Worldwide spending on Artificial Intelligence (AI) will hit $2.7 trillion this year, growing 49.5%. But Gartner buried the real story in its September 16 forecast: buyers now demand that vendors manage costs and build usage tracking directly into their tools to prove the tech actually works. [1]

Cost control is no longer a chore for the back office. It is a strict line item on every purchase order.

Gartner just raised its growth outlook for Generative AI (GenAI) models to 117% in a single quarter. [1] It also bumped AI application building platforms to 39%. [1]

Yet, those same analysts place GenAI firmly in the “Trough of Disillusionment” for 2026. [1] Money pours in faster than ever, while trust in the payoff sinks.

Why it matters: The leaders paying the bill cannot see what they are buying. A user asks a simple question, which wakes up a main planner, checks three search tools, makes four software calls, and prompts the model seven times. The invoice arrives as one massive number, hiding the real cost drivers deep inside the system.

By the numbers:

  • $2.7 trillion: Gartner’s 2026 worldwide AI spend forecast, rising from $1.79 trillion in 2025 toward $3.64 trillion in 2027. [1]
  • 117%: The revised 2026 growth rate for GenAI models, up from 110% in the prior quarter. [1]
  • $65.5 billion: The projected size of the AI agents market in 2027, more than double its $29.2 billion level in 2026. [1]
  • 98%: The share of Financial Operations (FinOps) teams managing AI spend today, up from 31% two years ago. [3]

Watching token prices will not save your budget. Token prices have dropped for two years, but total spending keeps climbing because the systems simply consume more units to get the job done.

The FinOps Foundation is blunt about the right fix for this problem. It calls the new discipline “use case economics,” shifting the focus away from raw technical metrics. [4] This means you divide the total cost of a real outcome by the number of outcomes produced.

The playbook:

  • Measure cost per resolved work-unit: Do not just track the cost per token. A workflow that fixes a ticket in one $0.05 step beats a system needing five $0.01 steps. The cheaper-per-token model often costs more per real outcome. [4]
  • Demand tracking at purchase time: Gartner notes enterprises already force this rule into deals. [1] Make cost visibility and per-model tracking a strict buying gate, not a later add-on.
  • Budget the hidden layers: Human review, software calls, and database lookups stack on top of raw token costs. An automation requiring a senior reviewer to check every answer carries a negative Return on Investment (ROI) even if the tokens are free.
  • Track spending before you scale: Cloud providers differ sharply in how they report data. Some show costs for each builder, while others hide it inside a broad platform. Pick vendors and gateways that let you tag every dollar to a specific team, feature, or business unit from day one. [5]

The catch: None of this is easy to measure. Teams keep putting it off because the math is hard and the tools are new.

Most enterprises still cannot track their AI spend past a single giant group bill. Gartner warns that buyers remain undeterred by the clear risk of runaway costs today. [1]

The Trough of Disillusionment is a warning, not a firm prophecy. The teams that survive it will tie every dollar to a resolved unit of work before the next budget cycle begins.

  • The Big Shift: Gartner projects worldwide AI spending will hit $2.7 trillion in 2026, while enterprises demand that vendors build usage tracking directly into their tools. [1]
  • Why It Matters: Token prices keep falling, but total bills climb while GenAI enters the Trough of Disillusionment. The gap between spending money and proving value is now a Chief Executive Officer (CEO) issue, landing squarely on the desks of FinOps teams. [1][3]
  • The Winning Moves: Run AI finance on unit economics, not token counts.
  • Cost per resolved work-unit: Divide total AI cost by completed outcomes, proving that a single $0.05 task often beats a five-step $0.01 task. [4]
  • Vendor usage tracking: Make detailed cost tracking and spend visibility a strict condition of every purchase. [1]
  • Budget the hidden layers: Count human review, system calls, and database lookups on top of token costs to find true profitability. [5]
  • The Catch: Tracking costs perfectly is difficult, and buyers often ignore the risk of runaway spending. The first team to track per-unit value will win the next budget cycle. [1]

Related reading

Sources

[1] https://www.gartner.com/en/newsroom/press-releases/2026-09-16-gartner-forecasts-worldwide-ai-spending-to-grow-49-point-5-percent-in-2026 [2] https://www.gartner.com/en/newsroom/press-releases/2026-05-19-gartner-forecasts-worldwide-ai-spending-to-grow-47-percent-in-2026 [3] https://data.finops.org [4] https://www.finops.org/wg/finops-for-ai-tools-services-considerations [5] https://developers.cloudflare.com/ai-gateway/changelog