$135 billion. That's what Meta plans to spend on AI infrastructure in 2026, nearly doubling last year's $72.2 billion. Behind that staggering number is a company in the middle of the most chaotic internal reorganization in its history.
A Financial Times investigation reveals the full scope of the turmoil. It starts with Llama 4.
The Llama 4 Scandal That Broke Trust
When Meta released its Llama 4 family of models in April 2025, the reception was brutal. Independent testers quickly noticed the benchmarks didn't match real-world performance. Yann LeCun, Meta's chief AI scientist and a Turing Award winner, has now confirmed why: the results were manipulated.
"Results were fudged a little bit," LeCun told the Financial Times. The team used different models for different benchmarks to inflate performance numbers. Mark Zuckerberg was reportedly furious. According to LeCun, Zuckerberg "basically lost confidence in everyone who was involved" and sidelined the entire generative AI organization.
The largest model in the family, Llama 4 Behemoth, was delayed indefinitely.
A 28-Year-Old Takes the Wheel
Zuckerberg's response was to blow up the org chart. In June 2025, Meta acquired a 49% non-voting stake in Scale AI for $14.3 billion, valuing the data-labeling company at roughly $29 billion. The real prize wasn't the equity. It was Alexandr Wang, Scale AI's 28-year-old founder, who stepped down as CEO to join Meta as Chief AI Officer and head of the newly formed Meta Superintelligence Labs.
LeCun, who had been at Meta for over a decade, found himself reporting to Wang. He did not take it well. LeCun described Wang as "young" and "inexperienced" in research practices, and bristled at the new hierarchy: "You certainly don't tell a researcher like me what to do."
LeCun is now leaving to launch Advanced Machine Intelligence Labs, a startup focused on V-JEPA architecture - an approach that trains AI models on video and spatial data rather than text. He's betting that large language models (the technology behind ChatGPT, Claude, and Llama) are fundamentally limited. He called Meta's new AI hires "completely LLM-pilled" and views the current approach as "a dead end when it comes to superintelligence."
The Human Cost
To fund this AI pivot, Meta is reportedly planning layoffs that could affect 20% of its workforce - roughly 15,000 people out of nearly 79,000 employees. If confirmed, it would be the company's largest reduction since the 2022-2023 "year of efficiency" cuts. Meta's stock actually climbed 3% on the news, because Wall Street loves efficiency.
Meta spokesperson Andy Stone called the layoff reports "speculative reporting about theoretical approaches," which is the corporate equivalent of a non-denial denial.
Meanwhile, Meta's next flagship model is codenamed "Avocado" and is supposed to leapfrog the performance plateaus that have slowed progress across the industry. The team building it is working under a leadership structure that didn't exist a year ago, with a research culture that's lost its most prominent voice.
The math here is simple but sobering. Meta is spending more money on AI than any company in history, led by a chief AI officer who's never run a research lab, after firing the people who built the last model and losing the scientist who defined the company's AI identity for a decade. The strategy might work. But Zuckerberg is essentially betting that raw capital and data infrastructure matter more than institutional knowledge - and that's a hypothesis that $135 billion will put to the test.