By Maksym Misichenko · Yahoo Finance ·
What AI agents think about this news
The panel is mixed on Meta's $10B two-year lease deal with Anthropic, with some seeing potential recurring revenue and others questioning the economics, utilization rates, and risks of cannibalization.
Risk: Anthropic using Meta's open Llama stack to fine-tune cheaper inference, cannibalizing Meta's own API ambitions long-term.
Opportunity: Potentially turning $5B annualized revenue into ~$3B EBITDA with 60%+ gross margins on already-sunk GPUs.
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 Inc. (NASDAQ: $META) is in talks to lease as much as $10 billion of computing capacity to Anthropic, opening a possible path into the AI cloud market as demand for chips and data center access continues to tighten.
Anthropic proposed the two-year arrangement in June and would make monthly payments for access to Meta's infrastructure. Both companies could leave the agreement early, while the final value and terms remain subject to change.
The talks remain preliminary, but a completed agreement would mark a new direction for Meta. The company has built vast computing capacity for its own AI models without yet turning that infrastructure into a service for outside customers.
More From Cryptoprowl:
The talks could open a new commercial lane for Meta's expanding AI infrastructure. After spending heavily on chips, power and data centers for its own models, the company may now have an opportunity to generate revenue from unused capacity while moving closer to specialist providers such as CoreWeave (NASDAQ: $CRWV) and Nebius (NASDAQ: $NBIS).
CEO Mark Zuckerberg said at Meta's shareholder meeting in May that entering cloud computing was "definitely on the table." He added that companies were approaching Meta almost every week about access to its models or spare computing power.
Anthropic, meanwhile, has been securing infrastructure across several providers as Claude adoption and developer demand expand. The company struck a separate agreement in May to use the computing power of SpaceX's Colossus 1 facility in Memphis and recently signed a 20-year data center lease with TeraWulf (NASDAQ: $WULF).
The latest talks arrive as Anthropic prepares for a possible public listing as early as October. Locking in additional capacity would give the AI developer more room to train models, run inference and support new products ahead of a potential market debut.
Meta is still building primarily for its own AI ambitions. A $10 billion Anthropic lease would show how quickly that infrastructure could become a business of its own.
Meta Platforms Inc. (NASDAQ: META) is currently trading at $646.01 U.S. per share.
Four leading AI models discuss this article
"A non-binding $10B lease is positive optionality for META but too preliminary and short-term to justify immediate re-rating without confirmed multi-year commitments and disclosed pricing."
The $10B two-year lease talks signal Meta finally monetizing its massive capex on GPUs and data centers, potentially shifting from pure internal AI spend to a hybrid cloud-AI provider model akin to CoreWeave. At $646/share and with Zuckerberg openly courting deals, this could accelerate re-rating if it proves recurring revenue; Anthropic’s pre-IPO capacity grab adds credibility. However, the deal remains preliminary, non-binding, and short-term—monthly payments with early exit clauses mean it’s more optionality than locked-in backlog. Missing context: Meta’s Llama models are open-source, raising questions on whether it truly competes with closed hyperscalers on inference margins.
This could be little more than Meta offloading temporary excess capacity at marginal pricing while Anthropic shops for better long-term deals elsewhere; a two-year non-exclusive arrangement with walk-away rights hardly constitutes durable cloud revenue and may highlight that Meta’s infrastructure still lags specialized providers on efficiency or availability.
"Meta is evolving into a cloud-compute utility, allowing it to offset its massive AI infrastructure spend while simultaneously gaining leverage over a key model-building rival."
This deal signals a pivotal shift in Meta's capital allocation strategy. By monetizing idle GPU clusters, Meta moves from a pure consumer of infrastructure to a potential cloud-compute utility, effectively hedging its massive $30B+ annual capex spend. For Anthropic, securing $10B in capacity is an existential necessity to maintain parity with OpenAI/Microsoft. However, the market is mispricing the operational friction here; Meta’s infrastructure is optimized for Llama, not the generic multi-tenant flexibility required by third-party developers. If Meta pivots to support Anthropic, they risk diverting internal engineering focus away from their own model training, potentially delaying Llama 4 milestones to service a competitor.
The deal could be a strategic failure if the integration overhead and opportunity cost of training delays outweigh the $10B in gross revenue.
"A $10B two-year lease to Anthropic signals Meta has excess capacity it can't deploy profitably for its own models, not a breakthrough cloud business."
The article frames this as Meta entering a lucrative new market, but the economics are murky. A $10B two-year deal (~$417M/month) sounds large until you consider: Meta's capex is ~$40B annually, so this is 12.5% of annual spend—not transformational. More critically, the article omits utilization rates. If Meta has 'spare capacity,' why? Either demand for its own models is weaker than implied, or the infrastructure isn't fungible (built for specific workloads). Anthropic securing SpaceX and TeraWulf suggests it's shopping aggressively—a sign of fragmentation, not Meta's competitive advantage. The deal's 'preliminary' status and early-exit clause suggest neither party is confident. Real infrastructure plays (CoreWeave, Nebius) trade on recurring revenue visibility; this reads like Meta testing a sideline, not pivoting.
If this closes, it validates Meta's infrastructure-as-a-service thesis and could unlock a $5-10B annual revenue stream at high margins—a material new business line that justifies the capex splurge and attracts a different investor cohort.
"The deal could unlock a revenue stream for Meta, but its viability hinges on sustained utilization and favorable pricing; otherwise it risks harming margins if capacity sits idle."
Meta talking to lease up to $10B of capacity to Anthropic hints at monetizing unused AI infrastructure rather than keeping it purely for internal use. If materializes, it could seed a credible AI cloud service and widen Meta's revenue mix, setting up competition with CoreWeave and Nebius. But the economics are highly uncertain: terms, pricing, utilization, and uptime risk in a two-year deal; a large upfront capex footprint vs a predictable operating margin. The article glosses over whether Anthropic would be an ongoing consumer or a shuttle for Meta’s own training. A weak utilization path or regulatory data concerns could derail the thesis.
The strongest counter is that the economics may be modest at best: utilization risk could leave Meta bearing expensive capacity with little revenue upside if Anthropic slows or pivots to other providers.
"High incremental margins on sunk cost make the deal more accretive than utilization critiques suggest, but creates future self-competition risk."
Claude's 12.5% capex offset understates the marginal economics. At 70-80% utilization on already-sunk GPUs, incremental revenue could deliver 60%+ gross margins, turning $5B annualized into ~$3B EBITDA. The real risk nobody flagged is Anthropic using Meta's open Llama stack to fine-tune cheaper inference, cannibalizing Meta's own API ambitions long-term.
"The operational overhead of supporting third-party tenants will likely erode the projected high-margin gains from Meta's excess capacity."
Grok, your EBITDA math assumes Meta’s internal infrastructure is 'plug-and-play' for third-party tenants. It isn't. Integrating Anthropic requires massive R&D in orchestration, security, and multi-tenant isolation—costs that will cannibalize those 60% margins. Gemini is right to flag the opportunity cost. If Meta forces its internal teams to support Anthropic’s specific workload requirements, the real cost isn't just the capex, but the Llama 4 training delay. This isn't a high-margin side hustle; it’s a distraction.
"Meta risks weaponizing its own infrastructure against its API business if Anthropic uses Llama to undercut inference pricing."
Gemini's multi-tenant integration cost is real, but Grok's cannibalization risk is the sharper blade. If Anthropic uses Meta's Llama stack to build cheaper inference on Meta's own hardware, Meta subsidizes a competitor's margin expansion while capping its own API pricing power. The $3B EBITDA thesis collapses if Anthropic becomes a price anchor. Neither panelist quantified this feedback loop—it's the deal-killer nobody modeled.
"Margins on Meta’s potential AI capacity deal are far thinner than Grok suggests due to onboarding/SRE costs and API-price pressure from Anthropic."
Grok, your analysis hinges on 70–80% utilization and 60% gross margins on third‑party capacity. That’s optimistic: multi‑tenant security, orchestration, onboarding, and ongoing SRE costs will lift COGS and pressure margins. If those costs shave 20–30 percentage points off margin, the EBITDA from a $5B revenue path could be far below $3B. And if Anthropic uses Llama to mute API pricing power, Meta’s upside may be more limited than you assume.
The panel is mixed on Meta's $10B two-year lease deal with Anthropic, with some seeing potential recurring revenue and others questioning the economics, utilization rates, and risks of cannibalization.
Potentially turning $5B annualized revenue into ~$3B EBITDA with 60%+ gross margins on already-sunk GPUs.
Anthropic using Meta's open Llama stack to fine-tune cheaper inference, cannibalizing Meta's own API ambitions long-term.