AI demand side is staggering

Goldman Sachs forecasts AI token consumption will multiply 24x by 2030, to 120 quadrillion tokens per month.

The mechanics explain why: an agent doesn’t answer one query, it runs a sequence of them — blowing up a simple chatbot request 10x, 20x, 50x in compute.

And even though we’re now talking quadrillions, the assumptions underneath are conservative. Goldman models just 12% of knowledge workers using agentic AI by 2030, rising to 37% by 2040.

Bill is already arriving

Per a new KPMG survey (via WSJ), only 26% of companies have a comprehensive view of their AI costs.

• 50% have partial visibility

• 22% have none - they only see it after billing

The challenge follows: Uber burned through its entire 2026 AI coding budget by April.

One KPMG client watched token usage grow 6X. Jellyfish says per-developer token consumption rose ~19X in nine months.

And one company reportedly ran up a $500M Claude bill after skipping usage limits.

KPMG’s Steve Chase in WSJ on AI token cost:

“It’s a new resource that needs to be managed that didn’t exist quite that way.”

AI is no longer just a CTO/platform conversation about model capability.

It is becoming a CFO, procurement and FinOps conversation about variable spend, auditability, routing, budgets, commits and governance.

Response is forming:

The Linux Foundation is standing up the Tokenomics Foundation to bring token spend into the same kind of discipline FinOps brought to cloud. Model routers, AI gateways and observability vendors are racing into the gap.

Meanwhile, spend is consolidating where governance already exists

Increasingly, that means cloud marketplaces: one bill, drawdown against existing cloud commits, procurement controls built in.

Marketplace teams tell us Anthropic is becoming one of their top revenue performers - and recent Ramp data shows it just passed OpenAI in business adoption (34.4% vs. 32.3%) for the first time in April.

Marketplaces won’t be the only control plane for AI spend. But they are the ones that already operate at scale.

Lessons for alliance and tech leaders:

• Your buyer now includes the CFO. AI deals (and others) will run through finance, so cost visibility and commit drawdown belong in your pitch — not just capability.

• ISVs that make spend predictable and controllable — private offers against commits, usage transparency, sensible defaults — will win deals and renewals that capability alone won’t.

Is AI cost control becoming the next big marketplace use case — or will FinOps tooling get there first?