Due to uncontrolled internal AI usage and projected related expenditures reaching several billion dollars in 2026, Meta has urgently imposed limits on token consumption, established a centralized dashboard, and mandated that employees switch to its proprietary tool, MetaCode, to reduce reliance on external vendors. Zuckerberg issued a rare apology for the situation and pledged to address the organizational crisis to prevent talent attrition.
Meta is hitting the brakes on its AI ambitions. After months of aggressively encouraging employees to use AI tools, the social media giant is now moving to restrict internal AI token consumption and address an internal crisis triggered by aggressive restructuring—soaring costs and collapsing employee morale are simultaneously pressuring management.
According to an internal memo reviewed by The Information, Meta informed approximately 6,000 employees this week that its internal AI usage alone is projected to cost several billion dollars in 2026, and the company plans to formally implement a token management system centered on budgets and quotas starting in 2027. Meanwhile, CEO Mark Zuckerberg acknowledged in another internal memo that the company had 'made mistakes' during its AI-driven team reorganization and pledged to offer 'meaningful roles' to affected employees.
The sequential disclosure of these two memos reflects the dual pressures Meta faces in its AI transformation: on one hand, internal AI costs are rising exponentially; on the other, employee discontent stemming from aggressive restructuring has reached a breaking point—one employee openly vented with profanity during a company-wide live meeting attended by thousands, directly exposing internal tensions to public view.
These developments have drawn market attention. AI researcher Gary Marcus noted, 'Tokenmaxxing is giving way to tokenminimizing,' and predicted this trend would cause Anthropic and OpenAI to report weaker third-quarter revenues compared to their second-quarter performance. For Meta, striking a balance between cost control and retaining AI talent has become its most urgent managerial challenge.
Internal usage spirals out of control, token consumption hits record highs
Meta’s internal AI usage has expanded far faster than anticipated. In April, an internal leaderboard titled 'Claudeonomics' showed that Meta employees consumed 60.2 trillion tokens over 30 days—a figure that subsequently rose to 73.7 trillion. Named after Anthropic’s flagship product, this leaderboard tracked AI usage among more than 85,000 employees and listed the top 250 'super users' by consumption volume. The highest single user consumed 281 billion tokens in 30 days, which, based on Anthropic’s published pricing, could amount to millions of dollars in costs.
This leaderboard fostered a phenomenon dubbed 'tokenmaxxing'—employees competed to inflate their token usage to demonstrate AI proficiency, with some even instructing AI agents to run multiple tasks in parallel, artificially amplifying consumption. Employees received gamified incentives through badges such as Bronze, Silver, Gold, Platinum, and Jade, as well as titles like 'Session Immortal' and 'Token Legend.'
Meta CTO Andrew Bosworth issued a warning in April that 'no one should use AI just for the sake of using AI' and emphasized that 'token consumption itself is not a measure of influence in any meaningful sense.' The company subsequently shut down the Claudeonomics leaderboard. Now, internal memos reveal that Meta is building a centralized dashboard called 'AI Gateway' to monitor employee AI usage and spending in real time. It will also introduce an automated alert system for anomalous consumption, track current costs to forecast future expenditures, and inform compute resource planning and vendor negotiations.
Shifting to in-house tools, reducing reliance on external vendors
Another avenue for cost control involves steering employees toward Meta’s proprietary AI tools. The aforementioned internal memo indicates that Meta plans to guide employees away from third-party AI programming tools—particularly Anthropic’s Claude—and toward its internally developed coding assistant, MetaCode (formerly Devmate).
According to reports, Meta’s newly established Applied AI Engineering (AAI) division has assigned engineers specifically to enhance MetaCode’s capabilities, including generating high-quality reinforcement learning data—by having MetaCode repeatedly solve programming challenges to train its coding response abilities. Meanwhile, the company stated it will continue to allow employees access to third-party AI models.
Meta currently faces dual financial pressures: on one hand, the company plans capital expenditures of up to $145 billion this year, partially allocated to expanding data centers, AI chips, and talent reserves; on the other, investors are continuously pressing for returns from its massive AI investments. Meta has already introduced paid subscription tiers on Facebook, Instagram, and WhatsApp and signaled intentions to charge businesses that use its AI-powered business agents. Against this backdrop, strategies to reduce internal operational costs have become increasingly valuable. Notably, Meta is not alone: reports indicate that Uber and ServiceNow exhausted their full-year Anthropic tool budgets within the first few months of 2026, and several venture capital firms have imposed caps on employee AI usage due to daily token costs reaching thousands of dollars.
Mandatory reassignments trigger internal crisis, prompting public employee protests
Beyond cost concerns, a deeper internal crisis at Meta stems from organizational restructuring driven by its AI transformation. The Applied AI division was established in March 2026 and currently comprises approximately 6,500 engineers and product managers, many of whom were reassigned with little to no advance notice. In May of this year, Meta laid off around 8,000 employees under the rationale of advancing its AI transformation, while another 7,000 were transferred into new AI-related initiatives.
The primary responsibilities of these reassigned engineers now involve generating puzzles, writing programming challenges, and conducting model testing and evaluation to supply training data for AI models. For engineers previously accustomed to product development and feature launches, this shift is widely perceived as a professional demotion. One current employee described it as follows:
"You suddenly lose your sense of purpose—you barely interact with anyone and just mechanically repeat these tasks every week."
Another employee bluntly stated:
"Most people find this work utterly suffocating."
Excessively flat organizational structures have further exacerbated tensions. Reports indicate that in some teams within the Applied AI division, each manager directly oversees approximately 50 employees, resulting in insufficient support, unclear promotion pathways, and limited visibility to leadership. Accumulated frustrations erupted during an all-hands livestream meeting this week: one participant lost emotional control, interrupted the speaker with profanity, and demanded that attendees relay criticism to a specific AI executive, calling them "a jerk." Previously, more than 1,600 Meta employees had signed a petition calling for the termination of an internal project that collected AI training data by recording U.S. employees’ mouse clicks, keystrokes, and screen activity. Under pressure, Meta subsequently scaled back the project slightly.
Zuckerberg acknowledges missteps; management initiates emergency remediation
In the face of escalating internal turmoil, Meta’s senior leadership has made a series of public statements. Chris Cox, Chief Product Officer of Instagram, described the past several months during an all-hands meeting as a “tough” and “brutal” period, likening employees’ experience to “running a marathon in hail, with your teammates swapped out midway, while someone films you the entire time.” He offered a rare, sober assessment of AI itself:
“It is neither a god nor a demon. It’s not as good as you imagine, nor as bad as you fear.”
In an internal memo, Zuckerberg was even more direct: “Given the complexity of these changes, we made mistakes.” He pledged to provide “as much stability as possible,” announced a large-scale hackathon to be held in July, and initiated a restructuring of the Applied AI division’s management framework. According to Reuters, Zuckerberg also stated that no further company-wide layoffs are expected this year.
Analysts note that these coordinated statements indicate Meta’s leadership has recognized that the current round of reorganization poses a tangible threat to its talent pool. Engineering talent is the scarcest resource in today’s AI race; if core employees continue to feel marginalized, their potential departure could create significant and lasting repercussions at critical competitive junctures. Whether the current remedial measures can genuinely restore morale remains to be seen.
Editor/Liam