Nscale, a London-based AI cloud company founded in 2024, is in talks to raise about $3.5 billion of pre-IPO financing ahead of a New York listing. Roughly $2 billion of it is expected to come from Nvidia. The other $1.5 billion is a convertible note led by Third Point, with Goldman Sachs running the process. An IPO of around $3 billion could follow.
Nscale is briefing investors that its total contracted value is approximately $103 billion, a figure that includes a $45 billion agreement to supply compute to Anthropic.
Two things in that paragraph deserve more attention than the headline number.
The chipmaker is funding the buyer
Nvidia putting $2 billion into an AI cloud provider is not a passive financial position. Nscale exists to buy Nvidia silicon and rent it out. Money that goes in as equity comes back out as hardware orders, and the revenue that results is recognised as third-party demand.
This is not unique to Nscale — it is the defining financial structure of the current buildout, and it is why the sector’s demand signals have become genuinely hard to read. When a supplier capitalises its customers, the order book stops being independent evidence of appetite. It becomes partly a function of how much the supplier chose to invest.
None of which makes the demand fake. Anthropic’s $45 billion commitment is a real contract with a real counterparty that has its own capital and its own reasons. The problem is narrower and more practical: from outside, you cannot easily separate the portion of an AI infrastructure order book that reflects end demand from the portion that reflects vendor financing. That distinction only matters when growth slows, which is precisely when nobody will be able to work it out quickly.
Our take: The most informative detail is not the $3.5 billion. It is the convertible structure. The notes convert at a double-digit discount to the IPO price, with that discount adjusting up to a $30 billion valuation — above which the conversion price stops moving. Read that backwards: the people structuring this deal built in downside protection all the way to $30 billion and capped the investors’ participation beyond it. That is a negotiated view on where the listing prices, expressed in terms far more precise than any banker’s range.
$103 billion of contracts, and a company two years old
A contracted value of $103 billion against a company founded in 2024 is a ratio that has no real precedent outside of utilities and defence. It is also the reason the pre-IPO round is being raised at all: contracts of that size require data centres, power and hardware that must be paid for years before the revenue arrives.
That is the actual business model, and it is worth saying out loud. An AI cloud provider is a project-finance business wearing a technology multiple. The contracts are long, the capex is front-loaded, the customers are concentrated, and the whole thing depends on funding markets staying open at each step. We wrote about the same dynamic when Crusoe tripled its valuation on the back of a single $13 billion contract with Jane Street, and about the debt side of it in $219 billion of hyperscaler bond issuance with thinning order books.
The concentration point is the one that gets least attention. A large share of that $103 billion sits with a small number of counterparties. Contracted revenue is only as durable as the customer’s ability to keep paying for it, and AI labs are themselves consuming capital faster than they generate it.
What to watch
- Whether Nvidia’s $2 billion actually closes. It is in talks, not signed. If the strategic leg shrinks and the convertible grows, the terms just got worse.
- The customer concentration disclosure in the F-1 or S-1. The filing will have to name what share of contracted value sits with the top customers. That number is the whole risk profile.
- Where it prices against $30 billion. The convertible cap is a live prediction. Watch whether the market agrees with it.
- The gap between contracted and delivered. $103 billion is what has been signed. Capacity actually online is a much smaller and much more useful figure.
- Power. Every one of these contracts assumes megawatts that must be secured, built and energised. That is the constraint that has been binding all year.
A two-year-old company raising $3.5 billion privately before raising $3 billion publicly is a reasonable summary of how AI infrastructure is being financed right now: quickly, at scale, and increasingly with money from the firm selling the equipment.