As AI Labs Burn Cash, the Real Question Is Which Suppliers Get Paid

Microsoft confirmed this week that it was pulling out of a planned artificial intelligence (AI) data center project in the United Kingdom. The facility, being developed by Nscale, was supposed to provide computing power for Microsoft's Azure cloud business.

The problem? The electricity needed to run the data center wouldn't be available until the 2030s.

Microsoft is maintaining its other agreements with Nscale, but the cancellation raises an important question about the enormous amounts of money being committed to AI infrastructure. What happens when a project doesn't work out?

Companies make large commitments to build AI infrastructure. But there's a difference between selling the equipment needed to build an AI data center and owning that infrastructure while waiting for customers to pay for it. And that difference should matter to investors.

Image source: Nvidia.

Three companies that get paid for hardware

Nvidia (NASDAQ: NVDA) is first on the list, and for good reason. Graphics processing units (GPUs) are considered must-have equipment for data centers, providing the computing power needed to train and operate advanced AI models.

Nvidia sells the processors and networking equipment. It generally recognizes revenue when the equipment is delivered under the applicable sales terms, rather than waiting years for the customer to earn a return.

But Nvidia isn't completely insulated from its customers' financial problems. The company has invested in more than 200 other companies, many of them key tech, cloud, or infrastructure companies. That creates financial exposure beyond selling chips.

Broadcom (NASDAQ: AVGO) approaches the AI business differently. It designs custom AI accelerators for major technology companies and supplies networking equipment that connects thousands of processors inside data centers. Broadcom gets paid for the semiconductors and networking products it supplies, rather than collecting monthly fees for operating the data centers.

If anything, Broadcom's risk comes in customer concentration. If a major customer reduces its AI spending, Broadcom can't necessarily sell those specialized processors to somebody else.

Advanced Micro Devices (NASDAQ: AMD) seeks to take market share from Nvidia with its Instinct AI accelerators and EPYC server processors. It has an agreement with OpenAI for deploying 6 gigawatts of AMD GPUs over multiple years, beginning with the MI450 series. That's an opportunity to establish AMD as a major alternative to Nvidia.

But there's an interesting wrinkle: AMD also granted OpenAI warrants to acquire up to 160 million AMD shares, subject to purchasing milestones and other conditions. That means AMD is doing more than selling hardware. It's offering OpenAI a potential equity incentive tied to the success of their relationship.

Three companies that wait for customers to pay

Oracle (NYSE: ORCL) is making an aggressive push into AI infrastructure through Oracle Cloud Infrastructure, its platform with more than 200 services for customers, including GPU computing and autonomous AI databases.

Unlike chip suppliers, Oracle builds and operates data centers and rents computing capacity to customers. While Oracle's long-term agreements provide recurring revenue, the company must also secure the equipment, power, and facilities to deliver its contracted services. And if a customer cannot meet its obligations, Oracle could be left with infrastructure that was built for a particular workload.

CoreWeave (NASDAQ: CRWV) is one of the clearest examples of the financial risks associated with AI infrastructure. The company operates GPU-powered cloud facilities and signs contracts with customers that need enormous computing resources.

Its relationship with Nvidia is particularly interesting. Nvidia supplies the chips, owns a stake in CoreWeave, and has agreed to purchase unused computing capacity under certain conditions through 2032. That arrangement provides CoreWeave with some protection against insufficient customer demand.

But CoreWeave still has to finance its infrastructure, service its debt, and deliver computing capacity on schedule. Even a large contracted revenue backlog doesn't put cash in the bank immediately.

Nebius Group (NASDAQ: NBIS) follows a model similar to CoreWeave, as it develops AI computing infrastructure and sells access to customers. Nebius has long-term contracts with its customers to justify investments in expensive equipment, but those agreements also impose obligations to deliver capacity. The company has to pay for its infrastructure even if a project is delayed or if a customer scales back its commitment. That means Nebius has to turn demand for computing power into reliable payments quickly enough to support its expansion.

Alphabet has a foot on both sides

Alphabet (NASDAQ: GOOGL) (NASDAQ: GOOG) has an advantage that separates it from most of these companies. Through Google Cloud, Alphabet sells computing capacity to outside customers, including access to Nvidia GPUs and Google's own Tensor Processing Units (TPUs). It uses its infrastructure to power products across its business.

But it also faces the cost of building data centers, and its AI investments must generate adequate returns. Unlike smaller infrastructure providers that depend heavily on contracted customers, Alphabet has established businesses that can help finance its expansion.

Who carries the risk?

There's plenty of money flowing into AI infrastructure, but not every company faces the same financial exposure.

Nvidia, Broadcom, and AMD generally get paid for supplying equipment. Oracle, CoreWeave, and Nebius must invest in infrastructure and collect revenue over time. Alphabet participates on both sides.

The Microsoft-Nscale cancellation shows why those distinctions matter. Even a major technology customer can walk away from a project when the infrastructure can't be delivered as planned.

For investors, the size of an announced AI agreement is only part of the story. The more important issue may be who has already been paid -- and who is still waiting for their money.

Patrick Sanders has positions in Nebius Group and Nvidia. The Motley Fool has positions in and recommends Advanced Micro Devices, Alphabet, Broadcom, Microsoft, Nvidia, and Oracle. The Motley Fool has a disclosure policy.