Tech companies are discovering that AI adoption without financial constraints creates unsustainable burn rates. Uber and other unnamed firms reported shocking quarterly token expenses after leadership demanded aggressive integration into workflows regardless of actual business value. This 'tokenmaxxing' phenomenon mirrors handing out electric drills to construction workers and expecting miracles rather than strategic application.

The fundamental problem lies in AI's unpredictable cost structure. Output tokens cost five times more than input tokens, yet developers cannot control response length. When agentic tools chain prompts between models, the number of required attempts becomes unknowable. Budget planners face an impossible task: calculating costs requires summing unknown input counts, unknown output counts, and unknown retry frequencies across varying model prices.

Major hyperscalers now face a looming crisis as enterprise customers push back on unlimited spending. Anthropic and OpenAI plan IPOs while carrying hundreds of billions in investor obligations, making reduced enterprise usage a critical threat. These companies pushed AI features onto startups as a revenue strategy, but spiraling costs may collapse this entire flywheel.

Apple's entry with privacy-focused Siri powered by Google Gemini adds consumer pressure—offering AI capabilities without per-token pricing. The disconnect between promised productivity gains and actual financial reality forces a reckoning: unlimited AI budgets are ending whether companies like it or not.