Majestic Labs cofounder Sha Rabii knows what it’s like to build AI infrastructure in an industry dominated by Nvidia.
Rabii spent years leading custom silicon inside Google and Meta before launching his own AI hardware startup.
When Rabii considers Nvidia’s biggest competitive advantage, he doesn’t point to its brilliant engineers, leading technology, or well-established sales channels.
“Where they’re kind of unmatched is in the giant bag of cash they’re sitting on top of,” Rabii said.
Nvidia had more than $80 billion in cash and marketable investments as of April, when it last reported earnings, and generated another $50 billion in operating cash flow that quarter. While other Big Tech giants have comparable stashes, Nvidia doesn’t shoulder the same massive data-center construction costs.
Even after returning billions to shareholders, it leaves it with an enviable problem: how to put its riches to work.
“What better use of the rest of the cash to invest and grow and strengthen the ecosystem around their products?” said Bernstein analyst Stacy Rasgon.
On Monday, Nvidia unveiled a new way to turbo-charge the power of its cash hoard by bringing in at least $500 billion of outside capital to finance far more Nvidia-powered AI infrastructure than what it’s already funding. This new initiative includes some of the biggest firms on Wall Street, including Apollo, Blackstone, and Goldman Sachs.
The company is also in talks to guarantee an OpenAI data center project worth hundreds of billions of dollars. It’s providing financial backstops to neoclouds, or specialized AI cloud providers, to buy its graphics processing units (GPUs) — the chips used to run AI.
It’s also making long-term commitments to secure scarce components, and investing in and supporting AI startups through its NVentures and Inception programs. If one bet takes off, Nvidia hits the lottery on both the investment and future chip sales.
The result is a self-reinforcing cycle: Nvidia’s technological dominance generates cash, and that cash is increasingly becoming a moat of its own.
Nvidia’s scale has boosted its financing playbook
Vendors using their balance sheets to help customers buy their products isn’t a novel strategy, said Bernie Margulies, who works on GPU financing at American Compute. The practice is spreading more broadly.
AMD has previously backstopped chips for customers, while Broadcom recently partnered with Apollo and Blackstone on an AI-infrastructure financing platform. Neither rival can match Nvidia’s scale, said D.A. Davidson analyst Gil Luria, calling the chipmaker’s financial might a different “order of magnitude.”
In addition, Meta agreed to purchase up to $15 billion of computing capacity from the neocloud Nebius, while Google has backed Anthropic’s data-center projects.
Historically, Nvidia hasn’t needed to do this, Margulies said.
As the AI buildout explodes, Nvidia helps finance projects that might not otherwise get built. Margulies said this could also be read as an effort to keep customers from experimenting with alternatives, like AI chips built by AMD or Google.
Alex Yeh, CEO of the neocloud GMI Cloud — which has a backstop agreement with Nvidia — said the deals help the chip giant build an “alliance” with neoclouds and AI startups as its biggest cloud customers develop their own chips.
They can also unlock massive orders for AI chips.
Yeh said that without Nvidia stepping in, he wouldn’t have been able to take on the AI startup Fireworks as a customer, with which GMI has a nine-figure deal.
Likewise, Rabii said financing is a complication for one of Majestic’s target markets: neoclouds. If a customer wanted to buy billions of dollars of Majestic hardware, his startup couldn’t backstop the financing the way Nvidia can.
“Nvidia’s real moat today is how freaking expensive it is to build data centers,” Rabii said. Instead, he said he’s betting Majestic’s power-efficient chips will convince customers to choose its systems.
Nvidia is looking beyond its own balance sheet
Nvidia’s growing financial ties have fueled concerns about circular financing.
Margulies said he isn’t particularly concerned, though the strategy “does cause a lot of investors to be worried.” If Nvidia loses market share and has to pay out on its backstops, “it’ll be brutal,” he said.
Michael Reid, CEO of Australian neocloud Megaport, which does not have a backstop agreement with Nvidia, said overall demand for GPUs provides another safeguard. If one customer falters, the neocloud could rent the capacity to another.
Nvidia CEO Jensen Huang said on X that the company’s latest Wall Street partnerships are designed to address concerns about circular financing. Analysts said that tapping outside capital means Nvidia doesn’t have to shoulder as much of the financial risk.
“They’re sharing the reward, but they’re also sharing the risk,” said D.A. Davidson’s Luria.
In this vein, the chipmaker’s next financial moat may be access to money that isn’t even its own.
Have a tip? Contact this reporter via email at [email protected] or Signal at @geoffweiss.25. Use a personal email address, a nonwork WiFi network, and a nonwork device; here’s our guide to sharing information securely.
