Nscale seeks $3.5B pre-IPO on a $103B backlog

Nscale, the British AI compute provider founded two years ago, is in talks to raise roughly $3.5 billion in pre-IPO financing ahead of a US listing. The structure is the interesting part, and so is the number it is being raised against: a contracted revenue backlog of about $103 billion. If your inference or training capacity routes through a neocloud, Nscale's pre-IPO financing is a live read on how that layer is actually funded.

What actually happened

Per Bloomberg, reported September 4, the package is roughly $1.5 billion in convertible notes plus about $2 billion from Nvidia. TechCrunch notes Nvidia is not a new name here — it participated in the $1.1 billion Series B that Aker led in March 2026, which followed a $155 million Series A in December 2024. Bloomberg previously reported the IPO could come as early as this month.

The $103 billion backlog figure comes from The Information, and the qualifier matters: it represents projections based on signed customer leases, not booked revenue. A large share of it traces to a single roughly $45 billion agreement with Anthropic for capacity, which TechCrunch reported on August 26.

So the shape is: a two-year-old company, a nine-figure equity history, an eleven-figure raise, a twelve-figure backlog, and a chip vendor on both sides of the transaction — supplying the GPUs and funding the buyer.

Why compute vendor financing matters for your business

You are not buying Nscale stock. You are buying tokens, and the question that reaches your P&L is whether the capacity you were quoted shows up on the date you were promised, at the price you were quoted.

Backlog answers neither. A signed lease is a commitment to pay for capacity that has to be built — substations, buildings, silicon, staff — and the delivery date on that is a construction schedule, not a software release. When a provider's forward book doubles in a month, that is demand signal, not supply. Read your own contract for what happens when the date slips: is there a credit, a termination right, or a paragraph that says "commercially reasonable efforts"?

The concentration point is sharper. When one customer accounts for a large fraction of a provider's book, your priority in a shortage is set by a queue you cannot see. That is not an argument against using neoclouds — the price is often genuinely better than hyperscaler list. It is an argument for keeping a second path warm. Keep your inference calls behind an interface you own, keep a fallback provider credentialed and tested on a real workload, and know your switching cost in hours before you need to spend them.

Sources: Bloomberg, TechCrunch.