Token Shock Hits Silicon Valley’s Biggest Spenders

In a surprising turn of events, major players in Silicon Valley are grappling with the financial implications of artificial intelligence (AI) consumption as traditional pricing models become increasingly obsolete. Uber and Microsoft, two of the largest tech companies, have hit significant hurdles in managing their AI budgets, leading to a reevaluation of how they approach AI investments and their overall financial strategies.

The Shift in AI Budgeting

Uber has found itself in a precarious position, having exhausted its entire artificial intelligence budget for 2026 by April of this year. The company's Chief Technology Officer, Praveen Neppalli Naga, indicated that Uber is now “back to the drawing board,” as reported by The Information. This situation has prompted Uber's Chief Operating Officer, Andrew Macdonald, to express concerns about the productivity metrics associated with their AI investments. On the Rapid Response podcast, Macdonald stated that the connection between AI usage and tangible productivity gains remains tenuous, making it difficult to justify the costs. He noted, “That link is not there yet,” emphasizing the challenge of correlating AI efforts with the production of more valuable consumer features.

During a recent earnings call, Uber’s CEO, Dara Khosrowshahi, revealed that autonomous agents were responsible for generating approximately 10% of the committed code, while the company's research and development expenditure reached $3.4 billion in 2025, marking a 9% increase from the previous year. Despite these investments, the underlying financial mathematics of AI consumption remains problematic.

Similarly, Microsoft has faced its own challenges, leading to the cancellation of most internal Claude Code licenses by mid-May. This decision redirected engineers from the Experiences and Devices division to GitHub Copilot CLI by the end of its fiscal year on June 30, as reported by Fortune. Notably, this cancellation does not impact Microsoft's broader partnership with Anthropic, which includes a substantial investment of up to $5 billion. This shift highlights a broader trend in the industry as companies reevaluate their AI strategies in light of rising costs and uncertain returns.

The Financial Friction of Token-Based Pricing

The financial challenges encountered by both Uber and Microsoft stem from a fundamental mismatch between traditional AI pricing structures and the operational frameworks of enterprise finance teams. Historically, annual licenses and seat-based pricing provided CFOs with a predictable cost structure that facilitated budgeting and forecasting. However, the emergence of token-based consumption—where charges accumulate based on the volume of text processed and generated—has disrupted this model.

As reported by PYMNTS, the unpredictability of token-based pricing can lead to unforeseen spikes in costs due to various factors, such as internal experimentation, new product features, or poorly optimized prompts. This volatility creates a direct impact on engineering decisions, which finance teams may not be equipped to monitor effectively. The implications of this shift are significant, as engineering teams must now consider the financial ramifications of their AI usage in real-time.

Moreover, token consumption as a metric for AI adoption presents its own set of challenges. While companies increasingly rely on token usage to gauge the intensity of their AI workflows, the measure itself has limitations. A poorly structured prompt that generates excessive iterations can consume more tokens without necessarily delivering useful output, complicating the assessment of AI effectiveness.

Nvidia's CEO, Jensen Huang, underscored the potential future landscape of AI budgeting at the GTC event in March, suggesting that engineers may soon require annual token budgets that could amount to half of their base salaries. This projection illustrates the growing financial stakes associated with AI development and deployment.

The Future of AI Pricing Models

The emergence of agentic coding tools exacerbates the cost exposure compared to traditional chatbot interactions. Unlike a single-turn conversation that generates one inference call, an agentic session—which involves planning, executing, verifying, and self-correcting—results in multiple calls, further driving up costs.

In response to these challenges, companies like Anthropic have begun implementing usage-based billing for enterprise customers, while SaaS firms such as Salesforce and HubSpot are preparing to adopt outcome-based pricing models. Adobe has also announced outcome-based pricing for its new AI product suite, reflecting a shift towards pricing structures that better align costs with actual outcomes. GitHub, too, is transitioning all Copilot plans to usage-based billing through AI Credits, moving away from flat-rate licensing that obscured true consumption levels.

Google's CEO, Sundar Pichai, provided context for the scale of token consumption during the recent I/O event, revealing that the company now processes an astonishing 3.2 quadrillion tokens per month, a sevenfold increase from the previous year's 480 trillion. Pichai acknowledged the implications of this consumption dynamic, suggesting that the term “tokenmaxxing” might aptly describe the current landscape.

As these developments unfold, it is clear that the financial landscape surrounding AI is evolving rapidly, necessitating a reevaluation of budgeting practices and pricing models within the tech industry. Companies must adapt to these changes to ensure sustainable growth and effective AI integration moving forward.