Nvidia's $500B GPU financing plan faces dark GPU oversupply warning

Nvidia's plan to convert GPU capacity into a $500 billion financeable asset class faces a new risk warning, with White House advisor David Sacks flagging "dark GPU" oversupply as the sector's biggest threat.

"The biggest risk is not on the demand side — it's compute oversupply and overbuilding," Sacks, a member of the Presidential Science and Technology Advisory Council, said on the All In Podcast. "Just like dark fiber after the internet bubble, if dark GPUs emerge, it will be a disaster for everyone."

Sacks cited Elon Musk's plan to add 6-8 gigawatts of compute capacity next year, requiring $300 billion to $400 billion in capital expenditure against just $100 billion raised through equity and debt. Nvidia signed memorandums of understanding with Apollo Global, BlackRock, Blackstone, Brookfield, Goldman Sachs and KKR to mobilize over $500 billion in third-party capital for AI infrastructure, with Nvidia offering residual value support of up to 25 percent on individual projects.

The financing structure — securitizing AI compute revenue into tradable instruments — could unlock massive downstream AI capex but concentrates systemic risk in GPU-backed collateral. Nvidia shares, trading at 34x forward earnings with a consensus price target of $305.94, have yet to reclaim all-time highs since the announcement.

Dark Fiber Parallel: $30-$50 Per Watt vs $20-$25M Per MW

Sacks drew a direct comparison to the early 2000s telecom bust, when carriers laid fiber optic cable far ahead of demand, leaving vast networks idle and prices collapsed. He argued that GPU infrastructure faces the same risk if too many players race to build compute capacity simultaneously.

The warning carries weight because it comes from inside the Trump administration's science advisory circle — the first direct risk assessment from that orbit on the AI financing boom. Sacks also noted that political resistance to data center construction may paradoxically protect the market from oversupply, since building hurdles slow capacity expansion.

Nebius, the AI infrastructure provider, disclosed that its multi-year cloud service agreements carry annual contract values of roughly $20 million to $25 million per megawatt. That figure suggests Musk's estimate of $30 to $50 per watt for AI compute value reflects short-term contract premiums rather than industry-standard pricing, potentially inflating revenue expectations across the sector.

GPU-Backed Loans Carry 9%-18% Rates

The financing mechanism works by packaging future AI computing service revenues into special purpose vehicles, similar to mortgage-backed securities, and selling them to institutional investors. Nvidia's CUDA software stack provides a moat by enabling older GPUs to shift from training to inference workloads, extending their revenue-generating life. This matters as AMD's Helios rack systems target inference workloads, where Nvidia faces its most direct competitive pressure.

GPU loan rates across the industry currently range from 9 percent to 18 percent, according to Ilya Kasyan, chief technology officer at American Compute, a firm providing GPU residual-value insurance. Repayment periods typically span three to five years with borrower equity of 20 percent to 40 percent. A 5-percentage-point rate increase on a $10 million GPU purchase adds roughly $540,000 in costs over three years.

Bloomberg analyst Simon White drew a parallel between Nvidia's 25 percent residual value guarantee and AIG's credit default swap exposure before the 2008 financial crisis. While Nvidia's risk profile differs — the chipmaker uses product cycle knowledge to offset depreciation risk — the higher credit ratings enabled by the guarantee open the door for conservative lenders like pension funds to participate.

CoreWeave, the neocloud operator in which Nvidia holds roughly 10 percent equity, warned in a regulatory filing that reducing reliance on Nvidia GPUs would require significant time and capital. The company reported a $104 billion backlog and its stock surged 19 percent to $107.73 on Wednesday. Meanwhile, average GPU utilization stands at just 5 percent, according to an April report from CAST AI, raising questions about whether the supply shortage narrative matches field reality.

For investors, the near-term question is whether Nvidia discloses specific terms on the residual value mechanism and actual committed amounts from participating financial institutions. Hyperscalers including Alphabet, Amazon, Meta, Microsoft and Oracle are expected to spend approximately $800 billion on capex this year, rising above $1 trillion by 2027. If the "dark GPU" scenario materializes, the fallout would extend beyond Nvidia to the entire AI infrastructure debt chain — including the very financial institutions now packaging these securities.

This article is for informational purposes only and does not constitute investment advice.