Nscale’s IPO prospectus says that, as of 31 August 2026, about 1 GW of its roughly 1.37 GW of active and contracted capacity sat at wholly-owned sites, with 200 MW at leased sites and 165 MW in colocation (Nscale S-1). That is about 73% at owned sites, by our arithmetic.
CoreWeave’s latest quarterly filing describes leasing third-party data centers, with $16.3 billion of operating lease liabilities at 30 June 2026 and a further $35.5 billion of undiscounted lease payments on leases signed but not yet started (CoreWeave 10-Q).
Business Pill · Vertical Integration
An evergreen Business Engineer explainer on vertical integration. It teaches the general idea only and does not discuss Nscale or CoreWeave.
The key insight: Nscale’s S-1 puts about 1 GW of its roughly 1.37 GW of active and contracted capacity at wholly-owned sites, while CoreWeave’s 10-Q shows $16.3 billion of operating lease liabilities and $35.5 billion of leases signed but not yet started; both companies paid for their software acquisitions mostly in stock.
Nscale’s S-1 lists seven wholly-owned sites, nine colocation sites and one leased site among five active and twelve contracted data centers, and says it also owns a portfolio of powered land. Its founder, Josh Payne, put the strategy this way in the Anyscale announcement: “Most infrastructure providers just buy GPUs and rent them. Nscale is doing something unique. We build and own every layer ourselves: the power, the data centers, the compute, and the software that turns them into an AI cloud.” (Nscale).
CoreWeave’s 10-Q says many of its leases for third-party data centers run for multiple years at fixed capacity, and that leases not yet commenced will start between 2026 and 2029 with terms of seven to sixteen years. Its operating lease liabilities rose from $8.2 billion at the end of 2025 to $16.3 billion at 30 June 2026, about 2x in six months, by our arithmetic. The filing does not give a share of capacity at owned sites.
Nscale’s S-1 puts about 1 GW of its roughly 1.37 GW of active and contracted capacity at wholly-owned sites, 200 MW at leased sites and 165 MW in colocation, as of 31 August 2026; 73% is our arithmetic. Data: Nscale S-1.
What Was Posted on X
neoclouds want to own more parts of the business themselves, rather than paying other companies for it.
makes sense, since every company they pay means less profit and can hold them up when there are shortages
there are different ways to be vertically integrated. @CoreWeave and @nscale are taking very different approaches
at the very bottom are the land and the power. then, the transformers and switchgear, the data center building, then the gpus. at the very top is the software
@CoreWeave owns its gpus and has been buying software companies, but it doesn’t own the buildings. they lease their data centers
@nscale started with the land and power. 73% of its capacity is at sites it owns. only recently are they adding software on top by buying anyscale for $1.65b
coreweave started in the middle of the stack and moved towards customers. nscale started closer to suppliers, and is moving up
X user @jessiedong_, whose bio lists Compute & AI and CS and Physics at Stanford, posted on 7 November that neoclouds are trying to own more of their stack. The post says CoreWeave owns its GPUs and leases its data centers, while Nscale started with land and power and has 73% of its capacity at sites it owns. The 73% matches our arithmetic from Nscale’s S-1; the CoreWeave characterisation is the poster’s.
Both Use Stock to Pay for Software
CoreWeave bought Weights & Biases, an AI developer platform, in May 2025 for a total purchase price of $1,029 million, of which $929 million was Class A stock and restricted stock awards, $96 million cash and $4 million replacement stock units, its 10-Q shows; that is about 90% in stock, by our arithmetic. The filing also names Monolith AI, acquired in November 2025.
Nscale agreed in July 2026 to buy Anyscale, the company behind the open-source Ray framework, “for consideration of $1.65 billion which will be settled in our equity”, its S-1 says. The deal is subject to regulatory approvals and to Nscale completing its IPO. Nscale’s own July announcement said financial terms were not disclosed; the price appears in the S-1. Anyscale’s roughly 200 staff will join Nscale as part of the acquisition, the company said.
Our Numbers
Owned-site share: 1 GW of about 1.365 GW (1,000 + 200 + 165 MW) is 73.3% for Nscale, by our arithmetic from its S-1; CoreWeave does not disclose a comparable figure. Stock share of the two software deals: about 90% for Weights & Biases and 100% for Anyscale, by our arithmetic from the filings.
Connecting the Coverage
The two companies sit in the basket of AI cloud stocks that fell 20.9% in four sessions (our piece), and CoreWeave insiders sold $6.08 million of stock over the past month (our piece). Microsoft has been reported to be tightening oversight of Nebius, IREN and Nscale (our piece) and to have rejected an option to expand an Nscale Texas deal (our piece). Our tally of neocloud debt is here.
The Structural Read
As we read the two filings, the companies started at different layers: Nscale’s capacity is mostly on land it owns, while CoreWeave’s 10-Q describes multi-year leases for third-party data centers.
As we read the deals, both used equity rather than cash to add software: about 90% of the Weights & Biases price was stock, by our arithmetic, and the Anyscale price is to be settled entirely in Nscale equity.
As we read Nscale’s S-1, much of its capacity is contracted rather than live: it counts about 25,000 active GPUs against 461,000 active and contracted.
Josh Payne, Nscale founder and CEO
“Most infrastructure providers just buy GPUs and rent them. Nscale is doing something unique.”
Three Implications
FOR NSCALE Its S-1 says the $1.65 billion Anyscale deal is subject to regulatory approvals and to Nscale completing its IPO.
FOR COREWEAVE Its 10-Q says leases not yet commenced will start between 2026 and 2029 with terms of seven to sixteen years.
FOR CUSTOMERS Nscale’s announcement says Anyscale will keep its brand and customers may keep running it on other infrastructure.
The analysis says: “The job of any operator, investor, or builder is to know which layer they are operating inside and which layer the binding constraint is rotating toward next.”
As we read it, Nscale and CoreWeave are answering that question from opposite ends, one from land and power, the other from GPUs and software. The analysis does not discuss either company.
What Is Not Established
Nscale’s capacity figures include contracted capacity that is not yet built; its S-1 counts about 25,000 active GPUs. CoreWeave’s Q3 10-Q was not yet filed when we read its filings, so its figures are as of 30 June 2026. The 73% figure was highlighted on X by Jessie Dong; the filing is our source.
This Week In The AI Supercycle
The Great Convergence of Capital, Power, and Geopolitics
This week’s issue: for two years the AI build-out was priced as a supply story, and last week as a credit story. This week it was priced as a permission story, and the answer arrived as a cost, not a ban.
Nscale’s S-1 puts about 73% of its active and contracted capacity at sites it owns, seven of its 17 sites, and has agreed a $1.65 billion all-stock Anyscale deal tied to its IPO; CoreWeave reports $16.3 billion of operating lease liabilities plus $35.5 billion of leases yet to start, does not disclose an owned-capacity share, and paid for Weights & Biases mostly in stock.
A note on sourcing. Nscale figures are from its S-1 registration statement filed with the SEC on 18 September 2026 and its July 2026 press release; CoreWeave figures are from its 10-Q for the quarter ended 30 June 2026. Percentages and ratios are our arithmetic. This is not investment advice.
Gennaro Cuofano is a CRO and tech executive who has worked in AI since late 2015, starting with bringing NLP, natural-language generation, voice and chatbot products to the marketing industry. His background is in law and finance: he holds a Master's degree in Law and an International MBA with an emphasis on corporate finance (LUISS Business School and the University of San Diego, 2012), and worked as a financial analyst at a real-estate investment firm in San Diego and as an assistant controller. His work focuses on business model strategy, business engineering and, more broadly, structural analysis: how companies make money, read from their own filings. He created FourWeekMBA and leads research there and at The Business Engineer, his newsletter on AI and business strategy, with over 95,000 subscribers and more than 1,000 published analyses. His writing has also appeared on Entrepreneur (for example, How to Digitize a Family Business in Europe) and HackerNoon. Find him on LinkedIn and Substack. See how we source.
