$1.02 billion. That’s the net loss Nscale posted against $140.6 million in revenue in the first half of 2026. Roughly $7.25 evaporating for every dollar that came in the door. And the company is asking public markets to value it at $30 billion.
I review AI tools for a living, which means I spend most of my time looking at products people actually touch. Nscale isn’t that. It’s the plumbing underneath the products, the British cloud provider renting out compute so somebody else’s chatbot can answer your email. But the numbers in this IPO filing say something about the whole AI buildout that no demo video will tell you, so let’s look at them honestly.
The growth number is real, and it’s almost meaningless
Revenue up 1,252% in H1 2026. That’s the headline, and Reuters ran it, and every finance newsletter will repeat it. It is also the least informative statistic in the filing.
Percentage growth off a tiny base is a magic trick. If you did $10 million and then you did $140 million, congratulations, you have posted a number that looks like a rocket launch. The absolute figure matters more: $140.6 million in six months. Annualize it naively and you’re at roughly $281 million. Against a $30 billion ask, that’s somewhere north of 100x forward revenue, and over 200x the half-year figure. Software companies with actual margins don’t trade there. Infrastructure companies with data centers, power contracts, and depreciation schedules definitely don’t trade there.
The 1,252% is doing a lot of load-bearing work in this story. I’d treat it as marketing copy that happens to be factually accurate.
Where a billion-dollar loss comes from
A $1.02 billion loss on $140.6 million of revenue is not a company with a sloppy expense policy. That gap is structural. Building AI cloud capacity means buying GPUs before customers exist, securing sites and power before racks are full, and financing all of it up front. The revenue arrives later, in installments, assuming the customers stay.
That’s a legitimate business model. It’s also the same model that has eaten capital-intensive infrastructure companies alive in every previous technology cycle. The question isn’t whether Nscale can grow. Clearly it can. The question is whether compute pricing holds long enough for those assets to pay themselves off before they’re obsolete.
Nobody in the filing can answer that, because nobody knows.
The Nvidia connection cuts both ways
Nvidia’s backing is the first thing every headline mentions, and I understand why. It’s a credibility stamp from the company that currently sets the terms for the entire AI hardware market.
Look at the structure of that relationship, though. Nvidia sells chips. Nscale buys chips. Nvidia has an investment in Nscale. That’s not a scandal, it’s how the sector works right now, but it does mean the endorsement carries less independent signal than it appears to. A supplier investing in its customer is, among other things, a supplier helping to fund demand for its own product.
If you’re an investor reading “Nvidia-backed” as a quality guarantee, read it instead as “operating inside Nvidia’s orbit.” Different thing.
What this tells us about the AI tool stack
Here’s why I care, from the applications side of the business. Every AI product I test runs on infrastructure priced by somebody like Nscale. When the compute layer is losing seven dollars per dollar of revenue and financing the gap with public equity, that cost eventually flows somewhere.
Two possibilities:
- Capital markets keep funding the buildout, compute stays artificially cheap, and the AI tools you use keep their generous free tiers and low per-seat pricing.
- Capital gets expensive or patience runs out, compute prices rise to cover real costs, and every AI product built on thin margins reprices in a hurry.
The second scenario is why I’m skeptical of AI startups whose entire pitch is a thin wrapper on somebody else’s model. Their unit economics are borrowed from a layer that hasn’t proven its own yet.
My read
Nscale is going public on the NYSE under “NSCL” because right now investor appetite for anything AI-adjacent is strong, and the filing is timed to that appetite rather than to any milestone in the business. That’s not cynicism, it’s standard practice. Companies raise when the window is open.
What I’d want before taking the $30 billion figure seriously: customer concentration, contract length, actual gross margin on compute, and how much of that $140.6 million comes from a handful of large deals that could walk. None of those are in the headline, and the headline is what most people will read.
The growth is genuine. The losses are enormous. The valuation asks you to believe the first number predicts the future and the second one doesn’t. That’s a bet, not an analysis, and anyone telling you otherwise is selling something.
🕒 Published: