Meta in talks for $10B Anthropic compute deal; neocloud stocks fall on competition fears

Meta Platforms has stirred concern in the neocloud sector with reports of a potential $10 billion（約1.6兆円） compute deal with Anthropic to run over 2 years. Unnamed sources described the talks as early-stage conversations, and neither company has officially confirmed the negotiations or disclosed specifics such as the gigawatt capacity involved. If realized, the deal would mark Meta's entry into the competitive AI compute-leasing market, a space currently dominated by specialized providers like Nebius and CoreWeave.

Meta's ambitions in AI infrastructure are already substantial. The company is building a 5-gigawatt data center in Louisiana and recently committed to a five-year, $27 billion（約4.3兆円） compute contract with Nebius for additional capacity. In January, CEO Mark Zuckerberg signaled even grander plans in a Threads post, stating that Meta expects to build tens of gigawatts of AI infrastructure this decade and "hundreds of gigawatts or more over time." These investments are driven by Meta's massive compute appetite: the company closed Q1 with 3.56 billion daily active users across its social networks and is also developing large language models and AI glasses, all of which require substantial computing resources.

The strategic question facing investors is whether Meta can transition from a net buyer to a net seller of compute capacity. Historically, Meta has used a copycat strategy to neutralize competitors, notably taking down Vine and limiting Snapchat's growth. However, the article argues this playbook is unlikely to succeed in the neocloud space, where the AI industry's rapid pace may mean market winners are already decided before Meta reaches meaningful operational scale as a compute provider.

Further complicating the picture, Hut 8—a data center provider—recently announced an expanded contract with an unnamed "high-investment-grade tenant" covering 352 megawatts with a 15-year term valued at $9.8 billion（約1.6兆円） (averaging $653 million（約1000億円） annually). If that tenant is Meta, it would signal that Meta still has significant unmet compute needs and is not yet in a position to sell excess capacity. Either way, the deal highlights that companies seeking AI compute are not waiting for Meta to launch its cloud business—they are securing capacity now.

Neocloud providers are taking steps to reduce their dependence on hyperscaler customers. Iren announced this week that it had secured $2.8 billion（約4500億円） in new contracts with "leading AI developers," without disclosing the megawatt volume but noting an average contract length of four years. The company now has 10 customers and revealed it has signed a new leading AI developer whose name it may disclose in its upcoming earnings report. Iren also raised its 2026 annual recurring revenue target to over $4 billion（約6400億円）, suggesting strong underlying demand. The article notes that even if hyperscalers eventually reduce neocloud usage, providers could mitigate the impact by securing one-to-two-year advance commitments from new customers, locking in revenue before existing hyperscaler contracts expire.

Meta Platforms finds itself at a critical juncture in the AI infrastructure market. As both one of the world's largest compute consumers and an aspiring compute seller, the company faces competing pressures. On one hand, Mark Zuckerberg's stated ambition to build tens of gigawatts of AI infrastructure this decade, combined with the 5-gigawatt Louisiana facility and the $27 billion（約4.3兆円） Nebius contract, signals aggressive capacity expansion. On the other hand, Meta's current needs are immense: serving 3.56 billion daily active users and developing its own large language models and AI hardware requires enormous compute resources.

The reported $10 billion（約1.6兆円） compute deal with Anthropic—if confirmed—would represent Meta's entry into the competitive cloud-leasing business. Yet the article suggests this move may come too late to reshape the industry the way Meta has disrupted social media platforms. The AI sector's rapid evolution means that market leadership may be decided before Meta reaches meaningful scale as a compute seller. More immediately, the uncertainty itself is affecting neocloud stocks, even as details remain sparse and unconfirmed.

Interestingly, neocloud providers appear better positioned to weather this threat than a straightforward competitive analysis might suggest. Companies like Iren have begun diversifying beyond hyperscaler clients, securing $2.8 billion（約4500億円） in new contracts with independent AI developers and targeting $4 billion（約6400億円） in annual recurring revenue in 2026. This strategic shift suggests that even if hyperscalers eventually reduce their reliance on neocloud providers, robust demand from smaller AI developers and startups may sustain the business—provided those providers can lock in multi-year commitments ahead of hyperscaler contract expirations.