Industry leaders and executives are pushing back against the token-based pricing models used by AI labs Anthropic and OpenAI.

This friction highlights a growing tension between the rapid deployment of generative AI and the actual economic returns for the businesses paying for the technology. As companies integrate these tools into core operations, the cost of compute is becoming a primary barrier to scalability.

Uber COO Andrew Macdonald has emerged as a prominent critic of this trend, which has been described as "tokenmaxxing" [1]. The debate centers on the way AI labs charge users based on tokens—the basic units of text processed by a model—rather than by the value or outcome of the work performed [1].

Critics argue that this pricing structure drives soaring AI expenses without delivering proportional productivity gains [2]. In Silicon Valley, the conversation has shifted from the capabilities of the models to the sustainability of the business models supporting them [2].

Reports indicate that the backlash intensified in May 2026 as companies realized that increasing token usage does not always correlate with higher efficiency [2]. The current model effectively incentivizes AI labs to maximize the volume of data processed, while the clients bear the financial burden of those costs [1].

Seema Mody of CNBC said she has tracked the growing sentiment among enterprise users who feel the pricing is decoupled from real-world utility [1]. The push for a new standard in AI billing suggests that the initial "growth at all costs" phase of the AI race is meeting a period of fiscal correction.

While Anthropic and OpenAI have not yet shifted their primary billing structures, the pressure from high-profile clients like Uber suggests a potential shift toward outcome-based pricing in the future [1].

“The backlash against 'tokenmaxxing' grows as AI race heats up.”

The shift in sentiment reflects a transition from the 'experimentation phase' to the 'operational phase' of enterprise AI. When companies move from small pilots to full-scale deployment, pricing models that scale linearly with data volume—rather than value—create unpredictable and potentially unsustainable overhead. This pressure may force AI labs to move away from raw token counts toward subscription tiers or value-based pricing to retain large corporate clients.