Nscale filed for a U.S. initial public offering on September 18 with one number that makes the AI-cloud boom look radioactive: the company lost $1.02 billion in the first six months of 2026 on just $140.6 million of revenue. Yet NVIDIA Corporation (NASDAQ:NVDA) is committing $1 billion to a $3.1 billion convertible financing, while Microsoft Corporation (NASDAQ:MSFT) is expected to become one of Nscale’s most important customers. The IPO therefore gives investors an unusually clean test of who captures the economics when AI demand is enormous but the infrastructure needed to serve it is even more expensive.
We recently asked whether the trillion-dollar AI capex gap could become Nvidia’s trap and Microsoft’s opportunity, and examined whether IREN’s Microsoft relationship and $14 billion of funding are enough to overcome execution risk. Nscale puts both questions inside one IPO—and the answer depends on who gets paid before the infrastructure earns an adequate return.
The contract book is huge, but so is the financing problem
As of August 31, Nscale had $103.4 billion of active and contracted total contract value under long-term take-or-pay contracts, with a weighted-average contract life of about 5.7 years. It operates in 14 regions and has more than 10 gigawatts of potential power capacity. That is the bull case for NVIDIA Corporation: each new Nscale campus can become another large deployment of Nvidia accelerators and networking, and Nvidia gets paid for hardware before Nscale proves it can earn an attractive return on the buildings. The catch is customer concentration. Nscale disclosed that one customer generated 52% of first-half revenue, and its future plan leans heavily on Microsoft and Anthropic. If financing tightens or deployments slip, Nscale absorbs the project risk while Nvidia’s exposure is partly insulated by selling the scarce compute.
Microsoft Corporation gets the opposite trade. Nscale can give Microsoft access to external AI capacity without Microsoft owning every site itself, useful while Azure demand remains constrained by power and data-center availability. But outsourcing capacity does not make it cheap. Long-dated take-or-pay style commitments can lock Microsoft into infrastructure whose economics depend on utilization, GPU pricing and the durability of AI demand.
The smart-money split is telling
Insider Monkey’s database counted 285 hedge funds with reportable Nvidia longs at the end of Q2 2026, up from 275 in Q1. Fisher Asset Management held 90.9 million shares after increasing its position about 3%. Microsoft moved the other way: 273 funds held the stock in Q2, down from 282 in Q1, while Arrowstreet Capital increased its stake about 14%. Those filings predate Nscale’s September IPO filing. Microsoft also had about 74.45 million shares sold short as of August 31, roughly 1.0% of float with 3.18 days to cover, which is not a crowded bearish setup.
The IPO’s most important question is not whether AI demand exists. It is whether a company can borrow, build and fill capacity fast enough for revenue to catch depreciation and financing costs. Nvidia can still benefit from Nscale’s expansion by selling the picks and shovels, although its $1 billion financing commitment gives it direct capital exposure as well. Microsoft can win if external capacity relieves a bottleneck at acceptable economics. Nscale has to make both sides work at once.
