Anthropic Signs $35 Billion Lambda Deal as Compute Commitments Top $89 Billion in Two Weeks

A $35 billion cloud agreement with Nvidia-backed Lambda is Anthropic's third major compute signing in a fortnight, pushing its total announced infrastructure commitments above $89 billion as the lab prepares for a reported October IPO on its first-ever positive operating income.

Anthropic on September 2 announced a $35 billion cloud computing agreement with Lambda, the Nvidia-backed infrastructure provider, anchored at data centers in Nueces County, Texas — with bitcoin miner Hut 8 separately disclosing it had secured a role in the arrangement. The signing is the third major compute commitment Anthropic has publicized inside two weeks. It follows a $45 billion, six-year contract with Nscale at the Monarch Compute Campus in West Virginia, structured around Nvidia's forthcoming Vera Rubin GPU cluster, and a $9.1 billion, 191-megawatt, 20-year lease with Riot Platforms in Rockdale, Texas. Add a 482-megawatt Kentucky campus being developed for Anthropic by TeraWulf — cleared by state regulators in late August — and the lab's total announced or reported compute commitments now total more than $89 billion in aggregate.

The pace of these signings reflects a shared industry conviction that securing GPU access now is cheaper than competing for it at scale later. Semiconductor analyst Dylan Patel argued in late August that OpenAI and Anthropic together will control the majority of the world's AI compute capacity by 2028 — a projection that, if directionally correct, goes some way toward explaining why Anthropic is willing to sign decade-long leases with energy miners and construction-stage operators. The Nscale contract is particularly notable: $45 billion locked in across six years before Vera Rubin chips ship at scale constitutes an explicit hedge against future GPU supply constraints. For the counterparties, the economics are already registering in equity markets: JPMorgan raised its price target on Riot Platforms after the $9.1 billion announcement, and Hut 8 shares climbed on news of its Lambda role — a signal that institutional analysts now price long-duration AI infrastructure contracts as credit-quality revenue, not speculative upside.

Founded in 2021 by Dario Amodei, Daniela Amodei, and colleagues who departed OpenAI over alignment concerns, Anthropic spent much of its first five years as a distant commercial second despite consuming GPU resources broadly comparable to the market leader. That picture shifted materially in the second quarter of 2026: the company reported that its quarterly revenue had surpassed OpenAI's for the first time, and separately reported its first positive operating income. Annual compute costs had already reached $19 billion by August, per reporting at the time, underscoring how thin the buffer between top-line growth and infrastructure burn remains. The sequential move from loss-making to break-even is meaningful on its own terms, but the scale of commitments being executed simultaneously suggests the company is simultaneously growing into profitability and building expenditure that runs ahead of it.

Parallel to the leasing campaign, Anthropic is moving on several fronts to reduce long-term dependence on third-party silicon. Reports in late August indicated the lab had explored acquiring MatX, an AI chip startup, at approximately $7 billion — no deal has been confirmed. In the same period it invested $250 million in Fractile, a pre-revenue startup specializing in inference accelerators, and recruited a former Google TPU architect to lead an in-house silicon program. Separately, reports indicate Broadcom is exploring an $80 billion debt facility tied to the production of custom silicon for Anthropic's roadmap. These moves track a path already taken by Google and Meta, both of which concluded that owning the inference compute stack — not merely renting GPU capacity — is the only route to durable margin control at scale. Beyond silicon, Anthropic in late August also unveiled a Model Hardware Standard, an open interface specification designed to give AI agents uniform access to physical devices such as robotic arms and laboratory instruments, pointing to platform ambitions that extend well beyond the data center.

The infrastructure surge is unfolding against an imminent capital markets event. Anthropic is reportedly targeting a Nasdaq listing as early as October 2026, with an H-round financing at a $965 billion valuation reportedly in progress; secondary-market pricing implies a figure above $1 trillion, and some analyses have floated a $2 trillion target. Chamath Palihapitiya and other market observers have flagged the tension between those figures and the capital intensity required to sustain frontier-model development, particularly as newer entrants and open-weight models continue compressing inference margins. There are also structural questions embedded in the deal flow itself: Riot Platforms disclosed that its bridge loan matures before Anthropic's rent payments begin — a reminder that the counterparty ecosystem Anthropic depends on carries its own financial risk. The Pentagon's August decision to reverse its earlier order removing Anthropic software from weapons systems is a positive signal for government revenue, but illustrates equally how enterprise and defense contracts introduce policy volatility. On the product side, reports indicate some premium users have moved toward competing services; if that churn dynamic persists heading into the public offering, it would complicate what is otherwise a strong revenue narrative.

Three signals will be most telling in the months ahead. First, whether Lambda and Nscale can deliver compute at the scale and cost Anthropic requires: large data-center commitments have historically run into permitting delays, power procurement bottlenecks, and hardware delivery slippage that gap committed capacity from live capacity. Second, the fate of the in-house chip program — a credible tape-out milestone or a confirmed acquisition would meaningfully de-risk long-term unit economics in a way that GPU leases, however large, cannot. Third, the IPO itself: Anthropic's debut will set a reference price for frontier-AI capital formation, and any gap between the $965 billion target and public-market reception will send ripple effects across the sector. What the evidence supports, taken together and without embellishment, is that Anthropic has compressed an aggressive infrastructure buildout into an unusually short window, betting that locking in compute today is cheaper than competing for it in 2028. Whether that calculus proves correct will depend less on the contracts signed this month than on the models — and the revenue — that must ultimately justify them.