Microsoft, which invested $13 billion in OpenAI, had its accounts blown up by its own engineers using Claude Code.

The sky is falling for Microsoft engineers!

Yesterday, Microsoft engineers were still using Anthropic's Claude Code to write code at full speed, modify code libraries, and run proxies. Today, they received a notice: all usage must be discontinued by the end of June, and they are forced to switch to their own GitHub Copilot CLI!

This is not a minor issue. Multiple departments at Microsoft have collectively cut off the supply of Claude Code.

Even more shocking data has emerged: Even a giant like Microsoft, with almost unlimited cloud resources, finds the bill for Claude Code too high.

At the same time, the internal memo from Uber CTO Praveen Neppalli Naga has directly shattered the ceiling:

The entire AI budget for 2026 has been completely exhausted in just four months!

This is not AI helping; this is AI burning money!

The era of AI subsidies has officially ended, and the real "computing bill" is starting to crush corporate budgets.

When giants like Microsoft and Uber, which are not short of money, start to say they "can't afford it", it means that AI is changing from an efficiency tool to a money - swallowing black hole.

This is no longer just a technical experiment; it's a survival crisis related to the cash flow of all enterprises.

Microsoft Can't Bear the Token Fees

Foreign media reports show that Microsoft plans to cancel all employees' subscriptions to Claude Code by the end of June.

The specific scope has been defined: the "Experience + Devices" department responsible for Windows, Microsoft 365, Outlook, Teams, and Surface.

They must stop using Claude Code by the end of June and migrate their workflows to their own GitHub Copilot CLI in the next few weeks.

Nearly 100,000 Microsoft engineers will be required to make the switch.

The reason is extremely straightforward: the bill is too expensive.

This supply cut - off is defined internally as an "emergency brake" rather than an orderly tool migration.

Last year, Nadella revealed that "30% of the code was generated by AI", but the token - based billing framework shows its cruel side.

In traditional software development, the more code engineers write, the more money the company earns; in the era of AI agents, the more code AI writes, the faster the bill paid to external suppliers increases.

The generation of each token is consuming Microsoft's cash flow in real - time.

This is like hiring a security guard with amazing efficiency but charging precisely for every minute of his work - when he works more diligently than the regular army, the employer's financial balance is broken.

According to the restored internal communication in the media, this supply cut - off was led by the finance department: the engineering line unanimously reported a significant increase in productivity, but the management chose to force the discontinuation.

This has gone beyond the debate over product preferences.

The logic of the ledger has finally defeated the technical taste of engineers.

The most ironic thing is that Microsoft has invested $13 billion in OpenAI and personally built most of the computing power base for its competitor Anthropic on Azure.

But when its own engineers make large - scale calls to Claude Code, the financial process is transparent and cold: Microsoft needs to pay Anthropic by token - which is like transfusing blood to its most direct competitor.

After seeing the bill, this giant with a market value of $3.5 trillion has decided not to be the "sucker" anymore.

For Microsoft, GitHub Copilot is an "internal cost accounting", and the resources flow within Azure, with extremely low marginal costs; Claude Code is a real "external bill". Even if the outputs of the two are exactly the same, their financial natures are completely different.

Of course, Microsoft's official statement is "strategic alignment".

The statement is decent, but the bill can't be hidden.

This is the first large - scale financial correction in the AI tool market.

Uber Burned Through Its Annual Budget in Four Months

Microsoft's situation is not an isolated case.

Uber is even more extreme: in the era of token - based billing, the customers who embrace technology most actively may be the first to hit a financial crisis.

The internal memo from Uber CTO Praveen Neppalli Naga is shocking:

• 95% of engineers use AI tools monthly.
• 84% have entered the "Agentic Coding" mode.
• 70% of the code submitted online is generated by AI.

But the cost is catastrophic.

The special AI budget originally set aside for the whole year of 2026 was completely exhausted in the first four months.

The prediction model established by the finance team based on "fixed seats" and "low - frequency calls" completely failed when faced with the agent - based workflows of 5,000 engineers working in parallel!

In the agent mode, the monthly cost of heavy users can reach up to $2,000 per person, and there is no stable equivalent relationship between this expenditure and productivity output.

Uber's case sends a systematic signal: The deeper you use, the more money you burn.

When corporate finance teams haven't learned to manage this "utility - based" pricing, the efficiency myth of AI is often accompanied by a budget black hole.

Paradigm Shift: From "Monthly All - You - Can - Eat" to "Pay - by - the - Gram"

In the past thirty years, we have been used to the generosity of the software industry: pay the monthly fee, and the software is as cheap as breathing. This is the essence of SaaS (Software as a Service).

But AI is violently destroying this logic.

Anthropic, OpenAI, and GitHub under Microsoft have all made a turn - GitHub has cancelled the "flat - rate" fixed plan and fully switched to usage - based billing.

In the past six months, the effective price of global AI software has increased by 20% to 37%.

Humanity has officially entered the "Utility Economy" from the "Subscription Economy".

Previously, using AI was like having a buffet. Pay the entrance fee, go in with a big appetite, and eat as much as you can.

Now, AI has become top - grade wagyu beef priced by the gram in the market, with a clear price tag.

Each line of automatically generated code and each invoked agent leaves a string of bloody numbers on the meter.

Intelligence is no longer a static tool but a high - energy "metabolic process" - every breath costs money.

Someone Has to Do the Math

So, where will this race lead? Logically, there are only two paths, and they lead to the same end:

Firstly, enterprises reduce their AI usage to fit the budget - but this will slow down the revenue growth rate of AI labs, and they won't be able to support their high valuations before IPO.

Secondly, AI labs voluntarily reduce prices and bear the losses themselves - but this will further worsen their already fragile unit economics at the worst time.

Whichever path is taken, the end is the same: the accounts don't add up, and someone has to pay for the losses. The current valuation of AI is based on the assumption that "revenue will continue to explode".

Once enterprises stop using AI because it's too expensive, or AI companies sell at a loss to retain customers, the two words hanging over the entire industry are "Writedown".

For every business owner, the baton is changing hands.

In the past, the promotion of AI was driven by CTOs and developers, aiming for "speed" and "coolness"; from now on, the CFO may become the real top commander of the AI team.

Reference materials:

https://www.theverge.com/tech/930447/microsoft-claude-code-discontinued-notepad

https://www.forbes.com/sites/janakirammsv/2026/05/17/uber-burns-its-2026-ai-budget-in-four-months-on-claude-code/

https://github.blog/news-insights/company-news/github-copilot-is-moving-to-usage-based-billing/

https://www.windowscentral.com/microsoft/microsoft-cancels-claude-code-licenses-shifting-developers-to-github-copilot-cli-a-move-likely-driven-by-financial-motives

https://www.thurrott.com/forums/microsoft/microsoft/thread/microsoft-cancels-claude-code-licenses-shifting-developers-to-github-copilot-cli

This article is from the WeChat public account "New Intelligence Yuan", author: ASI Revelation, published by 36Kr with authorization.