Nscale seeks $3.5 billion to deliver on the $45 billion contract it just signed with Anthropic

By Zendoric · September 5, 2026. Nscale, the AI compute infrastructure provider that has just closed a deal reported at $45 billion to supply capacity to Anthropic, is now negotiating a $3.5 billion pre-IPO round, TechCrunch reported. The deal would place it in the conversation among the big names building the AI industry's physical layer, alongside CoreWeave, which is already publicly traded after a similar trajectory.

A pre-IPO round serves several purposes at once: it strengthens the company's balance sheet, attracts strategic investors, sets a benchmark valuation ahead of the market debut and gives early investors a way to sell part of their stake. In Nscale's case, where the money is going seems obvious. Neither the data centers, nor the power supply contracts, nor the cooling and networking systems that support Nvidia's GPUs —the specialized chips on which AI models are trained and run— pay for themselves, and all that spending comes before a single dollar of revenue arrives from the Anthropic contract.

The pattern is not new. Nvidia has become an active investor across much of this ecosystem, an increasingly common practice through which the chipmaker secures future demand for its own hardware while spreading capital among the providers that package it as infrastructure for AI labs. CoreWeave paved the way with its stock market debut, and its share price is watched closely as a gauge of whether public markets reward such capital-intensive businesses. If Nscale completes the process, it will be one of the most visible tests of that appetite, above all because of the size of the contract underpinning its story.

The analysis published by TechBuzz points to the underlying risk: compute providers are betting that AI demand will keep growing, and they commit to hardware and energy spending in exchange for long-term contracts with labs whose business models are still settling. If AI spending cools, or if some lab decides to build its own infrastructure instead of renting it, companies like Nscale could be left overexposed. As sector context, that fear is not merely hypothetical: the same outlet reported in recent days a European Central Bank warning about a possible "worrying" correction in AI valuations, and a 10% plunge in Alibaba's share price after it announced a new round of AI spending worth $10.2 billion. None of those facts speaks directly to Nscale, but they capture the mood in which this deal is taking place.

Our reading is this: the round is not a sign of fragility at Nscale, it is an honest snapshot of how much it costs to build the physical base on which AI's promise rests. In the short term, the compute infrastructure business is exactly what it looks like: a leveraged bet concentrated in very few players —Nvidia as supplier and often as investor, a handful of labs as customers, and providers such as Nscale or CoreWeave in between— with a real risk of overbuilding if demand does not keep pace with what today justifies these figures. If the spending cycle corrects before revenue from AI applications matures, the ones footing the bill will be the investors who entered the chain late, not the labs that already have a signed contract.

But over the longer term, this infrastructure is precisely what is needed for AI to deliver on its most ambitious promise. The data centers, the energy and the GPUs this round finances are not an end in themselves: they are the physical plant required to train the models that, in time, could accelerate drug discovery, optimize power grids or bring down the cost of services that are out of reach for most people today. The question that really matters is not whether Nscale's stock market debut succeeds, but whether the capital being wagered today in the compute race ends up financing that second phase, or gets trapped in a tangle of cross-contracts among a handful of names financing one another.