1 hour ago
2
Meta is no longer the company that showed up late to the AI party. According to a new report from SemiAnalysis, Mark Zuckerberg’s empire is on track to hold more total AI compute capacity than either OpenAI or Anthropic by the end of 2026.
The research firm also projects Meta could surpass Google in frontier AI models within six months.
The infrastructure blitz behind Meta’s AI pivot
The centerpiece of this buildout is the Prometheus site in Ohio, a 1-gigawatt AI training cluster that represents the first of Meta’s next-generation facilities.
Meta plans to deploy two on-site natural gas plants, each rated at 200MW, to boost energy efficiency and reduce dependency on external power infrastructure.
The facilities themselves use what SemiAnalysis describes as innovative “Tent” datacenter designs. These structures are engineered for rapid deployment, allowing Meta to bring training capacity online faster than competitors using traditional datacenter construction methods.
Talent acquisition as a weapon
Meta’s most eye-catching move on this front was its 2025 deal to acquire a 49% stake in Scale AI, valued at approximately $14.3 billion. The deal was primarily motivated by a desire to bring Scale AI’s CEO Alexandr Wang and his team closer into Meta’s orbit.
SemiAnalysis notes that Meta has been offering compensation packages exceeding a billion dollars to secure leading researchers.
Internally, Meta is also experimenting with novel reinforcement learning techniques that reportedly utilize employee data tracking to enhance innovation.
What the capex trajectory signals
SemiAnalysis predicts that Meta’s capital expenditure in 2027 will see a significant increase, contradicting the slowdown thesis.
Disclosure: This article was edited by Editorial Team. For more information on how we create and review content, see our Editorial Policy.
English (US) · 