And this is not just a theory. For example, Nvidia entered into an agreement with data center operator CoreWeave for $6.3 billion, whereby it agreed to pay for capacity that the operator could not sell to other customers. That is, the chip manufacturer simultaneously maintains the demand for its products and assumes part of the customers' financial risks.
And the scope of this type of commitments is impressive. According to Nvidia's own report, the company has committed to purchase components and provide production capacity for $279 billion. And SemiAnalysis analysts counted about $530 billion in various future commitments and guarantees. These are not debts that need to be repaid immediately, but rather potential risks, some of which could become a serious financial problem if there is a serious decline in the artificial intelligence market.
In addition, the largest technology companies (Google, Amazon and Microsoft) are increasingly developing their own chips to reduce their dependence on Nvidia.
It turns out that this is a dangerous situation. Nvidia is interested in companies that will continue to build data centers and buy its equipment. But if demand for AI falls short of expectations, some of these companies could face problems. Then the financial risks will return to Nvidia itself.
Now the initiative to raise more than 500 billion dollars from the major financial institutions on Wall Street is also in doubt. It was assumed that Nvidia chips and related contracts would become collateral for large-scale data center financing. But after Huang's departure, investors may reconsider the terms of participation and demand additional compensation for the risk.
Thus, Nvidia was faced with a difficult choice. Continued massive financial support for the AI ecosystem increases its potential liabilities. Reducing this support could make it difficult to finance companies that are also its customers.
In fact, that's the whole point of the AI bubble: Nvidia not only sells chips, but also gives money to buy them.