The number is staggering: $45 billion. That's the reported value of the compute agreement between Nscale, a London-based AI cloud provider founded in 2023, and Anthropic, the AI safety-focused lab behind Claude. For context, that's nearly four times the size of CoreWeave's largest single contract—the $11.9 billion deal it signed with OpenAI in 2025. And it's roughly 1.5 times the size of Oracle's reported $25 billion agreement with the same company.
But here's what keeps me up at night: Nscale's entire public footprint could fit on a single page of CoreWeave's S-1 filing. The company has no disclosed GPU count, no published data center portfolio, no verifiable customer list. Yet it has allegedly secured a contract that would make it one of the largest AI infrastructure providers on Earth overnight.
Trust no one. Verify everything. This is the mantra that has guided my work since 2017, when I audited fifteen Ethereum-based whitepapers during the ICO frenzy and found that most "revolutionary" protocols were built on sand. The same skepticism applies here—perhaps even more so, because the stakes are higher and the information is thinner.
The Vera Rubin Gambit
Let's start with the technology, because that's where the story gets interesting. The agreement centers on Nvidia's Vera Rubin platform—the next-generation architecture following Blackwell, slated for 2026 production and 2027 delivery. This isn't a deployment contract; it's a futures contract on compute that doesn't exist yet.
Nvidia's official roadmap shows Vera Rubin combining a Vera CPU with a Rubin GPU, utilizing advanced packaging and HBM4 memory. The timeline is aggressive: initial production in 2026, with meaningful volume likely arriving in 2027. That means Nscale is committing to deliver hardware that hasn't been manufactured, using a supply chain that hasn't been tested, at a scale that hasn't been attempted.
Let me put the numbers in perspective. At current H100/H200 market prices of $25,000–40,000 per unit, $45 billion translates to roughly 1.1–1.8 million GPUs. Even at a premium Vera Rubin price point of $50,000+, we're talking about 900,000+ units. That's 50–100 large data centers, each housing 10,000–20,000 GPUs. The power draw alone would be 2–3 gigawatts—the electricity consumption of a mid-sized city.
Gold is heavy. Code is light. But compute infrastructure is heavier than both.
The Middleman's Burden
The commercial structure here is what I call the "compute intermediary gambit"—Nscale buys chips from Nvidia, builds infrastructure, and sells compute to Anthropic at a markup. CoreWeave proved this model works, but CoreWeave had billions in funding, years of operational experience, and a management team that had already delivered at scale.
Nscale has none of that. The company would need to raise at least $10 billion before 2026 just to cover data center construction and chip prepayments. That's a staggering ask for a company that, as far as public records show, has raised a fraction of that amount.
There's also the question of why Anthropic would choose Nscale over more established providers. The answer might be simpler than it appears: the majors are already locked up. CoreWeave's capacity is booked through 2027 by Microsoft and OpenAI. Lambda Labs has deep partnerships with xAI. AWS and Google are pushing their own silicon—Trainium and TPU respectively. If Anthropic wants Vera Rubin specifically, and wants it in volume, the secondary market might be the only option.
But this raises a critical question: is this a firm commitment or a framework agreement? My experience with large infrastructure deals suggests the latter. The $45 billion figure is likely a multi-year, multi-phase ceiling—the maximum value if every milestone is met, every option is exercised, and every contingency is realized. The actual committed amount could be a fraction of that.
The Liquidity Mirage
This brings me to a pattern I've observed across the industry, one that mirrors the Layer2 fragmentation problem in crypto. We're seeing dozens of AI cloud providers emerge—Nscale, Together AI, Lambda, CoreWeave, Crusoe, and a dozen others—each claiming to offer "scalable" compute solutions. But the underlying reality is that they're all competing for the same finite pool of Nvidia chips, the same limited data center capacity, and the same concentrated customer base.
Noise is cheap. Signal is rare. The signal here is that AI compute demand is real and growing. Anthropic's annualized burn rate exceeded $5 billion in 2025, and its compute needs are genuinely urgent. The noise is the assumption that every announced agreement represents actual, deliverable capacity.
I've seen this movie before. In 2021, I organized "Soulbound Berlin," a gathering of 40 artists and technologists to explore NFTs as tools for community building. I curated 12 non-transferable tokens for members, designed to encode identity without financialization. Within hours, 90% of participants had sold their tokens for profit. The gap between what we want to build and what the market actually does is always wider than we imagine.
The same principle applies here. The gap between a $45 billion announcement and $45 billion of delivered, operational compute is enormous. And that gap is where the risk lives.
The Contrarian View
Let me play devil's advocate for a moment. What if this deal is exactly what it appears to be? What if Nscale has secured Nvidia's backing—perhaps through NVentures, Nvidia's investment arm—and has a credible path to delivery?
In that case, the implications are profound. This would signal that AI compute procurement is entering a new phase where even second-tier providers can secure massive commitments. It would validate Nvidia's roadmap and potentially accelerate the entire industry's infrastructure buildout. It would give Anthropic the compute advantage it needs to compete with OpenAI, which has reportedly secured its own massive compute deals.
But even in this optimistic scenario, the execution timeline is brutal. Vera Rubin won't ship in volume until 2027. Data center construction takes 18–36 months. Power infrastructure requires years of permitting and grid upgrades. The earliest realistic deployment date is 2028, with full capacity arriving in 2029 or later.
By then, the competitive landscape will have shifted. Nvidia will be shipping its next architecture. Anthropic's competitors will have their own compute locked in. The market will have matured. The question isn't whether this deal makes sense today—it's whether it makes sense in 2028, when the hardware actually arrives.
The Verification Imperative
So what should we actually watch? Three signals matter most.
First, does Nscale announce a funding round? A company that needs $10+ billion to execute a $45 billion contract cannot stay silent for long. If we see a major raise in the next 6–12 months, that's a positive signal. If we don't, the deal is likely more aspiration than commitment.
Second, does Nvidia acknowledge the order? Nvidia has been careful about confirming specific customer commitments, but if Vera Rubin's order book is as strong as this deal suggests, we should see hints in earnings calls and supply chain reports.
Third, does Anthropic's financing activity align? A $45 billion compute commitment requires Anthropic to raise significantly more capital. If we see a massive new funding round, that's evidence the company is preparing to honor its obligations.
Summer fades. Builders remain. The AI infrastructure buildout will continue regardless of whether this specific deal materializes. The question is whether Nscale becomes a builder or becomes a cautionary tale.
I've been through enough market cycles to know that the most dangerous moment is when a narrative feels inevitable. The $45 billion deal feels inevitable right now—it's big, it's bold, and it fits the prevailing narrative of AI's unstoppable ascent. But inevitability is not the same as reality.
The next 12 months will tell us which one we're looking at. Until then, I'll be watching the funding announcements, the earnings calls, and the supply chain data. The truth is always in the details, and the details are always slower than the headlines.
Trust no one. Verify everything. And remember: in infrastructure, as in markets, the gap between announcement and delivery is where fortunes are made and lost.