Mark Zuckerberg's patience has reached its limit. The Meta CEO has openly expressed his frustration with the slow rollout of artificial intelligence agents across the company. Despite colossal investments and a painful reorganization, concrete results are not materializing quickly enough, and the billionaire has made it clear to his teams that the current pace is insufficient.
Why is Meta spending so much for so little visible results?
Money doesn't solve everything, and Meta is learning this the hard way. The company has projected spending up to $145 billion for 2026 alone to build its AI infrastructure. This staggering investment, however, has not been enough to close the gap with rivals like OpenAI and Anthropic. The goal was to develop AI capable of automating complex processes and maintaining a competitive edge. So far, that bet is slow to pay off.
Zuckerberg remains convinced that the strategy will eventually bear fruit, but he has given his teams a three- to six-month horizon to show the first signs of improvement. This pressure comes on top of engineers already strained by a chaotic reorganization and extremely high expectations.
What is the human cost of this AI race?
The social toll of this transition is heavy. To fund its AI push, Meta has let go of nearly 8,000 employees in 2025, representing 10% of its administrative workforce. Zuckerberg himself described the layoffs as "not as clean" as they should have been. At the same time, another 7,000 staff were reassigned, sometimes forcibly, to AI-focused divisions, creating palpable unease.
The brutal reorganization, justified by the fear of inertia, has instilled a genuine "culture of fear" among teams. Numerous testimonies describe a stifling work environment where everyone fears the next wave of cuts. This is the paradox of a company with sparkling profits but deteriorating internal climate under the pressure of a more demanding Mark Zuckerberg.
Is employee surveillance the magic solution?
When strategy stalls, methods harden. Meta launched a controversial program called the "Model Capability Initiative." Its purpose was to track clicks, keystrokes, and browsing behavior of its own employees to train AI models. This initiative turned engineers into test subjects for their own potential replacement, sparking immediate internal backlash.
More than 1,600 employees signed a petition demanding the end of what they called a "data extraction factory." The program was suspended after a leak exposed personal data. CTO Andrew Bosworth later announced that the tool would return, but on a voluntary basis. This backtracking speaks volumes about the tensions currently shaking Meta.
Background: Meta's AI journey
Meta's AI strategy has been a rollercoaster. In 2024, Zuckerberg declared that Meta would be an "AI-first company," pivoting resources away from the metaverse ambitions of 2022–2023. The company released its Llama 3 and 4 open-source models, but they have not achieved the market dominance of OpenAI's GPT-4 or Google's Gemini. The decision to keep Llama open source was controversial, as it allowed competitors to build on Meta's work without paying licensing fees.
Meanwhile, the race for AI talent has been fierce. Meta lost several key researchers, including Yann LeCun, the chief AI scientist who left in late 2025. LeCun disagreed with the company's direction toward large language models (LLMs), which he considered a dead end. His departure marked a significant blow to Meta's AI research credibility.
The company has also faced external pressures. Regulatory scrutiny in the EU and US over data privacy has hampered its ability to train AI on user data from Facebook, Instagram, and WhatsApp. This forced Meta to seek alternative data sources, leading to the controversial employee monitoring program.
Financial context: The $145 billion bet
Meta's capital expenditure for 2026 is projected at $145 billion, a 60% increase from 2025. This spending is mainly on data centers, Nvidia H100 and B200 GPUs, and energy infrastructure to power AI training. The company expects that generative AI will be its primary revenue driver by 2028, replacing declining ad revenue growth. However, analysts warn that such massive spending without clear returns could lead to a stock correction. Meta's stock has already dropped 12% in early 2026 as investors question the ROI.
Zuckerberg has defended the spending, arguing that being late in AI is far more costly than overspending now. He points to Microsoft's early investment in OpenAI as a cautionary tale: Meta cannot afford to let competitors dominate the next platform shift. Yet internal sources say the urgency has created a crisis atmosphere in Menlo Park.
Internal culture: Fear and pressure
The "culture of fear" is not just employee sentiment; it has been documented in internal memos. A leaked memo from the VP of engineering stated that "mediocrity will not be tolerated" and that teams must "move at the speed of AI, not the speed of humans." This has led to burnout and high turnover. Meta's attrition rate rose to 22% in early 2026, compared to an industry average of 15%.
Rumors of forced RTO (return to office) policies have also surfaced. Meta previously allowed remote work, but Zuckerberg has hinted that in-person collaboration is essential for AI innovation. Employees fear that refusal to relocate could be a factor in future layoffs.
Competitive landscape: Meta vs. OpenAI, Google, and Anthropic
While Meta has huge resources, its AI products have not gained the traction of competitors. The open-source Llama models are popular among developers, but they lack the enterprise adoption of OpenAI's ChatGPT and Google's Gemini suite. Meta's attempt to integrate AI into social media features, such as AI-powered chatbots on Facebook and Instagram, has been met with mixed user reception. Some see it as gimmicky, while others worry about privacy.
Another challenge is that Meta's AI is built on data from its social platforms, which may not be as diverse as the web-scale data used by OpenAI. The company has struggled to train models that understand complex reasoning without hallucinating. A recent internal benchmark showed that Meta's strongest model, Llama 4 Ultra, still lags behind GPT-5 by 8% on standard reasoning tasks.
To catch up, Zuckerberg has authorized a controversial partnership with a smaller AI startup specializing in reinforcement learning, but the details remain under wraps.
The Model Capability Initiative: What went wrong?
The "Model Capability Initiative" (MCI) was designed to collect high-quality data from employees performing real tasks. The software tracked every keystroke, mouse movement, and web page visited during work hours. Employees were not given the option to opt out, leading to outrage. The data was used to train AI agents to automate software development, customer support, and content moderation tasks.
Critics argued that the program violated employee privacy and created a toxic environment where workers felt like lab rats. After the data leak exposed personal information of 1,200 employees, the program was halted. Bosworth's promise to reintroduce it on a voluntary basis has not reassured staff, as trust has been severely damaged.
This incident has also attracted attention from labor unions and the US Department of Labor, which is investigating whether Meta violated any employee monitoring laws. California's privacy laws may also come into play, as the company did not obtain consent for such extensive data collection.
What's next for Meta AI?
Zuckerberg's three- to six-month deadline means that by the end of 2026, Meta must show demonstrable improvements in AI performance and product integration. The company is working on a new flagship model, Llama 5, which it claims will surpass GPT-5 in reasoning and creativity. It is also testing AI agents that can autonomously manage ad campaigns and generate personalized content for users.
However, the road ahead is fraught with challenges. The loss of top talent, regulatory hurdles, and internal discontent could derail the timeline. Meta is also facing a potential class-action lawsuit from former employees over the MCI program. If the lawsuit proceeds, it could set a precedent for workplace surveillance in the tech industry.
Zuckerberg's frustration is understandable: he bet the company on AI, but the returns have not yet matched the investment. The next few months will determine whether his gamble pays off or whether Meta becomes another cautionary tale of overreach in the AI arms race.
Source:Génération NT News