Nvidia has been the profit engine of the artificial intelligence boom, filling up data centers with its graphics processing units, which are also powering the biggest AI models.
But if there's one persistent concern among investors regarding Nvidia's ability to build on its $5 trillion market cap, it's customer concentration. Nvidia counts on the group known as the hyperscalers — Amazon, Google and Microsoft — for an outsized amount of revenue, as those companies buy GPUs in bulk for their own workloads, and sell access to them through their giant cloud businesses.
Add to that Meta and SpaceX, which have also been massive buyers as they build their own models, while also starting to rent out some of their Nvidia capacity to other companies that need it right away.
In its last earnings report in May, Nvidia changed the way it reports financials to split out hyperscalers from the rest of the pack, which it now groups as AI clouds, industrial and enterprise, or ACIE. Nvidia doesn't specify which companies fall into the former group, but CEO Jensen Huang has defined the group pretty clearly.
"The easiest go-to-market, of course, is the hyperscaler, because there are only five or six of them," Huang said on the May earnings call. "The rest of them, the rest of the industry, represents 250,000 companies around the world."
As Wall Street gears up for Nvidia's fiscal second-quarter earnings report after the bell on Wednesday, the ACIE portion of the customer base is coming under the microscope. In the first quarter, the two segments were almost equal, with Nvidia reporting $37.9 billion in hyperscaler sales and nearly $37.5 billion in ACIE revenue. Nvidia highlighted that its ACIE revenue grew at 31% from the prior period, topping 12% growth from hyperscalers.
Skepticism has started seeping into the stock. The shares fell 2.9% on Monday, dropping for a seventh straight day, the longest losing streak since 2022. They're down 7.5% over that stretch.
Nvidia stakeholders want the company to become less reliant on a small subset of customers, and they also worry about the ability of those companies to further bolster their spending. Amazon and Alphabet turned cash flow negative in the second quarter, and Meta's cash generation dwindled by more than 90% from a year earlier. Both of Elon Musk's public companies — SpaceX and Tesla — reported negative free cash flow as they pursue AI expansions.
"This story underneath the surface is investors' concern about just how sustainable the run has been for Nvidia, and feeling like the hyperscalers just can't give much more," Gene Munster, managing partner at Deepwater Asset Management, said in an interview. "They want to start to see the other segment starting to kick in."
Other business to grow faster
Nvidia in May provided nine restated quarters of history on its website, showing hyperscalers making up about 55% of data center revenue in the past year. Hyperscaler revenue soared 115% from a year earlier in the most recent quarter, topping the 74% growth from ACIE customers.
Analysts expect the trend to flip going forward. The ACIE side of the business is expected to show 149% annual growth over the past year to $43 billion in the second quarter, while hyperscaler revenue is likely to show growth of 83% to $43.6 billion, according to StreetAccount. One potential big driver for the hyperscaler business is SpaceX, which aims to rapidly build out Nvidia-based data centers over the next year, Munster said.
In total, analysts see Nvidia's revenue almost doubling from a year earlier to $92.2 billion, according to LSEG. Data center continues to account for a bigger chunk of sales, with that division expected to reach $86.3 billion, based on StreetAccount estimates. That would be 94% of the total, up from 92% in the first quarter.
For the full year, analysts expect 83% revenue growth to $396 billion, before slowing to 44% expansion next year, according to LSEG.
In its effort to diversify, Nvidia is being proactive rather than waiting for new customers to arrive. Recognizing that very few companies have the cash or balance sheet to support purchases of its rack-scale systems, Nvidia is working with Wall Street to turn GPUs into an asset class.
Earlier this month, Nvidia unveiled a program with six leading financial firms that could pull together up to $500 billion in financing from investors who view chips as an investable asset, like real estate. Such a project, Huang said, could enable more companies to borrow money for GPU purchases at lower rates, because the chips will generate returns.
Nvidia said in the announcement that it had signed a memorandum of understanding with the investment firms. Few specifics have been presented.
"There's been a lot of headlines and big numbers and not a lot of details on how this stuff is going to work yet," Stacy Rasgon, an analyst at Bernstein, told CNBC last week. "Color that they could give on that would be helpful to put some investors' minds at ease."
Another crucial number in Nvidia's earnings report will be sales of Vera Rubin systems, which have just recently started to ramp up. Earlier this year, Huang told attendees at the company's GTC conference that he expects $1 trillion in sales through 2027 from current-generation Blackwell and Vera Rubin.
Questions about the long tail of the customer base could ease if Vera Rubin looks like an even bigger success than projected.
"We expect Nvidia to report strong results and guidance with ramping Rubin GPU shipments representing the key drivers of upside," KeyBanc analysts, who recommend buying the stock, wrote in a note over the weekend.