Nvidia's $500 Billion AI Funding Plan Conceals Risks, Trump Advisor Warns of "Dark GPU" Surplus

Nvidia's grand plan to package GPUs into financial assets is facing risk warnings from within the White House.

Recently, David Sacks, a member of President Trump's Council of Advisors on Science and Technology, stated on the "All In Podcast" that Nvidia’s strategy to cooperate with investment institutions and turn GPUs into financeable yield-generating assets faces its biggest threat not from a lack of demand but from an oversupply of computing power—once a scenario of "dark GPUs" emerges, it would create a systemic shock to the entire AI infrastructure investment chain.

Sacks’ remarks represent the most direct risk warning to date from within Trump’s administration circle about this AI financing frenzy.

What is the “dark GPU” risk?

Sacks drew a parallel with the “dark fiber” of the internet bubble era to explain the scenario he’s concerned about.

In the early 2000s, telecom companies massively laid fiber-optic cables, but actual demand drastically underperformed expectations. A large volume of fiber was left unused and prices collapsed, a crisis known as “dark fiber.” Sacks believes the GPU market faces a similar risk.

On the show, he said directly: "For me, the biggest risk is not on the demand side; the biggest risk is an oversupply of computing power, resulting from overbuilding. Just like after the dot-com bubble burst and dark fiber appeared, if we see dark GPUs, it will be a disaster for everyone—especially for those who expected spot prices of $30 to $50 per watt when building computing infrastructure."

Here, "$30 to $50 per watt" refers to Elon Musk’s previously public estimates regarding the value of AI compute. According to reports, Musk stated during an internal SpaceX meeting that the value for AI compute is approximately $30 to $50 per watt. Based on this, he estimated that providing 1 gigawatt of compute could generate $300 to $500 billion in revenue by the end of 2027.

However, compute infrastructure provider Nebius later disclosed that the annual contract value for its multi-year cloud service agreements is about $20 to $25 million per megawatt. This figure indicates that Musk’s expectations actually reflect a premium price for short-term deals rather than common industry pricing.

Political resistance as “natural insurance”

Sacks didn’t stop at issuing a risk warning. He also made a counterintuitive point: it’s precisely the heavy political resistance faced by data center construction that could end up protecting the market from oversupply.

He said: "If suddenly too many people rush in to supply compute and an oversupply causes the market to collapse, that’s the real risk. But strangely, all this political resistance actually acts as insurance against such an outcome. Building data centers is just too difficult—we’ve discussed all sorts of reasons, and now there’s a moral panic, hysteria, even what you could call a farce unfolding. It’s precisely this political resistance, I believe, that almost guarantees, relative to the exponential growth in demand, that there won’t be an oversupply. So from a strange perspective, you’re actually protected."

In other words, Sacks’ logic is: the higher the barriers to building, the harder it is for supply to expand rapidly, and the lower the risk of oversupply.

The core logic behind Nvidia’s financing plan

Sacks also explained on the show the fundamental driver behind Nvidia’s financing plan: the capital expenditure gap for downstream buyers has grown so large it can no longer be covered by just equity and debt financing.

He took Musk’s expansion plan as an example: "Musk plans to add around 6 to 8 gigawatts of compute next year, which, as we know, requires $300 to $400 billion in capital expenditure. Yet his company has just raised $100 billion via equity and debt. Obviously, they need to raise funds through other means."

Sacks believes the most straightforward solution is simply to apply to Nvidia for vendor financing. He said: "Now, Jensen (Huang) is creating a credit line with the help of these large banks and major private equity institutions, opening this credit line to all downstream buyers, which will benefit them."

The core mechanism of Nvidia’s current plan is to turn GPUs into an asset class similar to financial securities, with Nvidia providing residual value support, thereby making GPUs financeable and income-generating. Sacks characterized this as a key move to help alleviate the current AI infrastructure financing bottleneck.