NVIDIA's ambitious plan to package its graphics processing units (GPUs) into financial assets is now facing a risk warning from within the White House.
David Sacks, a member of President Trump's Council of Advisers on Science and Technology, warned on the "All In Podcast" that the biggest threat to NVIDIA's initiative—which partners with investment firms to turn GPUs into securitized, yield-generating assets—is not a lack of demand, but an oversupply of computing power. He cautioned that a "dark GPU" scenario could create a systemic shock to the entire AI infrastructure investment chain.
Sacks' remarks represent the most direct risk alert from within the Trump administration's inner circle regarding this AI financing frenzy.
What is the "Dark GPU" Risk?
Sacks drew a parallel to the "dark fiber" era of the internet bubble to explain his concerns.
In the early 2000s, telecom companies laid massive amounts of fiber optic cable, but demand fell far short of expectations, leaving vast amounts of cable unused and causing prices to collapse. This became known as the "dark fiber" crisis. Sacks believes the GPU market faces a similar risk.
He stated directly on the show: "To me, the biggest risk is not on the demand side. The biggest risk is an oversupply of computing power, overbuilding. Just like dark fiber emerged after the internet bubble burst, if we get dark GPUs, that would be a disaster for everyone—especially those building computing infrastructure expecting spot prices of $30 to $50 per watt."
The reference to "$30 to $50 per watt" points to Elon Musk's public estimate of AI computing power's value. According to reports, Musk stated during an internal SpaceX conference call that AI computing power is worth roughly $30 to $50 per watt, estimating that providing 1 gigawatt of power could generate $300 billion to $500 billion in revenue by the end of 2027.
However, computing infrastructure provider Nebius later disclosed that the annual contract value for its multi-year cloud service agreements is approximately $20 million to $25 million per megawatt. This figure suggests that Musk's expected price corresponds to a premium for short-term contracts, rather than the industry's standard pricing.
Political Resistance as a "Natural Insurance"
Sacks did not stop at the warning. He also offered a counterintuitive insight: the extensive political resistance currently facing data center construction may actually protect the market from oversupply.
He said: "If suddenly too many people rush to supply computing power, creating an oversupply and a market collapse, that's the real risk. But oddly enough, all this political resistance acts as an insurance policy against that outcome. Building data centers is just too hard—for all the reasons we've discussed, and now there's a moral panic, hysteria, even a circus going on. It's these political barriers that I think almost guarantee we won't see an oversupply relative to exponential demand growth. So, in a strange way, you're actually protected."
In other words, Sacks' logic is that the higher the barriers to construction, the harder it is for supply to expand rapidly, thereby reducing the risk of oversupply.
The Core Logic of NVIDIA's Financing Plan
On the show, Sacks also explained the fundamental driver behind NVIDIA's financing plan: the capital expenditure gap among downstream buyers has grown so large that it cannot be covered by equity and debt financing alone.
Using Musk's expansion plans as an example, he said: "Musk plans to add roughly 6 to 8 gigawatts of computing power next year, which we know requires $300 billion to $400 billion in capital expenditure. His company just raised $100 billion through equity and debt. Obviously, they need other ways to finance the rest."
Sacks believes the most direct solution is to obtain seller financing from NVIDIA. He said: "Now, Jensen (Huang) is creating a credit line, leveraged by these large banks and big private equity firms, to make that credit line available to all downstream buyers. This will benefit them."
The core mechanism of NVIDIA's plan is to transform GPUs into a financial asset class, similar to securities, with NVIDIA providing residual value support, thereby making GPUs financeable and yield-generating. Sacks characterized this as a crucial step in alleviating the current financing bottleneck in AI infrastructure construction.
Comments