Why Nvidia Needs CoreWeave

AI cloud-computing player CoreWeave has been under pressure in the markets lately. If the company’s highly leveraged business truly blows up, though, Nvidia has strong reasons to come to its rescue.

Nvidia has never said that it will save CoreWeave, whose shares have lost about a third of their value in the past month. Its bonds have also fallen, pushing their yields well into the double digits.

But CoreWeave’s existence aligns with Nvidia’s interests so closely that it is hard to imagine it wouldn’t step in if the need arose.

One big reason is that despite the untold billions of dollars Nvidia has made during the AI boom, its base of AI chip customers has been somewhat narrow. About half of its data-center chip revenue in recent years has come from so-called hyperscalers—big tech companies like Amazon.com and Microsoft that operate sprawling computing operations.

Welcome though that revenue may be, that is a problem for Nvidia. Given the high cost of its chips and Nvidia’s healthy margins—its gross profit margins were about 75% in its latest quarterly report, an extremely high level in the chip world—those customers have always had financial incentives to look elsewhere.

They also have the financial and technical wherewithal to design their own AI chips and save money in doing so. All of the big tech companies and major AI labs now have AI-chip efforts, some further along than others.

Enter CoreWeave. For Nvidia, the company has long been an ideal customer: One that could both reduce its direct exposure to the hyperscalers and wouldn’t ever compete with it. In a securities filing ahead of CoreWeave’s IPO last year, the company explicitly said its partnership with Nvidia would “help to diversify their customer base.”

The benefits have flowed both ways. CoreWeave, which arose out of a cryptocurrency-mining operation, has leaned on Nvidia to grow. It has routinely gotten large allocations of Nvidia’s latest AI chips, and its financial support. Nvidia now owns about 10% of the company’s equity.

CoreWeave, in turn, hasn’t been very agnostic about whose AI chips it uses. While the company uses central processing units from suppliers that aren’t Nvidia, it hasn’t been eager to install AI processors made by Nvidia’s competitors.

For all of those reasons, CoreWeave is an important instrument for Nvidia, even if its debt-fueled business model occasionally causes investor trepidation.

Concerns about CoreWeave’s model have weighed on its shares, although it isn’t in any real trouble in the near term. CoreWeave makes money largely by signing up creditworthy tech companies to leases for AI computing power. It then uses those lease contracts and related computing assets to raise money and build out the infrastructure.

Those deals aren’t in much jeopardy. The main risk for investors is that CoreWeave could struggle to repay debt it has taken on to fund the general expansion of its operations. That has yet to become an issue, and as BNP Paribas analysts pointed out in a note on Monday, its parent-company debt largely matures after 2030, giving it time.

Still, CoreWeave’s future does look dimmer if prevailing interest rates keep going up, if appetite for AI-related debt keeps weakening and concerns about the AI boom’s sustainability turn into a dearth of funding options.

CoreWeave needs access to capital to keep growing. If that tap closes, though, Nvidia might be tempted to use its massive balance sheet to turn it back on.

Every kid in America is getting a single share of SpaceX, currently valued at around $146. Or so says Elon Musk. Musk said Monday he would make the donation next year, following a growing number of AI-boom leaders promising to share some of their immense wealth via so-called Trump Accounts. If Musk follows through with it, the donation would amount to about $10 billion in shares at today’s prices.

Amount of Anthropic’s shares the company is to commit to a public benefit fund that could be distributed to charitable organizations or directly to people whose livelihoods are displaced by AI.

Consumers are getting more comfortable using AI agents to help them make purchases. They’re not yet going all-in.

The vast majority of people in a recent Accenture survey said they’d use an AI agent to research potential purchases. A much smaller percentage said they’d be OK with an agent making purchases on their behalf autonomously.

The latter percentage is likely to increase over time, as users get comfortable with agents and more companies begin to offer agents that can spend money without hand-holding.

And that could be a tough environment for e-commerce players. Big online retailers have gotten good at directing consumer traffic to their sites through search-engine optimization and inserting themselves into social-media conversations.

It is far less clear how to adapt to a world where those companies market largely to machines and customer relationships are less direct.

In a product demo that looked more like a Hollywood stunt, Musk’s Tesla unveiled its long-awaited Roadster late last week. The car levitated, flipped and landed, showing off aerial capabilities that may or may not be advisable (or legal) in the real world. Tesla shares fell the day following the unveiling, suggesting investors didn’t see a huge new market for the flight-capable cars, which Musk says are going to roll off the production line in 2028.

WSJ AI & Business is a weekly look at AI’s transformation of the business world. This newsletter was curated and edited by Asa Fitch. Reach him at asa.fitch@wsj.com (if you’re reading this in your inbox, you can just hit reply). Got a tip for us? Here’s how to submit.