The strategic rationale for corporate artificial intelligence expenditure is undergoing a rigorous stress test as token consumption accelerates without commensurate commercial value realization. On May 22, Andrew Macdonald, CEO of Uber, a company valued at over $200 billion, publicly questioned the correlation between rising token usage and tangible product improvements during a podcast appearance. Macdonald highlighted the growing difficulty in justifying escalating AI costs, coining the term 'tokenmaxxing' to describe the wasteful behavior observed within engineering teams. This sentiment was echoed earlier in May when Microsoft began reducing internal licenses for Claude Code, citing the unsustainability of token-based billing models. These concurrent developments have elevated token economics from a marginal operational detail to a central pillar in the broader debate regarding AI investment viability.

Data emerging since April paints a stark picture of fiscal strain within the sector. Uber's CTO disclosed that the company exhausted its entire annual budget for Claude Code within just four months, with monthly usage rates among its 5,000 engineers ranging from 84% to 95%. The average monthly cost per engineer fluctuated between $150 and $2,000, while the CTO himself reportedly consumed tokens worth $1,200 during a single two-hour internal demonstration, a figure that left Macdonald 'shocked beyond words.' Concurrently, reports by Tom Warren of The Verge indicated that while Claude Code gained popularity among Microsoft engineers, the billing structure rendered large-scale adoption unsustainable, prompting license reductions. GitHub further signaled a market shift by announcing that starting June 1, all Copilot plans would transition from fixed subscriptions to a pay-per-use model. This move sparked significant backlash, with the official discussion thread receiving nearly 900 negative votes after users calculated that a single agent programming session could cost between $30 and $40, effectively exhausting a $10 monthly subscription in one use.

Analysis of the broader market landscape reveals divergent interpretations of these cost escalations. The developer productivity platform Entelligence.AI analyzed data from 2,444 companies and found that, LLM Token Expenditure Index, token prices have risen by approximately 65% since the end of February, while US AI software prices increased by 20% to 37% over the past year. Optimists argue this volatility represents a temporary phase of successful transformation. Jim Schneider of Goldman Sachs projects that by 2030, agent-based AI will drive a 24-fold increase in token consumption, reaching approximately 120 trillion tokens per month, with gross margins for cloud providers and model vendors expected to turn positive within 3 to 12 months. Rich Privorotsky of Goldman Sachs suggests that the first quarter of 2026 may mark the peak of 'tokenmaxxing' as a key performance indicator, signaling an industry pivot toward 'unit effective action cost' as a more sustainable metric.

Conversely, economic research by JPMorgan indicates that the surge in new and updated Python packages on PyPI in early 2026, a trend absent during the 2022 ChatGPT launch, suggests genuine productivity gains are occurring. Valuation metrics also support a non-bubble thesis; the Mag 7 currently trades at a price-earnings ratio of approximately 20 times future earnings, significantly lower than the 52 times seen at the 2000 tech bubble peak, the 67 times in Japan in 1989, or the 34 times during the 'Nifty 50' era.

However, pessimists like Jim Covello, a semiconductor analyst at Goldman Sachs, argue that the concentration of value in the AI supply chain is unprecedented and unsustainable. Covello notes that chip companies are profiting at the expense of increased consumption across the supply chain, with NVIDIA's net profit increasing approximately 20 times since ChatGPT's launch.

Meanwhile, major cloud service providers have exhausted operating cash flows, issuing about $182 billion in data center-related debt in 2025, double the 2024 figure. Research by MIT's Nanda further underscores the disconnect, showing that 95% of companies investing in generative AI are experiencing zero returns.

The financial architecture underpinning this ecosystem involves a complex cycle between large-scale cloud providers and AI laboratories. Documents cited by The Information reveal that OpenAI and Anthropic account for more than half of the approximately $2 trillion in future cloud service commitments from Microsoft, Oracle, Google, and Amazon. This structure creates an inherent feedback loop: Microsoft's $13 billion investment in OpenAI is largely fulfilled via Azure credits, which OpenAI uses to purchase computing power, subsequently counted as cloud revenue by Microsoft. This dual role as equity investor and service provider is reflected in recent financial reports. Alphabet recorded record first-quarter profits of $62.6 billion, with approximately $28.7 billion, or nearly half, derived from the appreciation of its investment in Anthropic. Similarly, Amazon's $30.3 billion first-quarter profit included $16.8 billion in pre-tax unrealized gains from its Anthropic stake.

However, Amazon's free cash flow plummeted by 95% to $1.2 billion due to $44.2 billion in capital expenditures for data centers during the same period.

The sustainability of this model hinges on AI laboratories' ability to secure external funding to meet cloud commitments, which depends on corporate customers' willingness to pay rising token fees. Reports indicate that Anthropic currently incurs costs of up to $3 for every $1 in revenue generated. If financing slows, the credibility of cloud revenue forecasts will erode, pressuring valuation multiples. While the current situation does not fit a typical bubble scenario given historical valuation comparisons, the myth that increasing token consumption equates to successful transformation has been shattered. High consumption does not guarantee commercial value, and the bill for AI investment has arrived. Optimists predict positive ROI within 1 to 1.5 years as technology matures, while pessimists foresee further budget cuts as executives like Macdonald voice concerns over low returns. Woofun AI analysis suggests that the outcome remains uncertain, but the era of blind token accumulation is definitively over. The critical question now is whether downstream savings will materialize quickly enough to sustain the valuations of AI laboratories and cloud giants, or if the disconnect between cost and value will force a structural correction.