Meta Bets Up to $135B on AI Infrastructure
Meta is planning a massive capital expenditure of up to $135 billion in 2026 to reclaim AI market share. The company is partnering with AMD to build out its GPU capacity, signaling a new phase of the infrastructure arms race that could impact cloud compute costs and availability for all large-scale ML training.
This spending is a significant jump from the $72.2 billion Meta allocated in 2025 and the $39 billion in 2024. The massive increase is aimed at building out data centers, servers, and networking gear to train and run more advanced AI models, with CEO Mark Zuckerberg stating the goal is to "deliver personal superintelligence to billions." This level of investment initially caused concern among investors before the company's strong revenue growth seemed to justify the strategy. The partnership with AMD is a multi-year, multi-generation agreement to deploy up to 6 gigawatts of AMD Instinct GPUs. The first shipments, based on a custom MI450 architecture, are set to begin in the second half of 2026. This move is part of Meta's strategy to diversify its AI chip suppliers beyond Nvidia and also includes a multi-billion dollar deal with Google to use its specialized TPU hardware. This infrastructure will power Meta's open-source Llama models and AI-driven features across its apps, which are used by over 3.58 billion people daily. The company is developing AI for a range of applications, from AI assistants in WhatsApp and Messenger to generative AI tools for creators and experiences in the metaverse. The ultimate vision includes providing every user with a personal AI agent. The scale of this investment reflects an industry-wide trend, with estimates suggesting Big Tech companies like Alphabet, Amazon, and Microsoft could collectively spend around $650 billion on AI-related infrastructure in 2026. This AI arms race is turning compute power into a foundational layer of the economy, fundamentally gated by infrastructure rather than ideas. The surge in demand is already impacting the supply chain for chips and data center capacity. This massive build-out has significant energy implications. The International Energy Agency projects that electricity consumption from data centers could more than double by 2030, with AI being a primary driver. Meta is constructing hyperscale data centers, like the "Prometheus Hyperscale" facility in Ohio, to handle these AI-intensive workloads, with plans to invest over $600 billion by 2028 in these large-scale campuses.
