After Amazon, Meta has also limited AI usage! When even big companies can't afford tokens, is it time for major AI companies to "control profit margins"?

Meta’s internal AI usage costs are spiraling out of control, and tech giants collectively hitting the brakes is bringing to the forefront a long-neglected industry contradiction: AI’s commercialization logic is being undermined by the very billing crisis it has created.

According to The Information, Meta sent an internal memo this week to around 6,000 employees, announcing an upper limit on employees’ Token usage and the creation of a real-time tracking platform to curb the exponential expansion of internal AI costs.

The memo states that Meta’s internal AI usage alone is expected to cost tens of billions of dollars by 2026. The backdrop is that Meta had previously strongly encouraged employees to integrate AI tools into daily work, but now has to hit the brakes hard.

Meanwhile, according to The Wall Street Journal, OpenAI is considering a substantial reduction in Token fees to compete for corporate clients, while the Silicon Data LLM Token spending index, which reflects market willingness to pay for AI, has fallen for seven consecutive trading days, marking the longest losing streak since January this year.

These signals combined have suddenly intensified market doubts about the sustainability of AI commercialization. For investors, the core question is no longer whether AI demand exists, but: When the tech giants themselves start pinching pennies over Token bills, how much profit margin is truly left for large model vendors?

Token Bill Out of Control: "Cash-Burning Race" Inside Meta

Meta’s tightening of AI usage was directly triggered by an internal incentive mechanism that led to a Token consumption frenzy.

Last November, Meta explicitly told employees that demonstrating "AI-driven work results" would be a "core evaluation requirement" this year, and top performers would be rewarded. This direction had an unexpected side effect this spring—some employees started participating in a competitive behavior dubbed "tokenmaxxing", racing to climb the company’s internal "Claudeonomics" leaderboard, which ranked the top 250 employees by Token usage.

According to The Information’s internal data, employees collectively consumed 60.2 trillion Tokens within one 30-day period, after which the number climbed to 73.7 trillion, and the leaderboard was taken offline. Some employees even deliberately instructed AI agents to run multiple tasks simultaneously to artificially inflate Token consumption.

Meta CTO Andrew Bosworth issued a warning in an April memo saying "No one should use AI for the sake of using AI" and emphasized that "Token usage is not, by any measure, a sign of impact". But clearly, mere warnings were insufficient to curb the relentless rise in costs.

Systematic Response: AI Gateway and Internal Tool Substitution

Facing runaway expenditures, Meta is establishing a systematic cost control mechanism.

According to the internal memo, product developers and engineers at Meta have built a centralized dashboard called "AI Gateway" to monitor the company’s AI usage and spending in real time, with automatic alert functions for anomalous consumption spikes. The company plans to officially roll out these control tools to a wider group of employees in the coming weeks and to implement a more structured Token budgeting mechanism by 2027.

At the same time, Meta is pushing employees to shift from third-party AI tools to internal solutions. According to two sources, Meta’s newly formed Applied AI Engineering division has tasked engineers with improving the internal programming assistant MetaCode (formerly Devmate), aiming to reduce dependency on Anthropic’s Claude—currently the main tool Meta engineers use in programming. The team is enhancing MetaCode through generated coding challenges and training reinforcement learning data.

It's worth noting that Meta hasn’t completely blocked access to third-party models; employees can still use tools from OpenAI, Anthropic, and Google, but internal tools will have a clearly elevated priority.

Industry Resonance: The Big Firms Hit the Brakes Together

Meta is not alone. Token bill pressure is reverberating through the tech industry.

According to the Financial Times, Amazon last month closed an internal AI leaderboard because employees were undertaking unnecessary activities to “score points,” sharply increasing the company’s computing costs. Amazon also started using a "standardized deployment" metric to assess whether engineers periodically use AI to generate useful code, rather than simply measuring Token consumption.

The Information also reports that Uber and ServiceNow exhausted their annual Anthropic tool budgets in the first few months of 2026; ServiceNow likewise began tracking daily employee usage to curb and control costs; venture capital firms are setting AI usage limits for employees as daily Token fees can easily reach thousands of dollars.

Against this backdrop, rumors of OpenAI’s price cuts carry deeper significance.

According to the Wall Street Journal, OpenAI is considering a steep reduction in the Token fees charged to users, partially to "get ahead" of Anthropic making similar moves.

OpenAI CEO Sam Altman recently publicly acknowledged that AI usage costs have become "a huge problem" and said the company will "help people get more value for less spending".

The timing of this statement is quite delicate—OpenAI secretly filed for an IPO this week, and Anthropic is similarly in the countdown to listing. Lower prices help win over enterprise customers but will directly erode both companies’ profit margins, and both are currently losing tens of billions of dollars due to the massive computing power required for AI systems.

Token Index Drops: What is the Market Repricing?

The capital market has already felt the chill.

Silicon Data’s LLM Token Spending Index has dropped for seven consecutive trading days as of June 11, marking the longest decline since January, with losses recorded in 11 out of the past 12 days.

This index measures the average payment per 1 million Tokens across the market, and since December last year, it more than doubled and kept rising until May 2026, when it suddenly turned south.

US macro strategist Andreas Steno Larsen called this chart "the most important chart in the market right now", warning that if Token pricing continues to weaken, the trading logic linking memory, broader hardware, and data centers may come to an end in this cycle.

Wall Street sees this differently. JP Morgan describes the current trend as "the smallest speed bump"; Citadel points out that the core constraint for AI deployment has shifted from "model with strongest capability" to "cost and scarcity", and users are quickly migrating to cheaper models.

Behind this divergence lies a more fundamental valuation question: Is the drop in Token consumption a sign of peaking AI demand, or is it simply the result of users rationally choosing cheaper models? The answer will directly impact the capital spending outlook and valuation logic for Nvidia, cloud providers, and the entire AI hardware chain.

Turning Point for the Commercialization Narrative

From a macro perspective, Meta’s tightening of AI usage reflects a deep rift in the commercialization narrative of generative AI.

Over the past three years, the AI industry has gone from subsidizing users, hiding costs in monthly subscriptions, to triggering an enterprise billing crisis based on Token pricing. Now, as the narrative of "the more Token consumption the better" comes to an end, the industry faces not just a pricing decision, but a more fundamental proposition of remodeling its business model.

Meta invested as much as $145 billion in annual capital expenditure, partly for expanding AI infrastructure, including data centers, AI chips, and talent recruitment. Meanwhile, the company is under pressure from investors to generate visible returns from its massive AI investments—Meta has launched paid subscription tiers on Facebook, Instagram, and WhatsApp, and plans to charge businesses using its AI commercial agents.

However, if even Meta’s own internal AI usage costs are unsustainable, the profitability prospect for its enterprise AI agent business will also be called into question. For the entire industry, what the next commercialization story will be remains unresolved.

Risk Disclosure and DisclaimerThe market has risks and investment should be cautious. This article does not constitute personal investment advice, nor does it take into account the individual investment goals, financial status, or needs of any user. Users should consider whether any opinions, views, or conclusions in the article suit their particular circumstances. If you invest accordingly, you assume your own responsibility.