A sharp selloff in global technology shares has put fresh pressure on the financing model behind the AI infrastructure boom, with more than $1 trillion erased from the combined market value of the world’s 20 largest tech companies this week. Nvidia chief executive Jensen Huang responded by arguing that demand for the company’s computing systems is far broader than a handful of heavily funded AI model developers, even as credit markets signaled growing concern over debt-fueled cloud expansion.
The tension was most visible around CoreWeave, the cloud computing provider that has rapidly added Nvidia-powered data-center capacity through large borrowing programs. CoreWeave’s shares rose 5.94% after Huang said in Stockholm that the company did not require financial backstop support from Nvidia and that its cost of capital was falling. The rebound followed a two-session decline of more than 18%.
Credit trading told a less reassuring story. ICE Data Services showed CoreWeave’s five-year credit default swap spread surged by 74.5 basis points intraday to a record 1,194.5 basis points during the selloff before moving sharply again. At that level, pricing implied a five-year default probability of more than 60%, according to standard CDS-implied calculations.
CoreWeave’s position has attracted unusual scrutiny because Nvidia is both a critical supplier and a major shareholder. Nvidia owns about 10% of the company, making it CoreWeave’s second-largest shareholder, while CoreWeave uses borrowed capital to buy the GPUs that power its cloud offering. That arrangement leaves the company exposed if financing costs rise or customers take longer than expected to turn rented computing capacity into recurring revenue.
Huang argues demand extends beyond frontier labs
In a roughly 140-minute BG2 podcast interview released on Oct. 15 before the US market opened, Huang pushed back against the idea that Nvidia’s business rests mainly on the spending of two frontier AI labs.
He described Nvidia’s demand base as three roughly equal segments. Frontier model developers account for roughly one-quarter to one-third of demand, he said, and include five or six major organizations rather than only OpenAI and Anthropic. Huang named Meta, Grok, Gemini and Anthropic among the large AI efforts buying advanced computing capacity.
Huang estimated that OpenAI and Anthropic each operate about 2 to 3 gigawatts of compute capacity today. That is an enormous amount of power for individual technology companies, though it also illustrates why cloud providers and developers increasingly face questions over whether their revenue can keep pace with infrastructure commitments.
A second third of demand comes from more established computing workloads, Huang said. He included molecular dynamics, physics simulations, quantum chemistry, conventional machine-learning applications such as fraud detection at Visa and Mastercard, and recommendation systems. According to Huang, that category has been growing by 30% to 40% annually.
The remaining third comes from thousands of startups that rent computing capacity, he said. About 80% of those companies use open-source AI models, according to Huang, with applications spanning physical AI, biology and materials science. Nvidia’s argument is that a more distributed customer base could cushion the company against a slowdown at any individual frontier-model developer.
That distinction matters for Nvidia’s own revenue outlook, but it does not remove the funding challenge facing companies that are building new data centers. Hardware suppliers receive payment as systems are delivered, while cloud operators must recover their spending over years through leasing contracts, software services and customer usage.
Debt and power are becoming the limiting factors
The AI buildout increasingly depends on two resources that cannot be expanded as quickly as demand for chips: financing and electricity.
Huang said the primary constraints by 2027 would be land, power, permitting and construction labor. He agreed with a projection discussed during the podcast that 43 gigawatts of AI compute capacity could be added next year, followed by another 40 gigawatts the following year. Such additions would exceed the current installed base of some national electricity systems.
The figures place grid access at the center of the competition between data-center operators. US server facilities are expected to require 76 gigawatts of power this year, while machine-learning-focused sites globally could consume roughly 175 terawatt-hours, according to estimates cited in the supplied material. Data-center developers must secure long-term electricity contracts, transmission connections and construction approvals before the facilities can host revenue-generating hardware.
Huang said tighter power availability favors more efficient systems, arguing that Nvidia leads the market in tokens produced per watt of electricity. A token is a unit of text or data processed by an AI model; improving tokens-per-watt output can lower the electricity required to serve a given amount of demand.
He also said the economic value associated with each gigawatt of Nvidia computing has risen from roughly $18 billion to more than $40 billion. The estimate reflects the growing value placed on AI services running on the systems, though it depends on cloud operators and their customers sustaining high usage rates.
Nvidia sets out a far larger spending forecast
Huang framed the current buildout as a replacement cycle as well as an expansion of computing capacity. He said annual global spending on compute infrastructure before the AI boom was about $500 billion, and argued that the slowing of Moore’s Law has made it necessary to build more specialized, accelerated systems.
By the end of the decade, Huang forecast annual infrastructure construction and refresh spending of $3 trillion to $4 trillion, including about $1.5 trillion for pure compute. Over a longer period, he said he would not be surprised if annual build-and-refresh spending reached $10 trillion.
Those projections are far above current infrastructure outlays and would require a steady supply of corporate borrowing, private credit and customer contracts. The recent movement in CoreWeave’s CDS spread shows that lenders are already drawing sharper distinctions between companies selling AI hardware and companies financing large fleets of that hardware.
Nvidia is also seeking revenue beyond initial chip sales. Huang said the company earns annual enterprise software licensing fees per GPU and receives revenue shares from some cloud providers. Those arrangements could give Nvidia recurring income tied to its installed base, while cloud operators carry much of the capital intensity of building and operating data centers.
Huang said Nvidia now represents about 40% of capital spending by cloud service providers and about 80% outside that group when on-premises deployments and newer cloud builders are included. Six years ago, Nvidia’s share of global capital spending was in the single digits, he said. The company has used the cash generated by that expansion to repurchase shares, with Huang saying more than 50% of free cash flow goes to buybacks and about 60% did so in the most recent quarter.
For cryptocurrency markets, the immediate connection is less about AI demand itself than liquidity conditions. A sustained deterioration in technology credit can force traders to reduce risk across growth assets, including digital tokens. The more direct exposure remains with proof-of-work mining operations and data-center projects competing for electricity, financing and specialized hardware as AI infrastructure spending accelerates.
Worried about AI-fueled volatility? Explore how Web3, AI and crypto intersect to understand long-term digital infrastructure trends.
Disclaimer: The content on this page is provided for general informational purposes only and does not represent the views or financial advice of Toobit. We make no guarantees regarding the accuracy or completeness of this information and shall not be held liable for any errors, omissions, or outcomes resulting from its use. Investing in digital assets involves risk; users should independently evaluate their financial situation and the risks involved. For further details, please consult our Terms of Service and Risk Disclosure.
