Anthropic Locks In $45 Billion with Nscale as Compute Ambitions and IPO Countdown Converge
A six-year, $45 billion compute lease with Nscale at West Virginia's Monarch Compute Campus caps a summer of infrastructure deal-making that reframes Anthropic's public offering story — and raises pointed questions about capital sustainability.
When Anthropic and European compute provider Nscale sealed a six-year, $45 billion contract to build out the Monarch Compute Campus in West Virginia — anchored by deployments of Nvidia's next-generation Vera Rubin accelerators — the scope of the commitment announced something about where the AI infrastructure race now stands. This is not a tactical rental agreement but a decade-scale capital lock-in, one that positions Anthropic alongside hyperscalers in infrastructure depth even as it remains, for now, a private laboratory weeks from a landmark public offering.
The Nscale agreement is the largest in a series of compute accumulation moves that Anthropic has executed across a compressed timeframe. Riot Platforms disclosed a $9.1 billion compute lease covering 191 megawatts at its Rockdale, Texas campus; TeraWulf received Kentucky Public Service Commission approval for a 482 megawatt facility tied to an Anthropic AI data-center campus; and Bitdeer entered a $4.7 billion data-center lease with a cloud startup linked to the same buildout. Collectively, these deals place Anthropic's infrastructure commitments well into the hundreds of billions of dollars over their contracted terms. Anthropic's annual compute expenditure is already reported at $19 billion — a figure that, combined with these forward obligations, makes the company one of the largest buyers of AI infrastructure on the planet. Semiconductor analyst Dylan Patel has stated publicly that OpenAI and Anthropic together will control the majority of global AI compute capacity by 2028; the current contracting pattern lends that view at least circumstantial weight.
The timing of this infrastructure surge is inseparable from Anthropic's preparations for a public market debut. The company confidentially submitted a draft S-1 to the Securities and Exchange Commission in mid-August 2026, targeting a Nasdaq listing in October at a valuation in the range of $965 billion — consistent with the terms of its $65 billion Series H financing round. Some secondary-market activity has reportedly implied a valuation exceeding $1 trillion; certain commentators have circulated a $2 trillion target, though figures of that magnitude warrant caution ahead of public disclosure. What substantiates the capital story is Anthropic's financial inflection point: the company reported its first quarter of positive operating income in Q2 2026, and, according to multiple reports, its quarterly revenue surpassed that of OpenAI for the first time in that same period. That milestone reframes a competitive narrative that had long positioned Anthropic as the well-funded but commercially trailing challenger.
Even as it locks in long-term GPU capacity through partners like Nscale, Anthropic has been quietly advancing a silicon independence strategy. The company has reportedly allocated funding to develop proprietary inference accelerators — with $19 billion in annual compute costs providing obvious structural incentive to reduce third-party dependence. According to reports, Anthropic has recruited a veteran of Google's TPU program to lead the initiative and has committed $250 million to Fractile, a pre-revenue chip startup. Separately, Broadcom is said to be exploring an $80 billion debt facility to fund custom silicon production aligned with Anthropic's requirements. These moves sit in deliberate tension with the Nscale commitment: Anthropic is simultaneously locking in Nvidia's latest architecture at scale while building toward the institutional capability to eventually reduce that dependency. The approach echoes what CEO Dario Amodei has said publicly — that artificial intelligence centralizes by its nature, and that open-source models merely transfer power to whoever controls the underlying hardware. By investing in its own silicon, Anthropic is trying to occupy both sides of that equation.
The risks are commensurate with the ambition. A structural concern surfaced this week in the Riot Platforms deal: Riot's bridge loan reportedly expires before its compute lease payments to Anthropic are scheduled to begin, introducing counterparty risk into what should be a revenue-generating asset. More broadly, the capital intensity of Anthropic's strategy — multi-year commitments spanning several major counterparties — means that durable revenue growth is not optional but load-bearing. Commentary from 24/7 Wall St. and market analyst Chamath Palihapitiya has questioned whether an IPO valuation at the high end of reported ranges can be supported by the current cost structure. On the product side, some reports have noted user migration from Claude to competing platforms, including xAI's SuperGrok, attributed in part to delays in frontier model releases. The Pentagon's reversal on a directive to remove Anthropic software from weapons systems, while favorable in outcome, is a reminder of the regulatory unpredictability the company faces as it enters public markets. Australia's government has also signaled it intends to formally present infrastructure and energy demands to both Anthropic and OpenAI.
Three signals will clarify whether the current trajectory holds. First, the public S-1 filing — expected before any October listing — will expose the unit economics, contract concentration, and debt structure underlying Anthropic's infrastructure commitments; the granularity of those disclosures will define the terms of public-market debate in a way that no amount of pre-IPO coverage can. Second, the progress of Anthropic's in-house silicon program will determine whether the company can structurally compress its compute cost curve or whether the $19 billion annual bill scales proportionally with revenue. Third, the revenue relationship with OpenAI warrants continued scrutiny: Q2's overtake was corroborated across multiple outlets, but OpenAI continues to iterate aggressively and a single quarter's lead is not a trend. Anthropic has compressed years of competitive distance into a narrow window. Whether it can hold that ground while managing the financial obligations of a $45 billion compute build — and satisfying public investors simultaneously — is the central question going into its market debut.