Microsoft Cancelled Claude Code. Uber Ran Out of AI Budget.

Author: Protik Ganguly

Published May 26, 2026·2 min read

image.jpeg

Uber gave 5,000 engineers access to Claude Code in December 2025. By April, the company had exhausted its entire 2026 AI budget. Uber's CTO told The Information: "I'm back to the drawing board, because the budget I thought I would need is blown away already" (Fortune, 2026). Microsoft wound down most internal Claude Code licenses in mid-May, ending access for its Experiences and Devices division by June 30. Token-based billing made consumption unsustainable at deployment scale — just six months after the pilot launched.

These are not small companies making naive mistakes. These are two of the most sophisticated technology operators in the world. And they both discovered the same thing: AI tools that make engineers more productive also make AI bills unpredictable in ways that traditional procurement logic cannot handle.

The mechanism is specific. Claude Code operates on token consumption — costs vary based on how engineers actually use it. An engineer running basic autocomplete generates minimal spend. One orchestrating parallel agents across a large codebase can run up thousands of dollars in the same period. Monthly per-engineer costs at Uber ranged from $150 to $250 on average, with heavy users reaching $2,000 (DesignRush, 2026). Multiply across 5,000 engineers and the variability destroys any budget forecast built on averages.

Uber compounded the problem with an internal leaderboard that scored engineers by how many tokens they consumed. The more tokens, the higher the score. Engineers had every incentive to use Claude Code aggressively and no incentive to hold back. The company built a system that rewarded consumption without measuring value. The bills followed logically.

Uber's COO Andrew Macdonald was asked whether the spending connected to consumer-facing innovation. His answer: "That link is not there yet. Maybe implicitly there's more that is getting shipped, but it's very hard to draw a line between one of those stats and actually producing 25% more useful consumer features" (Fortune, 2026).

That is the AI productivity paradox expressed by a company living it in real time.

A 2025 Mavvrik survey found 85% of companies miss AI cost forecasts by more than 10%. Goldman Sachs forecasts a 24-fold increase in token consumption by 2030 as agentic AI scales (Fortune, 2026). The bills are not getting smaller. The measurement problem is not going away. Companies deploying AI fastest are discovering that speed of adoption and clarity of return are not the same thing.

The tools work. The economics are still being written.


References

DesignRush. (2026, May). Uber burns 2026 AI budget on Claude Code in four months. https://news.designrush.com/uber-2026-ai-budget-claude-code-token-spend

Fortune. (2026, May 22). Microsoft reports are exposing AI's real cost problem. https://fortune.com/2026/05/22/microsoft-ai-cost-problem-tokens-agents/

Fortune. (2026, May 26). Uber burned through its entire 2026 AI budget in four months. Now its COO is questioning whether it's worth it. https://fortune.com/2026/05/26/uber-coo-ai-spending-tokens-claude-code/

Related Articles

The AI Boom's Profits Sit Furthest From the Customer

The AI Boom's Profits Sit Furthest From the Customer

• Morgan Stanley projects AI cloud spending to reach $980 billion in 2026 and $1.58 trillion in 2027, with Nvidia expected to capture nearly $700 billion of the 2027 S&P 500 AI‑related revenue. • While AI demand is strong—evidenced by Microsoft’s AI business hitting a $37 billion annual run rate—supplier profits are realized early, before customers see returns on their hardware investments. • Historical parallels to vendor‑financed telecom equipment show risks when customer revenues fall short, though current AI buyers are highly cash‑generative; Morgan Stanley warns 2027 could see the largest cash‑flow deficit in its forecast, driven partly by higher component prices.

The Bond Market Is Financing the AI Boom. Does It Know the Full Risk?

The Bond Market Is Financing the AI Boom. Does It Know the Full Risk?

• AI‑related global debt issuance surged to $236 billion through May 2026, projected to reach $570 billion by year‑end, with major tech firms issuing $159 billion in corporate bonds in the first five months—far exceeding their borrowing in the prior five years. • Bond investors have treated AI infrastructure debt as low‑risk, but the rapid expansion of data‑center and chip financing has created a potential maturity mismatch, where the value of physical AI assets could decline before the long‑dated debt is repaid. • Regulatory uncertainty is growing, highlighted by the proposed Sanders‑AOC AI Data Center Moratorium Act and multiple state and local actions that have delayed or blocked projects, prompting early signs of buyer caution in the bond market.

The AI Digital Divide Is Widening. Most Coverage Isn't Tracking It.

The AI Digital Divide Is Widening. Most Coverage Isn't Tracking It.

• The AI digital divide is widening, with developing nations facing significant barriers to accessing AI technology due to limited internet connectivity and infrastructure. Mobile data in Sub-Saharan Africa costs fourteen times more than in Europe. • In the United States, 19 million Americans lack broadband internet, with a disproportionate concentration in rural states, while connectivity gaps persist across rural Europe. This limits access to AI tools requiring reliable internet and cloud access. • The global distribution of AI benefits is skewed, with only 3% of projected $19.9 trillion in benefits expected to go to Latin America and a mere 8% to the combined populations of Africa, Oceania, and other Asian markets excluding China.