On Tuesday it became known that Nvidia invested $3 billion in a Meta share offering. This is just one link in a chain of investments, loans and guarantees that, according to some analysts, has turned the chipmaker into the "central bank" of the artificial intelligence (AI) boom.
These deals help sustain demand for Nvidia processors - the foundation of huge data centers around the world and the main component of the AI revolution. Thanks to demand, Nvidia has become the world's most valuable company by market capitalization, and its CEO Jensen Huang is a key figure in the world of AI.
However, skeptics on Wall Street are asking: is Nvidia helping to inflate a bubble around AI?
Nvidia holds stakes in leading AI developers, including OpenAI, Anthropic and Elon Musk's xAI (now part of SpaceX). It has also invested in so-called "neoclouds" - companies like CoreWeave and Nebius that rent out computing capacity on Nvidia chips. The geography is broad: in France - Mistral, in Britain - the self-driving car startup Wayve, in Asia - a partnership with South Korea's SK Group to build data centers.
In addition to shares, Nvidia provides financial guarantees, acting as a buyer of last resort. For example, it committed to buy up to $6.3 billion of unused CoreWeave computing capacity if other customers do not take it. According to US media, the company also has about $36 billion in cloud service guarantees and a guarantee of up to $105 billion related to a data center for OpenAI.
Why is Nvidia doing this? Experts say that with huge profits it has few alternatives. "They are essentially becoming a bank," believes Michael Cusumano, a professor at MIT Sloan School of Management. "Directing funds into infrastructure related to the purchase of Nvidia equipment is reasonable." According to him, building its own data centers would put Nvidia in competition with its largest customers and could trigger antitrust claims. Gil Luria, head of technology research at D.A. Davidson, notes: Nvidia finds areas in the AI ecosystem that are underinvested and fills the gaps. "It's unprecedented, but it seems to be working," he says. Huang defends the approach: cutting-edge AI labs need tens of billions of dollars to start, not millions, and investments give them more time to succeed.
Critics compare the deals to the dot-com era, when companies inflated growth by investing in startups that then bought their products. "There is nothing intrinsically wrong with Nvidia helping customers," says Paulo Carvao, a senior fellow at Harvard Kennedy School. "But the closer the ties, the harder it is to assess real sustainable demand." Luria emphasizes: not all demand is the same. "If Nvidia gives CoreWeave money to buy Nvidia chips, that is not the same as Microsoft buying Nvidia chips because it has an OpenAI customer that has an enterprise customer willing to pay," he explains.
The main concern is concentration. According to the annual report, two direct customers accounted for 22% and 14% of Nvidia's revenue in fiscal 2026. Carvao warns: the intertwining of investments increases the risk of contagion if a weaker company like CoreWeave runs into difficulties. "If AI demand disappoints, spending cuts could hit chipmakers, data center builders and their financiers at the same time," he says. In October, shares of several Nvidia-linked "neoclouds", including CoreWeave, Nebius and IREN, fell by at least 30% amid fears about tighter lending for AI projects. Carvao notes that the AI chip market is supply-constrained, and estimates that demand will grow for another 12-18 months before it begins to level off. Cusumano expects high demand for Nvidia chips for another three to five years.