AI’s trillion dollar token reckoning | Tech Radar
Overview
News, deals, reviews, guides and more on the newest computing gadgets
Start exploring exclusive deals, expert advice and more
Details
Unlock and manage exclusive Techradar member rewards.
Unlock instant access to exclusive member features.
Get full access to premium articles, exclusive features and a growing list of member rewards.
AI cost economics now matter more than model access
When you purchase through links on our site, we may earn an affiliate commission. Here’s how it works.
Enterprise AI strategy spent two years chasing a single objective: reach the frontier before competitors do.
The default path was a public cloud account, an API key from Open AI or Anthropic, and a willingness to absorb cost in exchange for speed.
The numbers tell the story. Gartner forecasts worldwide AI spending reaching
In fact, in mid-2025, they claimed that procurement of AI had entered a “Trough of Disillusionment,” where scaling depends on predictable ROI, rather than visionary pilots.
Tokenomics: AI Has an income statement - it’s time to read it
Beyond Tokenmaxxing: the rising token tax on enterprise AI
The pressure has now shifted from how fast enterprises can pilot AI to whether they can sustain, govern, and defend it in production.
The Race to the Front is Over – Now Comes the Bill
We are now past AI 1.0, where simple access to cutting-edge AI was the differentiator. Now it’s AI 2.0’s turn, where inference economics, data gravity, latency and control decide the outcomes. Token prices have fallen almost tenfold annually since 2021, but AI spend overall by organizations has increased. That’s because more capable models have enabled greater ambition.
Anthropic, Open AI, and Mistral are now stratifying offerings between flagship reasoners and lower-cost workhorses precisely because customers refuse to pay flagship prices for every task. Mc Kinsey’s 2025 State of AI survey confirms the pattern - adoption is increasing, but impact at scale remains elusive for most organizations.
Now CIOs have stopped asking which model, but where each workload needs to run and how much it’s going to cost.
Banks delivering the next best action are a good example: the in-app, in-branch, or call-center recommendation served in milliseconds against a customer’s live context. The best banks prove that personalization at this layer can lift revenue by 5-15%. To give a firsthand example, a global bank we work with launched an AI assistant that has already resolved more than 1.5 million customer inquiries in its first year, driving huge efficiencies.
Beware the token trap: Why saving on inference might put your ADLC at risk
Why enterprise AI is forcing a rethink in cost control
Why AI infrastructure costs keep surprising IT leaders
But the inference economics are unforgiving at this scale. A single agentic decision can chain five to twenty model calls, each carrying its own context window. The cost gap between £0.50 and £3 per million input tokens seems trivial in a single-turn demo. Spread across hundreds of millions of customer events, it becomes the difference between a money-making feature and a money-burning one.
This isn’t a hypothetical either. Uber’s 5,000-strong engineering team’s use of Claude Code burned through the company’s entire annual AI budget in the first four months of this year. And AI companies are responding to this market shift. Decagon, after re-architecting onto an open-source multi-model stack on NVIDIA Blackwell, dropped cost per voice query by sixfold. Next best action isn’t a marketing decision anymore; it’s an economic decision.
Organizations making the structural shift now will outcompete those treating model selection as an afterthought.
The hardest lesson of the past 18 months is that model commoditization does not reduce enterprise complexity but relocates it. Open weights from Mistral or Deep Seek cut experimentation cost, but orchestration, governance, evaluation, and integration burdens move up the stack and sit with the buyer.
Enterprise leaders should be measuring unit economics per useful task, operational burden per deployed agent, and the ratio of inference spent on the governance scaffolding around it. That ratio is typically 1:5 or worse.
A second architectural shift is arriving: sub-quadratic attention.
Approaches from Deep Seek, Google, and Cartesia are collapsing the cost of long-context reasoning by orders of magnitude, with recent benchmarks showing 100x to 300x cost reductions at comparable accuracy.
Large banks will now be able to run whole-portfolio risk modelling, multi-decade fraud detection and cross-jurisdiction Know-Your-Customer (KYC) as single-pass operations - no more chunked retrieval workarounds.
Telcos can make network operations, predictive maintenance and multi-year customer journey reasoning more economically viable at scale. And manufacturers can move full-plant simulation and supply-chain disruption forecasting from periodic batch jobs to continuous reasoning.
The architecture that wins will not be the one with the cheapest token. It will be the one that places compute closest to the data, under the right jurisdiction, with governance that holds.
Sustainable, sovereign, controlled - that’s the new triad. The enterprises that build for it now will define the next decade.
We've reviewed, rated, and ranked the best business plan software.
This article was produced as part of Tech Radar Pro Perspectives, our channel to feature the best and brightest minds in the technology industry today.
The views expressed here are those of the author and are not necessarily those of Tech Radar Pro or Future plc. If you are interested in contributing find out more here: https://www.techradar.com/pro/perspectives-how-to-submit
You must confirm your public display name before commenting
Lanterns episode 1 ending explained: is [spoiler] really dead in the HBO Max show?
The US Army is opening up its training centers for private firms to test out new drones
Russian body, Desi brain: How India is turning the Su-30MKI into a lethal stealth hunter
US Air Force wants AI help to keep its Minuteman III nukes ticking until they're 66, or maybe 80
Quote of the day by Elon Musk: "[The Tesla Bot] has the potential to be a generalized substitute for human labor over time"
Tech Radar is part of Future US Inc, an international media group and leading digital publisher. Visit our corporate site.
© Future US, Inc. Full 7th Floor, 130 West 42nd Street, New York, NY 10036.
Key Takeaways
- News, deals, reviews, guides and more on the newest computing gadgets
- Start exploring exclusive deals, expert advice and more
- Unlock and manage exclusive Techradar member rewards
- Unlock instant access to exclusive member features
- Get full access to premium articles, exclusive features and a growing list of member rewards
