentic workloads move toward production.

In our latest thought leadership report, The Off Ramp From Per-Token Pricing: How Enterprises Regain AI Cost Control With Reserved Bare Metal, completed in partnership with QumulusAI, Futurum Research examines how organizations progress from token-metered experimentation to reserved, hybrid AI infrastructure as their AI applications mature, and how that progression helps them balance cost, model control, privacy, and resource flexibility.

In this report, you will learn:

  • How AI workload deployment has already gone hybrid, and why reserved and owned compute account for 66% of AI compute consumption
  • The four-stage progression organizations follow, from serverless inference APIs to hybrid multi-tier infrastructure, as workloads move from experimentation to production
  • Why AI-first cloud, the tier spanning bare metal GPU providers, specialized AI clouds, and GPU marketplaces, is forecast to grow faster than any other deployment tier
  • A decision framework for determining which workloads belong on reserved bare metal versus hyperscaler infrastructure
  • How compute leaders at Qubrid AI, Runpod, and Amberd.ai describe their customers’ migration from token-metered APIs to dedicated, reserved infrastructure

If you are interested in learning more, be sure to download your copy of The Off Ramp From Per-Token Pricing: How Enterprises Regain AI Cost Control With Reserved Bare Metal today.

Published by Futurum.

About the author

Brendan Burke, Research Director

Brendan Burke

Research Director, Semiconductors, Supply Chain & Emerging Tech

Brendan is Research Director, Semiconductors, Supply Chain, and Emerging Tech. He advises clients on strategic initiatives and leads the Futurum Semiconductors Practice. He is an experienced tech industry analyst who has guided tech leaders in identifying market opportunities spanning edge processors, generative AI applications, and hyperscale data centers. Before joining Futurum, Brendan consulted with global AI leaders and served as a Senior Analyst in Emerging Technology Research at PitchBook. At PitchBook, he developed market intelligence tools for AI, highlighted by one of the industry’s most comprehensive AI semiconductor market landscapes encompassing both public and private companies. He has advised Fortune 100 tech giants, growth-stage innovators, global investors, and leading market research firms. Before PitchBook, he led research teams in tech investment banking and market research. Brendan is based in Seattle, Washington. He has a Bachelor of Arts Degree from Amherst College.

Coverage on QumulusAI
  • Analyst Insights5
  • Research Reports1
All QumulusAI coverage
See the full picture in the platform

Vendor scoring, the benchmark data behind these findings, and the analyst team that produced them.

Open the platform (opens in a new tab)Become a client
The Weekly

Subscribe to The Weekly

The week's market signals, one email, no filler. Unsubscribe any time.