Is the Big Business of AI Dominated by Too Few Big Businesses?

U.S. companies expect to spend hundreds of billions of dollars on AI over the next few years, but some investors are worried those plans—and the stock market rally they've fueled—depend too much on just a handful of companies.

Chip giant Nvidia (NVDA) revealed in a regulatory filing last month that two direct customers accounted for nearly 40% of total revenue in the second quarter, with "Customer A" representing 23% of sales and "Customer B" 16%. At no other point since 2022, when ChatGPT sparked the AI craze, have two individual buyers represented a greater share of Nvidia’s sales. 

But experts note that the 40% figure likely overstates Nvidia's reliance on those two customers. Nvidia doesn't disclose who its customers are, but Customers A and B are likely distributors or system integrators that package Nvidia products in larger systems for sale to end users like hyperscalers and software companies.

Nvidia has less visibility into how much of its revenue comes from the hyperscalers. In the most recent quarter, the company estimated that two unnamed end users each accounted for 10% or more of total revenue. Unless they each account for well over 10%, the estimate suggests Nvidia's pool of final users could be more diverse than its pool of direct customers.

“Feast or famine,” is how Bill Kleyman, Senior Data Center Analyst at HostingAdvice, characterizes Nvidia’s sales. The company’s fortunes, Kleyman told Investopedia, “are increasingly tied to how much the big cloud platforms spend on AI infrastructure.”

Competitor Broadcom (AVGO) is in the same boat. One distributor accounted for nearly 30% of sales in the past two quarters, and Broadcom estimated that about 40% of revenue came from its five largest end users. Broadcom's sales have become more concentrated during the AI arms race of the past few years.

Nonetheless, according to Kleyman, the unique dynamic of the AI infrastructure buildout makes this level of concentration less risky than it otherwise would be. “Overall AI demand is so strong across many sectors that a shortfall from one large customer would likely be offset by others ramping up their investments,” he said.

Chipmakers aren’t the only companies in the AI supply chain that could be forced to grapple with the consequences of customer concentration. 

Enterprise software and cloud computing company Oracle (ORCL) reported earlier this month that its backlog swelled by nearly $320 billion in the most recent quarter. Shares soared as Wall Street hailed the “historic” quarter. But it quickly became apparent that nearly all of the backlog could be attributed to a 5-year, $300 billion cloud computing deal with OpenAI. 

The terms of the agreement are unknown, but experts say flexibility is always written into contracts of that size and duration. How much of Oracle's backlog is converted into revenue will depend on OpenAI’s usage, which hinges on factors beyond how much the public uses ChatGPT. 

“These deals are only as sticky as the models they serve,” says Rory Bokser, Head of Product at Moken.io, a decentralized computing network data provider. “If OpenAI changes architectures, moves off Oracle’s flavor of cloud, or pivots deployment strategy (edge vs. centralized), those ‘backlogs’ get revised real fast.”

Another risk of inking a $300 billion deal with one company is that the customer may not have $300 billion to spend. OpenAI's annualized recurring revenue reportedly reached $12 billion in July, and the company is targeting $125 billion in revenue by 2029. Whether OpenAI hits those aggressive growth targets will likely depend on the ability of other companies to use its models to expand their profit margins or develop revenue-generating applications. 

Without rapid revenue growth, OpenAI may need to lean on investors to supply it with the billions of dollars it expects to burn through by the end of the decade. A recent funding round put the company’s valuation at about $500 billion, making it one of the world’s most valuable private enterprises.

“It remains to be seen whether or not OpenAI will continue to receive the lavish funding it has received so far,” said Greg Osuri, founder and CEO of Akash, a decentralized computing marketplace. Future funding could become scarce if AI implementation fails to live up to Wall Street's lofty expectations or economic stress forces investors and creditors to pull back.