Borrowed Billions and the Unglamorous Plumbing Behind Your AI Agents

Nscale wants to raise $3 billion when it lists in the US. It just raised $3.36 billion before doing so. A company that hasn’t gone public yet has already pulled in more money than its own IPO is designed to bring in.

That’s not a typo, and it’s not a mistake on Nscale’s part either. It’s a signal about how the machinery under AI agents is being paid for right now, and it’s worth understanding even if you never plan to read a balance sheet for fun.

First, what on earth is a “neocloud”

If you use AI agents at work, whether that’s a customer support bot, a coding assistant, or something that reads your documents and answers questions about them, that agent isn’t running on your laptop. It’s running in a data center somewhere, on specialized chips built for AI work.

For years, if you wanted access to that kind of computing power, you rented it from one of the big three: Amazon, Microsoft, or Google. Neoclouds are the newer, smaller companies doing one thing instead of a hundred. They don’t sell you email hosting or a database service or a video streaming product. They sell AI compute. That’s the whole business.

Nscale is a British one. And the number attached to it right now is roughly $35 billion in expected valuation.

The financing, in plain terms

Here’s what was announced on Friday, stripped of jargon:

• Nscale secured $3.36 billion in convertible financing ahead of its planned US IPO
• The round was led by Third Point, a hedge fund
• $2.36 billion has already closed and is available immediately
• Nvidia, already an investor, committed an additional $1 billion expected in November
• The IPO itself targets $3 billion at an expected $35 billion valuation

“Convertible financing” sounds intimidating. It isn’t. A convertible note is a loan that can turn into ownership later, usually when a specific event happens, like a company going public. Investors hand over cash now, and instead of being paid back in cash, they can convert what they’re owed into shares.

Why structure it that way? Speed. A company that needs money this quarter can’t wait for an IPO process to finish. A convertible note lets the cash arrive now and sorts out the equity details later.

The Nvidia detail is the interesting one

Nvidia makes the chips that AI agents run on. Nvidia is also putting $1 billion into a company whose business is buying and renting out those chips.

I’m not going to characterize that as scandalous, because it isn’t unusual for a hardware company to invest in the businesses that expand demand for its hardware. But it’s a useful thing for a non-technical reader to notice. When you hear that “AI infrastructure is booming,” part of what’s happening is that the companies selling the shovels are helping fund the people digging. That’s a real dynamic, and it shapes how fast capacity gets built.

Why this matters if you just use AI agents

Most people reading agent101.net aren’t building data centers. You’re trying to figure out whether an AI agent can handle your invoicing, or why your chatbot got slower last Tuesday. So here’s the connection.

Every agent you use has a cost floor set by compute. Chips, power, cooling, buildings. Companies like Nscale exist because that floor is expensive and somebody has to own it. When billions of dollars flow into neoclouds, a few things follow downstream:

• More capacity. More places to run agents means fewer waitlists and rate limits for the tools built on top.
• More price competition. The big three having specialist rivals tends to be good for anyone paying per token.
• More geographic choice. A British provider matters to European companies with rules about where their data physically sits. That’s not a small detail if you work in healthcare, finance, or government.
• More concentration risk. The flip side. If your agent stack depends on one provider and that provider has a bad quarter, you feel it.

What I’d watch, carefully

A $35 billion expected valuation for a company that sells compute capacity is a bet that demand for AI agents keeps climbing steeply. That bet may well be right. It’s also the kind of number that assumes the future rather than describing the present.

The honest read is that we’re watching infrastructure get built ahead of confirmed demand, funded by instruments designed to move fast. That’s how a lot of genuinely useful infrastructure has been built historically. It’s also how some spectacular write-offs happened.

For now, the practical takeaway for anyone using agents is smaller and more useful than the headline number. The plumbing is getting bigger, more of it is being built outside the US, and the people who make the chips are helping pay for it. Keep an eye on who hosts the tools you depend on. That question used to be boring. It isn’t anymore.

🕒 Published: September 26, 2026