Nvidia's Huang defends CoreWeave as AI debt jitters hit markets

Today’s Focus

Shares of CoreWeave, the AI cloud provider that leases Nvidia-powered computing to clients like Microsoft and OpenAI, slid sharply this week as investors questioned whether the company can carry its growing pile of debt. Reuters reported CoreWeave stock and bonds were both headed for steep weekly losses, with its credit default swaps widening to levels CNBC described as “distressed.”

The selloff prompted Nvidia Chief Executive Jensen Huang to publicly defend the company. Speaking to Yahoo Finance, Huang said he remained “optimistic” about CoreWeave and about demand for AI compute broadly. Nvidia both supplies CoreWeave’s chips and holds an equity stake, a circular arrangement that has drawn scrutiny from analysts.

CoreWeave went public in March 2025 and has borrowed billions against long-term contracts with hyperscale customers. Axios reported the company’s debt stack now exceeds $11 billion, much of it secured by Nvidia GPUs whose future resale value is uncertain as new chip generations arrive.

The anxiety extends beyond one company. Reuters noted that lenders financing specialized “neocloud” firms are reassessing risk as the sector’s capital needs balloon. CNBC reported that CoreWeave’s five-year credit default swap spreads moved past 1,000 basis points this week, a threshold bond traders typically associate with elevated default risk.

CoreWeave has not revised its revenue guidance, and Huang pointed to the firm’s long-term contracts with blue-chip tech customers as evidence the business remains sound. But the price action has reignited a broader debate about whether the AI buildout is a durable investment cycle or a leveraged bet that could unwind if demand slows.

The Debate

Supporters argue

Huang told Yahoo Finance that AI compute demand continues to outstrip supply and that CoreWeave’s contracted revenue gives it a cushion other startups lack. “I’m optimistic,” he said, pointing to the company’s customer book.

Wall Street bulls make a similar case. A Yahoo Finance column argued investors nervous about CoreWeave can still own Nvidia, whose chip orders are backed by diversified hyperscalers including Microsoft, Google, Amazon and Meta. The piece framed CoreWeave’s wobble as a single-name credit story rather than a verdict on AI itself.

Dan Ives of Wedbush Securities, quoted across financial outlets this year, has repeatedly described the AI capital cycle as a multi-year expansion with trillions in incremental spending ahead. Supporters note CoreWeave’s largest customers have signed multi-year take-or-pay deals, meaning revenue is largely locked in even if spot demand softens.

Industry executives also point to utilization. Microsoft and OpenAI have both said publicly in 2025 earnings calls that they are compute-constrained, not demand-constrained, which supporters argue means CoreWeave’s capacity will stay rented.

Critics argue

Skeptics contend the financing structure is the problem, not the demand. Axios reported that much of CoreWeave’s borrowing is collateralized by Nvidia chips, assets that depreciate quickly as newer architectures ship. If resale values fall faster than loans amortize, lenders are exposed.

Short seller Jim Chanos, speaking on CNBC earlier this year, called the circular flow of capital among Nvidia, its investees, and their customers a “self-funding loop” that resembles vendor financing bubbles of past tech cycles. Critics argue that when Nvidia takes equity in a buyer that then uses the proceeds to buy more Nvidia chips, revenue growth can look stronger than underlying end-demand warrants.

Reuters quoted fixed-income analysts noting that the widening in CoreWeave’s credit default swaps reflects real worry among debt holders, not just equity traders. The Financial Times has reported that private credit funds, which have supplied a large share of neocloud debt, are tightening terms on new AI infrastructure loans.

Senate Banking Committee members including Elizabeth Warren (D-MA) have asked regulators to examine concentration risk in AI lending, arguing that a shock to one hyperscale customer could cascade through the sector.

What the experts say

Independent researchers caution that the AI infrastructure boom is historically large and historically fast, which makes outcomes hard to model. A September 2025 Bain & Company report estimated global AI compute spending will need to reach roughly $500 billion annually by 2030 to meet projected demand, a figure Bain itself called “stretching.”

The Bank for International Settlements, in its September 2025 Quarterly Review, flagged rising leverage in data-center financing as an area to watch, noting that collateral values for specialized hardware are difficult to mark to market.

Erik Brynjolfsson of the Stanford Institute for Human-Centered AI has argued that AI productivity gains are real but typically arrive with a lag, meaning near-term revenues may underperform the capital invested to produce them. That dynamic, documented in his research on prior general-purpose technologies, can leave infrastructure providers overbuilt in the interim.

Historical comparisons from the Federal Reserve Bank of San Francisco show the late-1990s telecom buildout saw similar collateral-backed lending against fiber assets whose values collapsed when demand lagged capacity.

By the Numbers

$11 billion: approximate total debt on CoreWeave’s balance sheet, according to Axios.

1,000+ basis points: level CoreWeave’s five-year credit default swaps reached this week, a threshold CNBC associated with distress.

March 2025: when CoreWeave completed its initial public offering.

$500 billion: annual global AI compute spending Bain & Company estimated will be needed by 2030 to meet projected demand.

$4 trillion: Nvidia’s approximate market capitalization in October 2025, per Reuters market data.

4: number of hyperscale customers (Microsoft, Google, Amazon, Meta) that account for the bulk of Nvidia’s data-center revenue, per the company’s filings.