Tech stock sell-off reopens debate on who will foot the AI bill

The recent sell-off exposed a financing problem behind the AI boom. Investors are no longer asking only how quickly AI will grow. They are asking which companies can pay for the enormous data centers, processors, and expansion plans that demand increasingly costly funding.

CoreWeave became the clearest example. Its shares fell 36% in one month, while investors demanded a 16% return on its debt. The company has $50 billion in debt and plans to keep raising money. Semiconductor makers also lost 19% in 12 trading days, while Firmus and Nscale postponed IPO plans.

The concern is not that AI demand has disappeared. TSMC beat forecasts, and ASML reported record sales and strong future demand. Instead, restrictive monetary policy, 20-year-high debt yields, and expected debt issuance make financing harder. Investors are now demanding evidence that AI infrastructure can become profitable enough to justify its cost.

A neocloud is an AI-focused cloud company that buys processors, installs them in data centers, and sells the resulting computing capacity. Its business depends on building infrastructure first, then earning enough from customers to cover that investment and its financing costs.

CoreWeave illustrates the model. It had orders worth $104 billion and revenue of $2.6 billion at the end of the second quarter. Yet it also reported a $626 million net loss, including $640 million in interest payments. This shows how quickly expansion can create financial pressure when companies borrow heavily to purchase processors and build capacity.

The major cloud companies, or hyperscalers, include Amazon, Microsoft, and Alphabet. They have impressed investors with their profits and ability to invest. Neoclouds are newer and financially more fragile. Their growth may be strong, but they face greater dependence on outside financing and must prove that future revenue can support their infrastructure costs.

CoreWeave’s figures reveal the financial scale of the AI infrastructure race. The company had $50 billion in debt, $2.6 billion in revenue, and orders worth $104 billion at the close of the second quarter. Despite those orders, it reported a net loss of $626 million, up from $290 million the previous year.

Interest costs explain much of the strain. CoreWeave paid $640 million in interest on its debt. That amount exceeded its reported net loss, showing how borrowing can absorb revenue even when demand and orders appear strong. The company also plans to keep returning to financial markets to fund expansion.

These numbers do not prove that AI demand is weak. They show that infrastructure companies must spend heavily before receiving enough cash from customers. CoreWeave therefore represents the central investment question: whether future AI revenue will be large and reliable enough to repay debt and justify continued expansion.

Doubts spread because AI infrastructure companies are financially connected. Neoclouds borrow to buy processors and build data centers, while chipmakers and other suppliers depend on continued investment. If investors question one heavily indebted buyer’s ability to pay, they may reassess the whole expansion chain.

CoreWeave had $50 billion in debt, a 16% demanded return, and a $626 million net loss. Its shares fell 36% in a month. Semiconductor makers lost 19% in 12 trading days. Firmus and Nscale postponed IPO plans, while South Korea’s Elice lost 32% when it went public.

The reaction reflects a tougher market, not necessarily weaker AI demand. TSMC beat forecasts, and ASML reported record sales of €11.7 billion. But investors are no longer funding every AI project automatically. Companies now need to show that future revenue can cover construction, equipment, debt, and interest costs.

Neoclouds are considered fragile because their business requires large spending before revenue fully arrives. They buy processors, install them in data centers, and sell capacity. That model leaves them dependent on financing to keep expanding and on customers to generate enough income afterward.

CoreWeave shows the pressure clearly. It had $104 billion in orders and $2.6 billion in revenue, but still recorded a $626 million net loss. It paid $640 million in interest on $50 billion of debt. Its plans also require continued access to financial markets, increasing its exposure to changing investor confidence.

Hyperscalers operate from a stronger position in the article. Amazon, Microsoft, and Alphabet have produced surprising profits and shown substantial investment ability quarter after quarter. Neoclouds may benefit from strong AI demand, but they must prove that growth can become durable profitability rather than merely increase borrowing and financial risk.

High interest rates make AI expansion harder because data centers and processors require large upfront investments. Companies that borrow for those projects must pay more to raise money and more interest while waiting for customers to generate revenue. The result is greater pressure on profitability.

CoreWeave illustrates the mechanism. It had $50 billion in debt and paid $640 million in interest. Investors demanded a 16% return from the company’s debt, a level the article associates with growing solvency concerns. CoreWeave also plans to keep using markets to finance its expansion.

The broader market is especially difficult because central banks are maintaining restrictive policy to contain inflation. Debt yields are at 20-year highs, while expectations of more debt issuance are growing. AI demand remains strong, but expensive financing can slow projects, weaken valuations, and expose companies whose revenue may not cover their investment obligations.

Strong chip demand and weak company profitability can exist at the same time. Customers may urgently need AI processors, but companies supplying the infrastructure must first spend enormous sums on equipment, facilities, and financing. Revenue growth therefore does not automatically mean that the business earns enough after costs.

The article gives both sides. TSMC beat market forecasts and raised its outlook. ASML recorded €11.7 billion in sales and said expected demand for the next two years exceeds available capacity. Yet CoreWeave had $2.6 billion in revenue, a $626 million loss, and $640 million in interest payments.

This suggests the pressure is concentrated in highly leveraged neoclouds rather than the entire AI industry. Enrico Chinello of Julius Baer said financing risks affect specific areas of the ecosystem. Investors may continue funding strong demand, but increasingly selective financing will favor companies that can turn orders into profits and manage debt.