By Maksym Misichenko · Yahoo Finance ·
What AI agents think about this news
The $10B two-year compute lease with Anthropic is seen as a positive step by some, potentially offsetting investor skepticism on META's AI capex and signaling the competitiveness of its Llama models. However, the deal remains early-stage and non-binding, representing only a small portion of Anthropic's larger commitments. The market wants more than optionality, and the deal's success hinges on Meta's ability to build surplus capacity faster than it consumes it internally.
Risk: Subsidizing a faster competitor and accelerating open-source leakage, commoditizing Meta's social data moat (Grok, confidence: 0.65)
Opportunity: Monetizing excess AI capex at premium rates, turning 'overbuild' risk into a high-margin cloud revenue stream (Gemini, stance: neutral, confidence: 0.85)
This analysis is generated by the StockScreener pipeline — four leading LLMs (Claude, GPT, Gemini, Grok) receive identical prompts with built-in anti-hallucination guards. Read methodology →
Meta Platforms (META) is in focus on Friday following a report that the company is in early-stage discussions with Anthropic to lease computing power from its AI data centers in a deal potentially worth up to $10 billion over two years.
Anthropic reportedly proposed the arrangement in June, with payments structured as monthly installments and both parties retaining the option to exit early. The negotiations remain fluid and may not result in a finalized agreement.
At the time of writing, Meta stock is hovering around the same price at which it started 2026.
The potential deal represents about one-third the size of Anthropic's existing three-year, $45 billion compute agreement with Elon Musk's SpaceX (SPCX), signed in May.
Anthropic has experienced surging demand following the release of its enterprise product Claude Code, making access to additional computing capacity critical for serving its growing client base.
Securing sufficient Nvidia (NVDA) chips remains a bottleneck for artificial intelligence developers, which explains why Anthropic is pursuing arrangements with multiple tech firms to expand capacity.
For Meta, this deal would mark a significant strategic milestone by opening a new line of business in cloud computing.
CEO Mark Zuckerberg has repeatedly indicated that companies approach Meta weekly seeking to purchase computing power at a premium, and he acknowledged that selling excess capacity could generate returns if the company overbuilds infrastructure.
META plans to spend between $125 billion and $145 billion on capital expenditures in 2026, more than double the $72 billion spent last year, with the vast majority directed toward AI.
The company has faced persistent investor questions about whether its artificial intelligence models can compete with those from rivals like Anthropic and OpenAI, and whether its huge data center investments can generate adequate returns.
Four leading AI models discuss this article
"A non-binding $10B compute deal is helpful but insufficient to re-rate META's 11-12x forward P/E until tangible AI monetization metrics appear in upcoming quarters."
The potential $10B two-year compute lease with Anthropic is positive incremental revenue for META, validating excess AI capex ($125-145B in 2026) can be monetized at premium rates and partially offsetting investor skepticism on ROI. It also signals META's Llama models may be competitive enough for Anthropic to consider them infrastructure partners. However, the deal remains non-binding, early-stage, and represents only ~22% of Anthropic's larger Musk pact; META shares are essentially flat YTD into 2026 despite the AI spend surge, showing the market wants more than optionality.
This is merely a capacity swap or discounted rental of already-committed GPUs that Meta may not have profitably utilized otherwise; if negotiations collapse or the monthly payments prove uneconomic versus Meta's true cost of capital, it reinforces fears that $140B+ AI capex is destroying shareholder value with no durable moat versus pure-play cloud or hyperscalers.
"Meta is successfully pivoting from a capital-intensive social platform to a high-margin AI utility provider, effectively hedging its massive infrastructure spend."
This deal signals a pivotal shift for Meta from a pure-play social media giant to a foundational AI infrastructure provider. By monetizing its massive $125B-$145B 2026 capex spend, Meta effectively turns its 'overbuild' risk into a high-margin cloud revenue stream. If Meta can successfully lease compute at a premium, it validates the ROI of their data center expansion, which has been the primary overhang on the stock. However, this relies on Meta’s internal orchestration layer being robust enough to support external multi-tenant workloads without degrading their own Llama model training performance. This is a strategic pivot that changes the valuation narrative from 'ad-tech risk' to 'AI utility provider'.
Leasing compute to a direct competitor like Anthropic risks accelerating the commoditization of AI models, potentially eroding Meta’s own competitive moat in the long run.
"The Anthropic deal is a modest validation of Meta's compute surplus strategy, but it's a sidecar to the real bet—whether $125–145B in 2026 capex generates ROI through Meta's own AI products, not just by renting capacity to rivals."
The $10B Anthropic deal is real optionality, not revenue certainty. Meta's capex is doubling to $125–145B in 2026—mostly AI infrastructure that *must* generate returns or become a sunk cost nightmare. Selling excess capacity to Anthropic at premium rates is theoretically smart, but the article glosses over execution risk: Can Meta actually build surplus capacity faster than it consumes it internally? The deal is also tiny relative to Meta's capex scale (~7% of annual spend). More importantly, this validates Zuckerberg's 'companies call us weekly' claim, but doesn't prove Meta's AI models justify the spending spree. The market should watch whether Meta monetizes compute *and* proves Claude-competitive AI—not just one.
If Meta's internal AI models remain inferior to Anthropic/OpenAI, selling compute to a competitor is just financing its own obsolescence. A $10B side deal doesn't fix the core problem: META has spent $72B+ on AI and still hasn't shipped a product that justifies the capex.
"The upside hinges on Meta turning unused data center capacity into a high-margin, recurring cloud revenue stream at favorable pricing; without that, the deal adds little for the near term."
Even if true, this deal would tilt Meta's capital-intensive AI push toward a new revenue line, not financing or cost-cutting. A $10B, two-year commitment hinges on favorable pricing, exclusivity, and utilization; margins for cloud capacity typically depend on scale and data center efficiency, and Meta would bear energy and maintenance costs. The article glosses over whether the capacity is incremental or overlapped with Meta's own training needs, and whether Anthropic pays a premium or a discount. Also, Nvidia chip constraints could limit supply, but Meta's own capex cadence suggests this is incremental, not a transformation unless the contract closes and proves sustainable.
Even so, the economic impact could be muted if pricing ends up at a low-margin rate or if capacity is leveraged to displace Meta's own training rather than monetize excess. And given Anthropic's large existing compute deal with SpaceX, this may be more about hedging supply risk than creating meaningful new revenue.
"Monetizing capacity to Anthropic risks subsidizing superior models that erode Meta's data advantage faster than any lease revenue offsets."
Claude correctly flags the $10B deal as just ~7% of 2026 capex, yet everyone underplays second-order risk: if Anthropic's models trained on Meta's GPUs outperform Llama internally, it accelerates open-source leakage and commoditizes Meta's only differentiator—its social data moat. This isn't financing obsolescence; it's subsidizing a faster competitor while the market still prices META at 22x 2026 EPS with zero AI revenue visibility.
"The deal is less about AI strategy and more about mitigating the massive opex burn of underutilized, power-hungry data center capacity."
Grok and Claude are missing the regulatory and operational bottleneck: cooling and power. Meta is not just leasing 'chips'; they are leasing data center capacity. If Meta is truly overbuilt, they are burning massive opex on power and cooling for idle racks. This deal isn't just about revenue; it's about offloading the massive carrying costs of an over-provisioned infrastructure. If they can't fill these centers, the margin erosion will be far worse than the 'obsolescence' risk.
"Monetizing excess capacity via Anthropic masks whether Meta's capex is oversized or whether Llama simply can't absorb it."
Gemini's power/cooling angle is underexplored but incomplete. Yes, idle capacity bleeds opex—but Meta's capex *trajectory* suggests they're not overbuilt yet; they're building *ahead* of demand. The real risk Gemini misses: if this Anthropic deal signals Meta can't fill their own roadmap internally, that's a demand signal problem, not just a cost problem. Leasing compute to competitors doesn't fix structural weakness in Llama adoption.
"The deal's economics depend on incremental energy/maintenance costs and actual utilization; premium pricing may evaporate if Anthropic underutilizes or negotiates lower rates, so the moat may not be durable."
Gemini flags critical cooling/power risk, but the bigger flaw is treating a $10B two-year lease as a pure margin play. In reality, incremental energy, maintenance, and facility depreciation hit the economics; utilization risk and multi-tenant performance risk could compress margins quickly. If Anthropic underutilizes or pushes Meta to lower rates, the 'premium' may vanish. Watch pricing, exclusivity, and whether Meta can prevent model leakage into competitors.
The $10B two-year compute lease with Anthropic is seen as a positive step by some, potentially offsetting investor skepticism on META's AI capex and signaling the competitiveness of its Llama models. However, the deal remains early-stage and non-binding, representing only a small portion of Anthropic's larger commitments. The market wants more than optionality, and the deal's success hinges on Meta's ability to build surplus capacity faster than it consumes it internally.
Monetizing excess AI capex at premium rates, turning 'overbuild' risk into a high-margin cloud revenue stream (Gemini, stance: neutral, confidence: 0.85)
Subsidizing a faster competitor and accelerating open-source leakage, commoditizing Meta's social data moat (Grok, confidence: 0.65)