The token bill is coming. Here's who's going to pay it.

In April 2026, Uber's CTO told a technology publication something that should have made every CFO in the world sit up: "I'm back to the drawing board because the budget I thought I would need is blown away already." The company had burned through its entire annual AI coding budget before June. Before Q3 planning. Before anyone had a chance to course-correct.

01What tokenmaxxing actually did to companies

For much of late 2025 and early 2026, "tokenmaxxing" was the unofficial productivity philosophy inside some of the world's largest technology firms. The logic was deceptively simple: more AI token usage means more productivity. Track the tokens, find your most innovative employees, win.

Companies installed leaderboards. Engineers raced to consume more. At Amazon, some employees reportedly spun up AI agents to complete wholly meaningless tasks just to keep their usage stats up — because managers had started using token consumption as a performance metric. At Uber, engineers were encouraged to use AI "as much as possible," with usage ranked competitively on internal boards.

The reckoning arrived quickly. Microsoft cancelled Claude Code subscriptions across several key product divisions. Meta took down its internal tokenmaxxing leaderboard. Amazon shut down its own after discovering the gaming behaviour. And Uber capped individual monthly spending at $1,500 per AI coding tool, with an internal dashboard so engineers can track their real-time consumption.

The conclusion reached by Uber's COO Andrew Macdonald sums up the entire era neatly: "If you're not actually able to draw a direct line to how much useful features and functionality you're shipping to your users, the token costs are harder to justify."

"Even as the price per token falls, total spending keeps rising — because usage grows faster than efficiency gains."

02The structural problem nobody wants to say out loud

Here is what the token cost conversation tends to obscure: this is not just an efficiency problem. It is a capital transfer problem.

Every time a company outside the United States adopts a public AI tool — whether Claude, ChatGPT, or Gemini — several things happen simultaneously. Sensitive operational data flows to servers outside the country. Every query and document potentially trains a model that benefits a foreign provider's future product. And the productivity gains from AI are partially offset by token bills that accumulate offshore, flowing from local economies to a small number of hyperscalers.

This is not a hypothetical. The tokenmaxxing era saw companies encourage maximum AI spend, frame usage as productivity, and reward the heaviest consumers. The bill came due in 2026, and it landed on enterprises who had already cut the people those tools were supposed to replace.

03The private token factory playbook

The companies that anticipated this are not capping usage — they are building their own inference infrastructure. The reasoning is exactly what it sounds like: if AI is going to be as foundational as electricity or cloud computing, you do not rent it from one overseas supplier indefinitely. You build the capability in-house, on your own terms, with your own data.

This is precisely what Trusted Hub is doing right now. We have built our own inference cluster. We are actively testing every viable open-source model and are pushing our engineering team hard to resolve concurrency issues so we can pivot away from paid model APIs as quickly as possible.

Talk to us about how to run a fully local inference cluster using Trusted Hub's Multiverse.

This is not just a cost play. It is a sovereignty play. The ability to run powerful models locally — at a fraction of the API cost, with none of the data exposure — is the logical endpoint for any enterprise serious about AI that does not want to become permanently dependent on a foreign utility.

04What this means for Singapore specifically

Uber can absorb a runaway AI budget as a rounding error. Most Singapore enterprises cannot. And the stakes here are higher than they appear.

Singapore has long been a sophisticated adopter of enterprise technology. That sophistication now needs to extend to how AI infrastructure is procured. The question is not whether to use AI — that ship has sailed. The question is whether the AI you run is yours: your data, your knowledge, your compute, your costs, on infrastructure you govern.

Trusted Hub's platform is designed around this premise from the ground up. The Trusted DataVerse pipeline keeps your multimodal data in your environment. The Trusted KnowledgeVerse hosts your intelligence securely with full digital rights control, allowing you to gain insights directly from it. Trusted ActionVerse understands your objectives and creates agentic AI workflows to automate your processes. As we complete our inference cluster build, the ability to run capable open-source models locally — purpose-built on your own knowledge — becomes available to every client on the platform.

The token bill is coming for everyone. The only question is whether you will be paying it to someone else indefinitely — or building the infrastructure to own it. Singapore companies have a narrow window to act before AI infrastructure costs become as locked-in as cloud dependencies were in 2015. The time to localise is now.