
Meta has committed $35 billion(約5.6兆円) to CoreWeave and up to $27 billion(約4.3兆円) to Nebius for AI computing capacity, but the company is now building its own cloud service to sell excess computing power and is negotiating to lease $10 billion(約1.6兆円) to Anthropic—potentially turning it from customer into competitor for two neocloud providers whose growth has depended heavily on Meta's orders.
Summaries like this, in your inbox every morning.
Sign up free →What happened
Meta signed a $21 billion(約3.4兆円) expanded AI infrastructure deal with CoreWeave (bringing total commitments to $35 billion(約5.6兆円)) and a $27 billion(約4.3兆円) multi-year deal with Nebius, but Bloomberg reports Meta is now building its own cloud business to sell excess computing capacity—and is in talks to lease up to $10 billion(約1.6兆円) in computing power to Anthropic over two years.
Why it matters
CoreWeave and Nebius, which operate high-performance data centers rented to AI developers, now face the risk that Meta—one of their largest customers—could become a direct competitor selling the same computing capacity to the same client base. This threatens the growth trajectory of both companies, whose revenues have surged (CoreWeave to $2.07 billion(約3300億円) in Q1 2026, Nebius to $399 million(約640億円) in Q1 2026) partly because of Meta's massive commitments.
What to watch
Meta will address its AI strategy on July 29 when it reports quarterly earnings. The key question is whether Meta will primarily be a customer buying computing power, a competitor selling it, or both—a shift that could reshape the economics of the neocloud sector.
Meta Platforms has signed substantial commitments to two neocloud providers—CoreWeave and Nebius—over the past several months, signaling heavy investment in AI computing capacity. In total, Meta has $35 billion(約5.6兆円) in commitments to CoreWeave, consisting of a previous $14 billion(約2.2兆円) deal and a recent $21 billion(約3.4兆円) expanded agreement. In March, Meta signed a long-term deal with Nebius valued at up to $27 billion(約4.3兆円) over five years. Neocloud companies like these operate high-performance data centers powered by Nvidia graphics processing units and rent the computing power to customers building, training, and operating AI programs.
CoreWeave is the larger of the two providers. As of the end of March, it operated 43 data centers with 850 megawatts of active power and 3.1 gigawatts of contracted power. In the first quarter of 2026, CoreWeave reported $2.07 billion(約3300億円) in revenue, up from $982 million(約1600億円) a year earlier, though it posted a net loss of $740 million(約1200億円). Nebius is smaller, with 11 data centers and only five currently operational, but reports contracted capacity exceeding 3.5 GW. Nebius revenue surged from $50.9 million(約81億円) in Q1 2025 to $399 million(約640億円) in Q1 2026, though it incurred a net loss of $100.3 million(約160億円) for the quarter.
However, Meta appears to be reconsidering its reliance on external infrastructure providers. According to Bloomberg, Meta is now building a cloud business of its own to sell excess AI computing capacity. The move was foreshadowed when CEO Mark Zuckerberg indicated that Meta might consider selling overbuilt computing capacity to other companies at a premium. Adding to this shift, The New York Times reports that Meta is in talks with Anthropic, the startup behind the Claude large language model, to lease as much as $10 billion(約1.6兆円) in computing power to Anthropic over two years. This raises a critical question for investors: is Meta Platforms a valued customer or a potential competitor to CoreWeave and Nebius?
Meta's strategic repositioning reflects its broader ambition in AI infrastructure. The company has previously made massive bets on emerging technologies; in 2021, it changed its name from Facebook to Meta Platforms to emphasize its shift toward building the metaverse, a virtual reality space for work and socialization. That wager has largely failed—after investing $80 billion(約13兆円), Meta has begun laying off employees in its Reality Labs division. As the company shifts its focus away from the metaverse and toward AI dominance, it has demonstrated a willingness to spend aggressively and take on large technology companies such as Alphabet, Microsoft, and Amazon. Meta is expected to address its AI strategy and infrastructure plans when it reports quarterly earnings on July 29.
Meta's position in the AI infrastructure market reveals a strategic pivot that could reshape the neocloud sector. The company has become a driving force in the growth of CoreWeave and Nebius—two specialized data-center operators that build and rent computing capacity to AI developers. CoreWeave operates 43 data centers with 3.1 gigawatts of contracted power, while Nebius operates 11 data centers with over 3.5 GW of contracted capacity; both have achieved dramatic revenue growth, largely fueled by Meta's massive orders. Yet Meta's recent moves—building its own cloud service and negotiating directly with Anthropic—suggest the company is shifting from being a customer to becoming a seller of computing capacity, potentially at premium prices.
This move reflects Meta's willingness to take large, strategic bets. The company has previously committed $80 billion(約13兆円) to the metaverse (a bet that has since stalled), and is now determined to secure a dominant position in AI infrastructure. By owning its data centers and selling excess capacity, Meta can retain more profit and control over its AI operations. However, the strategy creates a fundamental conflict: the very customers (like Anthropic) that CoreWeave and Nebius depend on may now prefer to rent directly from Meta at competitive terms, or may become customers of a lower-cost Meta cloud service. This positions Meta as both a major client and a potential rival to its own infrastructure partners—a tension that investors in CoreWeave and Nebius must now reckon with.
AI-summarized, only the topics you pick — one digest a day via Email, Slack, or Discord.
Free · takes 30 seconds · unsubscribe anytime
No comments yet. Be the first to share your thoughts!
Log in to join the discussion





Get curated AI news from 200+ sources delivered daily to your inbox. Free to use.
Get Started FreeFree · takes 30 seconds · unsubscribe anytime