Tae Kim
AI/Semiconductor Journalist, Author of Key Context
Meta is doubling down on AI, not retreating, contradicting last week's bear narrative that tanked markets. Per an internal Meta memo reviewed by Reuters, it will deploy 7 GW of computing this year and double that to 14 GW in 2027, backed by multi-year supply deals with Samsung Electronics, Sandisk and Sumitomo Electric. Mark Zuckerberg, tweeting for the first time in three years, signaled that Meta's AI model efforts are just getting started. x.com x.com x.com x.comÂ
Meta's data edge and poaching of top talent support its in-house AI development against rivals buying projects to integrate. Only Google matches its data; Anthropic and OpenAI cannot really compete without funding. Meta is taking some of Google's ad share, which should reduce Google's free cash flow. x.comÂ
Meta's negative free cash flow from AI capex is an expected investment phase, not a broken thesis. Analysts had forecast deterioration through 2027 as infrastructure spending accelerates, followed by a sharp recovery in 2028 as those investments begin generating cash. The market is treating known spending as a surprise, creating an opportunity while Meta builds infrastructure for the next decade of AI. x.comÂ
Meta's hundreds-of-gigawatts compute ambition looks hard to justify against the revenue a much smaller build might support. If Anthropic can generate over $70B in annualized revenue on 3GW or less, including training, hundreds of gigawatts is not a serious figure. Earlier Muse monetization than in past product cycles also makes little sense without direct integration into ad recommendations or an immediate shift in user behavior. x.comÂ
Meta's AI push is just getting started, with Alexandr Wang's team delivering. Per Wang, Muse Spark 1.1 is an industry-competitive agentic and coding model that rivals GPT-5.5 and Opus 4.8 on many agentic evaluations, and it is now available through Meta's model API and in Meta AI. x.com x.comÂ
Meta's advertising is less vulnerable to AI agents and LLMs because users must scroll Instagram themselves. AI may even give them more time to do so, while Meta retains other mega-caps' AI tailwinds; it remains the best AI stock. CapEx is still a problem. As @diyinvestor1 noted, Meta also has no exposure to OpenAI or Anthropic. x.com x.com x.comÂ
Meta building a cloud business to sell excess AI compute signals it is dropping out of the frontier AI race, which is very bearish for Meta alone. Per Wall St Engine, Meta is setting up that business, yet its compute is only equal to three CoreWeaves. x.com x.comÂ
Meta's AI-driven ad revenue can justify sustained AI spending, arguably a bigger NVIDIA positive than its slight capex-guidance lift. Meta raised the low end of 2026 capex guidance from $130B to $135B, left the $145B ceiling unchanged and lifted the midpoint just 1.8%. Its goal of over 10x better Muse compute-cost efficiency over the next year complicates GPU-order growth even as total serving costs are expected to rise. Spending looks durable, not necessarily set for faster NVIDIA orders. x.comÂ
Fresh funding for AI expansion supports a continued bid for AI stocks as compute spending shows no sign of slowing. Meta committed to $95B in shareholder dilution for AI, Anthropic is set to have $200B in cash in a couple of weeks, and SpaceX got a $40B Apollo loan solely to buy more Nvidia GPUs. The Anthropic IPO might even be a positive catalyst rather than the negative one many fear. x.comÂ
Meta looks unattractive because its fines, capital raises and AI capex make the math ugly. The stock could trade back to the $500s, while hyperscalers probably have to make those raises with equity down 15-25%. Muse looks like a fad and average users may not need AI agents, matching LogicalThesis's view. x.com x.com x.comÂ
Meta's 26% growth and still-unmonetized Muse make the dip buyable. ZeroHedge estimates $250 billion of AI capex next year; Meta would not spend that much unless Muse continues to unlock a massive opportunity in personal AI agents. x.com x.comÂ
