The same campus, twice
Nvidia-backed data center developer Nscale is reportedly in talks to sell additional computing power to Anthropic in a deal that could be worth about $45 billion, according to Bloomberg News, which cited people familiar with the negotiations. The talks concern a second facility at Nscale's Monarch data center campus in West Virginia, which would deliver 460 megawatts of computing power built on Nvidia's Vera Rubin AI chips once online. Anthropic also holds a right of first refusal on Monarch's expected third building, giving it an ongoing claim on the campus's future capacity beyond whatever this specific deal covers.
The number matters because it is not the first $45 billion figure attached to this relationship. In August, Anthropic agreed to spend roughly that same amount renting AI cloud computing power from the same West Virginia campus. If the reported talks close on similar terms, Anthropic's total committed spend at a single physical location would approach $90 billion, concentrating an enormous share of its compute strategy in one developer's footprint rather than spreading that risk across multiple providers and regions.
Why Nscale, and why West Virginia
Nscale has grown unusually fast for a data center developer: its total contracted value reportedly rose from $100 million to more than $103 billion in two and a half years, a trajectory that has made it a central player in the AI infrastructure buildout alongside rivals Nebius, Crusoe, and Lambda. Its own prospectus names Microsoft and Anthropic as its expected major future customers, which makes this reported deal less a surprise addition and more a continuation of a relationship Nscale itself has been counting on.
West Virginia's appeal for this kind of buildout is largely about power availability rather than any traditional data center advantage like network latency or customer proximity. Microsoft's earlier withdrawal from anchoring a flagship Nscale facility, reportedly because power connection delays pushed the expected supply date into the 2030s, is the clearest evidence that grid capacity, not capital, is now the binding constraint on how fast AI compute can actually come online.
The financing backdrop this deal lands in
This reported talk surfaces at an uncomfortable moment for AI infrastructure financing broadly. Days earlier, CoreWeave, a direct Nscale rival, saw its shares fall roughly 16% in a single session as yields on its bonds climbed close to 15%, a level that puts its debt closer to distressed territory than to a growth company's normal cost of capital. Oracle, Nebius, and Dell all fell alongside it, and S&P Global rates CoreWeave at B+, sub-investment-grade, with analysts describing its capital spending as close to hyperscaler scale without a hyperscaler's balance sheet to support it.
Anthropic itself is reportedly facing related pressure: separate Bloomberg reporting says some investors have urged the company to lower its IPO valuation target after a series of AI market setbacks. A company facing investor skepticism about its valuation while simultaneously reported to be negotiating a second $45 billion infrastructure commitment at the same campus is a tension worth watching closely, since the two data points pull in opposite directions on how confident the market should be in Anthropic's near-term trajectory.
What this says about compute concentration risk
Enterprise technology leaders evaluating their own AI vendor dependencies should note the structural choice Anthropic appears to be making here: betting heavily on one developer's single campus rather than diversifying compute commitments across providers and geographies. That concentration can pay off if Nscale's buildout executes on schedule, but it also means any site-specific problem, a power delay, a permitting fight, a construction setback, carries outsized consequences for Anthropic's model training and inference capacity.
For any enterprise relying on Anthropic's models in production, this is a useful lens for vendor risk assessment going forward: ask whether your AI provider has enough committed compute, and separately ask how concentrated that compute is geographically and contractually. A provider with $90 billion riding on one campus carries a different risk profile than one that has spread similar spending across five independent sites, even if the total dollar commitment looks comparable on a slide.
The open questions that matter more than the headline number
Neither Nscale nor Anthropic has confirmed the reported talks, and Reuters notes it could not independently verify Bloomberg's reporting, so the $45 billion figure, the facility timeline, and the final terms could all still change before anything is signed. Enterprise leaders should treat this as a signal of direction, evidence that Anthropic's compute appetite keeps growing, rather than as a confirmed commitment solid enough to build planning around, and should revisit it once either company issues an official statement or a regulatory filing locks the terms in place.
The more consequential question for the broader market is whether this kind of repeat, campus-specific mega-deal becomes the template for how frontier AI labs secure compute going forward, or whether a separate bond-market selloff among AI infrastructure builders marks the start of a broader repricing that makes deals this size harder to finance on the terms Nscale and Anthropic are reportedly discussing. Upcoming quarterly results from the major neocloud providers will be an early test of whether the market still believes this financing model works at the scale these deals assume.
