The $45 Billion Futures Contract: Nscale, Anthropic, and the Art of Selling What Doesn't Exist Yet
The announcement landed with the usual fanfare: Nscale, a London-based AI cloud provider with a fraction of the market presence of CoreWeave, has signed a $45 billion agreement with Anthropic to deploy Nvidia's next-generation Vera Rubin chips. The number is staggering. It is roughly four times the size of CoreWeave's largest single contract, and it comes from a company that has yet to prove it can operate a data center at scale.
Check the source code, not the roadmap. This deal is not a deployment; it is a futures contract on silicon that has not been manufactured. Vera Rubin is Nvidia's post-Blackwell platform, slated for 2026 production and 2027 delivery. The agreement, signed in 2025, means at least a 12-to-18-month waiting period before any hardware hits the ground. Nscale is essentially selling Anthropic a promise, backed by Nvidia's roadmap, that enough chips will exist to satisfy a 90,000-plus GPU deployment.
Let me run the numbers, based on my experience auditing infrastructure deals. At a conservative $50,000 per Vera Rubin unit, $45 billion implies roughly 900,000 GPUs. Each of those units will draw between 25 and 35 kilowatts under load. That is 2 to 3 gigawatts of power consumption, enough to light up a mid-sized city. The deployment would require 50 to 100 hyperscale data centers, each housing 10,000 to 20,000 GPUs. Construction timelines for that scale run 18 to 36 months. The math does not close before 2028, and that is assuming Nvidia's supply chain holds.
Here is the structural problem. Nvidia allocates its early production capacity to its largest customers: Microsoft, Meta, xAI. A newcomer like Nscale, with no publicly disclosed GPU fleet size, is not at the front of that queue. The company will need to secure not only hardware allocation but also the financing to prepay for it. Based on my 2020 DeFi audit experience, when a protocol promises returns that exceed its collateral base, you look for the hidden leverage. Nscale's collateral is its balance sheet, and there is no public evidence it can absorb a $100 billion capital raise by 2026.
The commercial logic mirrors CoreWeave's playbook, but at a scale that strains credibility. CoreWeave signed a $10 billion deal with Microsoft in 2024 and an $11.9 billion deal with OpenAI in 2025. Its market valuation after IPO was around $23 billion. Nscale is signing a contract four times larger than CoreWeave's biggest, with a fraction of the operational track record. The pricing model matters. At current AI cloud rates of $2 to $4 per GPU-hour, a single GPU generates $17,000 to $35,000 annually. Payback takes two to three years, assuming the contract spans at least five years. If Anthropic negotiated take-or-pay terms, Nscale's revenue is secured, but its execution risk remains unchanged.
Anthropic's motivation is easier to parse. The company burns through $5 billion annually and needs every compute advantage it can get against OpenAI. Its existing partnerships with AWS and Google give it access to Trainium and TPU chips, but not Nvidia's latest silicon. The deal with Nscale is a hedge, a way to secure Vera Rubin capacity without relying on the hyperscalers. It is also a supplier diversification play, reducing dependence on any single cloud provider. But the financial strain is real. A $45 billion commitment, spread over five years, is $9 billion annually. Anthropic's projected 2025 revenue is $2 to $3 billion. The gap requires continuous, massive fundraising. This is not a purchase; it is a bet on future capital markets.
The industry impact, if this deal executes, is a step-change in the AI arms race. It validates Nvidia's roadmap and cements its 80% market share. It forces OpenAI to respond with even larger commitments. It creates a new tier of AI cloud providers, with Nscale leapfrogging from obscurity to relevance overnight. But it also concentrates AI capability in fewer hands. The power requirements alone will strain grid infrastructure and carbon targets. Regulators will take notice, not for the technology, but for the concentration of compute in entities with no proven operational history.
Now, the contrarian angle. The bulls will say that Nvidia has an incentive to make this work. Nvidia's venture arm, NVentures, could provide seller financing to Nscale, reducing the initial capital burden. The deal could include equity components, with Anthropic taking a stake in Nscale, similar to the Microsoft-OpenAI structure. If Nvidia sees Nscale as a channel to lock in demand and diversify away from hyperscaler concentration, it has the tools to support this. Nvidia has historically backed multiple AI cloud providers to expand its ecosystem. Nscale could be the next CoreWeave, and the $45 billion figure, while inflated, represents a real, multi-year framework agreement with milestone-based triggers.
But here is what the hype misses. The source is Crypto Briefing, not Reuters or Bloomberg. The information asymmetry is massive. The lack of any official confirmation from Nvidia, Nscale, or Anthropic within 48 hours of the announcement is a red flag. In my 2017 ICO experience, projects with the biggest promises had the least code to back them. The same principle applies here. A $45 billion headline is noise. The signal is in the contract structure, the financing terms, and the supply agreements. None of that has been disclosed.
Hype is just noise in the signal. The real signal is that AI infrastructure deals are entering a new scale, and the risk profile is shifting. The question is not whether Anthropic needs the compute; it does. The question is whether Nscale can deliver. The answer, based on everything publicly verifiable, is that it cannot, at least not at this scale, within the promised timeline.
If the math does not work, the deal gets renegotiated. The $45 billion becomes $20 billion, the timeline extends, and the milestones get pushed. Or the deal collapses entirely, and Anthropic goes back to the hyperscalers with a stronger negotiating position. Either way, the market will learn something about how much of this is real demand and how much is speculative positioning. The contract is signed. The chips are not. That is the only fact that matters.
We are entering an era where compute is the new oil, and every major player is signing futures contracts on it. But unlike oil, the commodity here is still in the lab. The next 24 months will reveal whether this is a strategic masterstroke or a paper tiger. Watch the financing rounds, watch the supply agreements, and watch who actually takes delivery of the first Vera Rubin units. Trust the hash, not the hand. And check the source code, because the roadmap is not a delivery date.