Pioneer, the summit where CX leaders redefine what’s possible, is on October 7th.

Join leaders from Fin, Anthropic, and Clay for an insightful conversation on the state of AI transformation.

You’ll discover how some of the most innovative minds in CX have transformed their organizations, learn how they think about CX, and hear how they're planning for what's next.

Join the conversation in San Francisco, or tune in virtually.

Reserve your spot

If you’re new, not yet a subscriber, or just plain missed it, here are some of our recent editions.

_{Interested in sponsoring these emails? See our partnership options here.}

Subscribe to the Life Self Mastery podcast, which guides you on getting funding and allowing your business to grow rocketship. 

Previous guests include Guy Kawasaki, Brad Feld, James Clear, Nick Huber, Shu Nyatta and 350+ incredible guests.

Nscale files to list on the NYSE under NSCL. Price range still blank, with a reported $35 billion target, up from $14.6 billion in March. Revenue of $140.6 million for the first half, up 1,252%. Gross margin of negative 35%, and about negative 159% after depreciation. Net loss of $1.02 billion, $457 million of it from warrants marked up. Operating cash flow of positive $1.7 billion, almost all of it customer prepayments.

It is not a crypto miner, and it would like you to stop saying so. It is a two year old UK neocloud, spun out of an Australian bitcoin hoster, that rents Nvidia GPUs by the hour on six year take-or-pay contracts. It has $103.4 billion of contracted value, and $2.6 billion of that is live. Microsoft and Anthropic are 85% of it.

The fine print: going concern language, material weaknesses in accounting controls, and a $44.6 billion Anthropic deal that runs through an unbuilt West Virginia campus with no committed financing and walk rights if delivery slips. Nvidia is supplier, investor, rent guarantor and customer. The ask is about 90x today's run rate. CoreWeave, the closest comp, marketed at $35 billion and priced at $23 billion.

A company that was hosting a bitcoin miner two years ago filed this month to list on the NYSE at a reported $35 billion valuation.

Nscale is a UK neocloud. It buys Nvidia GPUs, plugs them into data centers with cheap power, and rents them out by the hour. It has signed $103.4 billion of contracts, mostly with Microsoft and Anthropic. The company would very much like you to call it a full-stack AI hyperscaler, "building the engine of superintelligence."

Here is what it actually is. Revenue of $140.6 million for the first half, up 1,252%, and a gross margin that is negative even before depreciation. Start with the backlog. Only $2.6 billion of the $103.4 billion is live, because the meter starts only when a cluster is delivered, and two customers make up 85% of it. Then the cash. Customers prepay about 23% at signing, which is why operating cash flow was $1.7 billion while the company lost $1 billion. The contracts are the collateral. The balance sheet is the business.

The net loss was $1.02 billion, and $457 million of it came from warrants rising in value. Nvidia sold Nscale the chips, invested $2.2 billion, guaranteed rent on a building full of its own GPUs, and rents capacity back as a customer.

Nscale began inside a bitcoin company. Arkon Energy, an Australian operator of crypto data centers, had the one thing every AI lab was about to need: sites with power already hooked up.

Nscale launched as an Arkon spinout with 40 employees, led by Josh Payne. The crypto roots ran deep. In 2024, hosting a bitcoin miner was 93% of revenue. That business ended in December 2024, which made 2025 Nscale's first full year as an AI company, at $33 million of revenue.

Early money was expensive. Sandton Capital, a New York credit fund, lent at 13% to 15% and collected warrants along the way. Aker, the Norwegian industrial group, partnered on a Norway joint venture, including a data center in Glomfjord that pipes its waste heat into a fish farm. Nvidia invested, then guaranteed rent.

Then the bottleneck in AI moved from chips to electricity, and a company with megawatts ready could bring clusters online faster than a hyperscaler waiting in line for the grid. The backlog went from $38 billion in December to $103.4 billion. In March 2026 Nscale bought a West Virginia campus. In August, Anthropic signed for up to $44.6 billion.

The risk factors read like a list of everything that has to go right, in order.

Start with supply. Every GPU is Nvidia, some customers contractually require Nvidia, and Nscale buys through three suppliers that took 53%, 37% and 10% of purchases in the first half. Orders go in just in time, with no long-term contracts guaranteeing capacity or price. Upstream of Nvidia sits TSMC and Taiwan. Upstream of the chips sits helium, which the filing says is exposed to the US-Iran conflict and the Strait of Hormuz.

Then delivery. The Anthropic agreements offer only limited relief for supply chain delays, and a late tranche can be terminated without liability. Other customers have similar delay exits. Miss a milestone and the penalties are cash, paid immediately.

Then money. There is still no binding financing for the Anthropic build. Nscale has lost $1.86 billion since incorporating in May 2024.

Then the stuff nobody controls. Tariffs now hit transformers, generators and switchgear, not just chips, and in at least one case the duty treatment changed after terms were agreed.

The last risk is the biggest. If AI gets more efficient, or moves off GPUs, the demand this entire balance sheet is built on shrinks.

Every S-1 picks a number, and the number is always big. Nscale didn't pick one. It picked a unit instead: gigawatts.

There is no dollar market size in the summary. The filing calls AI the fourth industrial revolution and the largest infrastructure build-out in history, then measures its own opportunity in power. 1.37 GW contracted. Line of sight to 10 GW. A West Virginia campus that could reach 8 GW of gross power by 2031. The pitch is that power, not chips, is the bottleneck, and Nscale already has its place in line.

The customer map has three buyers. Hyperscalers and frontier labs buy for scale, and they are nearly all of the backlog. AI natives buy for speed. Enterprises and governments buy for control, and this is where the British passport matters. Nscale pitches itself as the sovereign option for a Europe that doesn't want to rent its AI from American clouds.

Then the new edges. Figure AI signed for up to 100,000 GPUs to train humanoid robots. Anyscale brings Ray, with 740 million downloads, as a cheap developer funnel into the paid cloud.

The honest read is in the risk factors. If enterprises don't adopt AI as expected, the opportunity is smaller than it looks.

The GPU is not the product. The megawatt is the product.

I had assumed a neocloud was a GPU rental shop, a reseller with better cooling. It's the reverse. Nscale picked the power first, then built everything around it. It started assembling cheap powered land in 2022 and 2023, before it had an AI customer, because electricity is the one input you can't order from Nvidia. Where it can, it generates that power on site, off the grid.

Which means you cannot think of this as a server farm. Each cluster is engineered as one machine: prefabricated liquid-cooled halls, Grace Blackwell and Vera Rubin GPUs, and software that spots a failing node, moves the workload and swaps the hardware before the customer notices. The same cluster runs training or inference without changes.

Then it splits into two businesses off that one stack. Nscale Infrastructure rents a whole cluster to one tenant for about six years, and it is nearly all of the backlog. Nscale Cloud sells inference, fine-tuning and managed Kubernetes by the hour to everyone else, and on-demand revenue is not yet material. Anyscale is the bet that makes it material.

The GPUs are the price of admission. The power is what they actually sell, and they sell it by the hour for six years.

The contract is the financing in this company. The GPU-hour is where the money eventually shows up.

A customer signs a take-or-pay contract for about six years at a fixed price per GPU-hour and pays roughly 23% of it up front. That cash funds the build. It is why deferred revenue went from $2.0 billion to $6.5 billion in six months, and why operating cash flow was a positive $1.7 billion while the company lost $1 billion. GAAP treats a prepayment that big as a loan from the customer, so Nscale even books interest on it.

So the customer is not just the buyer. It is the first lender. Revenue only starts once a cluster is delivered and accepted, then bills monthly whether the GPUs are used or not. Contracts average 5.7 years and GPUs depreciate over five to six, so the customer pays for roughly the whole life of the chip.

The model has one dependency. No delivery, no revenue. Miss the date and some customers, Anthropic included, can walk. Everything runs on building on time.

Joshua Payne, Founder, Chief Executive Officer and Chair

Payne is a capital and power guy, not an engineer. He founded Arkon Energy in 2020 to build data centers for bitcoin mining, co-founded a SPAC chasing battery minerals, and ran seed funds on the side. By 2023 he had stitched together hundreds of megawatts of cheap land and power, then realized AI wanted it more than bitcoin did. The résumé is deals and megawatts. Which fits, because Nscale is really a power business that rents out chips. He still sits on Arkon's board. He is 32.

Alice Takhtajan, Chief Financial Officer

Takhtajan has spent her career on the other side of this exact table. More than 23 years at J.P. Morgan, most recently running equity capital markets coverage for tech, media and telecom, which means taking tech companies public. She joined in December 2025. J.P. Morgan now shares lead-left on the deal. MIT economics.

Philippe Sachs, Chief Business Officer and President, EMEA

Sachs is the sovereign guy. He led Nscale's early financing as a venture partner, and before that spent a banking career at S&P, J.P. Morgan, Goldman and Standard Chartered selling to governments. He now runs the European rollout and sovereign AI. Georgetown foreign service.

Nscale raised over $3.3 billion in equity across three series rounds, which is small next to what it plans to spend. The valuation moved fast. The last private round in March put it at roughly $14.6 billion, with Series C shares at $20.39. Goldman ran the Series B and co-ran the Series C with J.P. Morgan. The IPO target is about $35 billion, more than double in six months.

The real funding is debt, and it got cheaper as the backlog grew. Sandton Capital lent at 13% to 15% in 2023 and 2024. The newest facilities are investment grade at SOFR plus 2.375%. On top sit a $1.4 billion GPU facility, a $900 million revolver, $2.54 billion of Dell equipment financing and $3.1 billion of convertible notes signed three days before filing.

Look at who sits on the cap table. Sandton lent the money, took warrants and became a 5% holder. Aker partnered on the Norway joint venture and got paid in stock. Nvidia sells the chips, is investing about $2.2 billion, guaranteed a lease and rents capacity back. Arkon's shareholders swapped into Nscale in May.

Almost every major investor got in by doing business with the company first. The equity came second.

Nscale is not fighting for customers. Its biggest customer is also its biggest competitor.

The filing names the field: CoreWeave, Nebius, Crusoe and Lambda, plus SpaceX, which means the xAI side of the house and its plan to put data centers in orbit. Every one of them rents Nvidia GPUs by the hour. Every one of them is racing for the same powered land. Nscale's edge is cost, power it says runs about 70% cheaper than major US markets.

But the threats do not come from that list. They come from two directions the list hides.

Above Nscale sit the hyperscalers. Azure shows up as a competitor in the same filing where Microsoft is up to $43.8 billion of backlog. Microsoft rents today because it is waiting in line for power. Once it has its own grid connections, the risk factors admit customers face low switching costs when contracts end.

Below Nscale sits the chip. Every contract assumes Nvidia. The filing warns that custom accelerators from the big clouds could pull workloads off GPUs entirely.

The real comps are CoreWeave, which marketed at $35 billion and priced at $23 billion, and Nebius. They trade at 4.6x and 9.3x forward revenue. Nscale is asking for about 30x.

Nscale did $140.6 million in revenue for the six months, up 1,252%. But cost of revenue, before depreciation, was $189.6 million, mostly rent and power as new sites came online. So Nscale spent about $1.35 to deliver every dollar it billed. Add $174 million of depreciation and the gross line is deeply negative. That is the number the business actually runs on, for now.

Below it, the spending tells the real story. Sales, general and administrative ran $217.7 million, up 1,305%, and $93 million of that increase was stock comp. Product and technology rose 549% to $51.3 million. That left an operating loss of $492 million, a margin of negative 350%.

Then the fair value line does something strange. Nscale booked a $457 million loss because its warrants, SAFEs and convertible notes got more valuable as the company did. The better the valuation, the bigger the loss. Interest expense of $95 million is odder still, because most of it is imputed on customer prepayments, not real debt. Together they turn a $492 million operating loss into a $1.02 billion net loss.

Nscale's positioning underscores its ambition to turn cheap power into the foundation of the AI economy, owning the stack from electricity to tokens rather than renting GPUs out of someone else's building. With customers prepaying about 23% of every contract to fund the build, it has scaled from $100 million to $103.4 billion of contracted value in roughly two years, with line of sight to 10 GW of power.

Bull Case: Microsoft and Anthropic have committed up to $88 billion between them. Contracts average 5.7 years, roughly the whole life of the chip, and customers pay whether they use the capacity or not. If power becomes the scarce input as compute commoditizes, the lowest-cost producer wins, and Nscale buys electricity about 70% cheaper than major US markets. Price the next year's run rate, not today's, and 10x looks reasonable.

Bear Case: Only $2.6 billion of the backlog is live. The largest contract runs through an unbuilt West Virginia campus with no committed financing and walk rights if it's late. Gross margin is negative, controls have material weaknesses, and going concern language sat in the notes. Nvidia is supplier, investor, guarantor and customer. CoreWeave marketed at $35 billion and priced at $23 billion.

Renat Gersch is the co-founder of 10xGTM and a product marketing and go-to-market strategist with 15 years of experience scaling B2B SaaS companies.

If you enjoyed our analysis, we’d very much appreciate you sharing with a friend.

Here are the options I have for us to work together. If any of them are interesting to you - hit me up!

What do you think about my bi-weekly Newsletter? Love it | Okay-ish | Stop it