Nvidia lost approximately $365 billion in market value last week as rising borrowing costs cast doubt on the sustainability of Wall Street’s artificial intelligence spending boom.
The chipmaker’s shares fell 6.6% over the five sessions ending 16 October, closing at $214.13. The Philadelphia Semiconductor Index dropped 11.4%, underscoring mounting concern across the industry.
Demand for Nvidia’s processors remains strong. The bigger worry is whether companies building the data centres that house those chips can continue financing their ambitions.
CoreWeave exposes Nvidia’s financing risk
CoreWeave, the Nvidia-backed cloud provider, lost nearly a quarter of its market value last week as its borrowing costs surged.
Yields on one of its bonds climbed above 15%, up from around 9.6%, while investors demanded substantially more protection against default.
Paul Chew, head of research at Phillip Securities, estimates CoreWeave has roughly $30 billion in net debt and another $36 billion in lease commitments, supported by a $104 billion revenue backlog.
A four-percentage-point increase in financing costs could add almost $500 million to annual interest expenses, Chew warned in a research note published Monday.
“Nvidia cannot afford CoreWeave to fail,” he wrote.
Nvidia owns approximately 11% of CoreWeave, which buys its processors and helps broaden chip demand beyond technology giants.
However, the risks remain prospective, as CoreWeave’s contracted backlog offers some protection, and Nvidia chief executive Jensen Huang has defended the company’s financial strength.
AI demand survives but funding gets tougher
Building data centres requires enormous upfront expenditure on processors, energy and cooling systems. Revenue often arrives years later, leaving operators exposed when borrowing becomes more expensive.
My Bui, an economist at AMP told Reuters on Monday that rising government bond yields and wider corporate credit spreads were increasing financing costs, complicating borrowing and refinancing.
She also questioned whether future AI revenues would recover construction costs before equipment depreciates.
The strain extends beyond individual borrowers. The Financial Times reported on Sunday that at least 47 major US data-centre projects have been stalled or cancelled this year, reflecting financing, power and planning constraints.
Yet the pressure is uneven. It is concentrated among heavily leveraged operators. Microsoft, Alphabet and Meta have much stronger financial resources than smaller providers dependent on debt markets.
That distinction matters for Nvidia, as its customers may still want more computing capacity, even if some lack affordable capital to build it immediately.
Meta earnings offer Nvidia’s next test
Investors will look to Meta’s earnings on Wednesday for reassurance that spending by major customers remains intact.
Meta expects capital expenditure of $130 billion to $145 billion this year. Goldman Sachs estimates that five major AI hyperscalers could spend approximately $1.2 trillion combined in 2027.
Those commitments support Nvidia’s growth outlook, but greater spending is not automatically good news if profits and cash generation deteriorate.
Technology investor Michael Parekh wrote in his AI: Reset to Zero newsletter on Monday: “This is a financing cycle more than a demand cycle.”
He expects financing strains to delay construction and drive consolidation rather than eliminate demand for computing power.