Enterprises Push Back on Token-Based AI Costs

Uber COO Andrew Macdonald says rising AI spending hasn’t produced proportional productivity gains, and Uber exhausted its annual AI token budget within months. The post argues that if other companies report similar ROI disappointments, the AI investment bubble could burst. It cites related signs: Microsoft curtailing Claude Code licenses reportedly over costs, Target worrying about AI agent pricing models, and Starbucks halting an AI inventory pilot due to trust issues. The author warns that lofty valuations and upcoming mega-IPOs assume endless customer demand; weakening real-world returns could derail projections, strain banks and index funds, and trigger market corrections.

Uber president and COO Andrew Macdonald said the company is struggling to justify rising AI spending after reportedly exhausting its annual AI budget within four months of 2026. Macdonald told Rapid Response that growing token consumption for Claude Code hasn’t translated into clear productivity or more useful consumer features, making it hard to draw a direct return-on-investment line. Uber spent $3.4 billion on R&D in 2025 (up 9%), and CEO Dara Khosrowshahi has signaled the company is offsetting AI costs by hiring fewer employees. Macdonald framed future decisions around the trade-off between token costs and headcount if feature gains remain unclear over coming quarters.

Uber president Andrew Macdonald warned that the company is struggling to justify rising AI costs after reportedly burning through its annual AI budget within four months of 2026. Macdonald said Uber sees soaring token consumption for models like Anthropic’s Claude Code but cannot draw a clear line from that usage to more or better consumer features, making the trade-off between token costs and headcount harder to justify. Uber spent $3.4 billion on R&D in 2025 and CEO Dara Khosrowshahi has signaled hiring reductions to offset AI spend. The comments underscore broader industry questions about measurable ROI from large-scale generative AI adoption.

Uber says it burned through its entire 2026 AI budget in the first four months and is now questioning whether the spending is delivering measurable user-facing benefits. COO Andrew Macdonald told Rapid Response that despite spikes in token usage for Anthropic’s Claude Code, Uber has not been able to tie that backend consumption to a concrete increase in useful features or a 25% uplift in functionality. The company spent $3.4 billion on R&D in 2025 and CEO Dara Khosrowshahi recently noted hiring slowdowns to offset rising AI costs. Uber is reassessing the trade-offs between token costs, personnel, and demonstrable product impact as it seeks clearer ROI from AI investments.

Microsoft and Uber have paused or rethought large-scale AI coding tool use after runaway token-based costs made continued licensing unsustainable. Microsoft is moving many teams from Anthropic's Claude Code to GitHub Copilot CLI by June 30, and Uber's CTO says their 2026 AI coding budget was exhausted quickly, forcing a rebuild. The article argues current AI coding economics reward token-heavy behavior—bigger context windows and repeated re-reading of codebases—so providers earn more as users consume more. It says durable, compressed memory for models (selective, persistent representations of code and decisions) would cut costs and improve quality, but that poses a conflict of interest for model vendors who profit from token billing. If hyperscalers won’t build memory, everyone else faces a steep cost barrier.

Enterprises are pushing employees to use AI to boost efficiency, but growing compute and token costs are raising new financial pressures. Microsoft recently rescinded broad employee access to Anthropic’s Claude Code after internal usage ballooned; Microsoft still maintains its strategic Anthropic investment and Azure purchase commitments. Uber exhausted its AI tooling budget for 2026 within four months after encouraging heavy use. Nvidia executives warn that compute costs now exceed staff salaries for some teams. Analysts from Goldman Sachs and Gartner project massive token consumption and falling per-token costs by 2030, but total spending may still rise as agent-style models consume far more tokens. The trend complicates companies’ AI-driven cost-saving expectations.

Microsoft has canceled its internal Claude Code pilot after token-based billing consumed the Experiences & Devices division’s full annual AI budget within months; the pilot, launched in December 2025, will end June 30, 2026. The cost shock came when flat seat licenses masked token consumption until usage-based pricing revealed unsustainable spend, prompting Microsoft to redirect developers to GitHub Copilot. The episode highlights systemic risks of token-metered billing for enterprise deployments and undercuts Anthropic’s pitch as it raises at a reported $900B valuation while losing a marquee internal customer. The story signals opportunities for AI FinOps tools and for vendors offering predictable flat-rate pricing, and suggests procurement will demand tighter spend controls on future AI pilots.

Microsoft has canceled its internal Claude Code pilot after token-based billing exhausted the Experiences & Devices division’s full annual AI budget within months; the pilot, started in December 2025, will end June 30, 2026. The cost shock came when flat-seat licensing hid token consumption and a switch to usage-based pricing revealed unsustainable spend, prompting Microsoft to redirect developers to GitHub Copilot. The episode exposes a wider enterprise risk: frontier AI APIs billed by tokens can create runaway costs that procurement teams aren’t set up to forecast or cap. It also complicates Anthropic’s $900B fundraising narrative by showing a marquee customer walking away and highlights market opportunities for FinOps tools and predictable pricing tiers.