NVIDIA Blowout Earnings Still Raise Questions

NVIDIA closed its fiscal year with a solid earnings report last night, though investors continue to question whether AI demand is inflated by unrealistic expectations.

First, the good news: NVIDIA beat Wall Street expectations in revenue and profit. Quarterly revenue was $68.1 billion, up 20% sequentially and 73% year-over-year (y/y). Quarterly EPS was $1.62 and gross margin was 75.2%. Annual revenue was $215.9 billion, up 65% y/y. Gross margin for the year was 71.3% and annual EPS came in at $4.77. NVIDIA is projecting approximately $78 billion in revenue for its upcoming fiscal year, along with 75% gross margin.

Quarterly datacenter sales of $62.3 billion increased 22% sequentially and 75% y/y. Of that, compute sales (GPUs) accounted for $51.3 billion, or 82% of the total, growing 58% y/y. Networking products, including NVIDIA’s Spectrum-X Ethernet switches, accounted for over $11 billion and grew at an astonishing 263%.

But the positive numbers and outlook failed to raise NVIDIA’s stock price. As of this writing, shares were down more than 4% Thursday morning, pointing to a range of concerns. Let’s look at a few:

Hyperscaler capex. NVIDIA was questioned last night about the potential for collective hyperscaler capex of nearly $700 billion to compress cash flow for the leading public cloud providers. Can they keep up this kind of capex momentum? And if not, what happens to NVIDIA’s prospects, since hyperscalers account for over half of its revenues? CEO Jensen Huang was adamant in his reply: “I am confident in their cash flow growing!” he said. “In the new world of AI, compute equals revenues.”

Revenue isn’t profit. While NVIDIA’s hyperscaler customers may well require more compute and tokens to achieve AI results, it’s not a given that those results will translate to profits, particularly if capex continues to swell and debt grows as hyperscalers extend their datacenter infrastructure. Further, depreciation of NVIDIA assets is an issue that could weigh on customers' profits, given the vendor’s yearly cadence of replacement systems.

Supply chain worries. Shortages of memory chips and snarls in the supply chain for other components needed by NVIDIA could potentially hinder future sales. But NVIDIA says it’s gathered sufficient supply to meet demand for its Grace Blackwell and upcoming Vera Rubin products through 2027. What happens after that remains a question. CFO Colette Kress acknowledged that supplies could be tight.

NVLink’s future. Throughout last night’s call, CFO Kress and CEO Huang continually pounded home the success of NVIDIA's NVLink 72 technology. Installations of Grace Blackwell chips with the NVLink 72 switch accounted for two-thirds of datacenter revenue in the quarter, Kress said, noting that "NVLink scale-up fabric has revolutionized computing." Still, competition is building in technology that links GPUs within the rack. Efforts such as Broadcom’s open-source Scale-Up Ethernet (SUE) and the consortium-driven Ultra Accelerator Link (UALink)—the latter supported by AMD, Astera Labs, Cisco, Google, HPE, Intel, Meta, Microsoft, and others—will challenge NVLink.

Circular deals. The question of NVIDIA’s ongoing investment in customers such as CoreWeave and its imminent investment in OpenAI raise red flags for many observers, who see the deals as circular: NVIDIA provides infrastructure, hyperscalers in turn deliver AI factories running tokens from NVIDIA-supported model builders, all buying NVIDIA products. The risk, of course, is that these mutual dependencies spell mutual risk if one lags expectations. NVIDIA CEO Jensen Huang has called these concerns “ridiculous.”

The China question. While the U.S. government has approved the sale of NVIDIA’s H200 chips in China, there’s no guarantee that the PRC will allow the chips into its enormous market. This has disappointed NVIDIA, which sees China demand as a rich potential source of revenue. Participation in the China market will be needed if the U.S. is to maintain its leadership position in AI compute, NVIDIA says.

While NVIDIA will likely continue to dominate the market for AI infrastructure for the foreseeable future, competition is building, and the above concerns and others, including NVIDIA’s key-man risk in its reliance on the ironclad leadership of 63-year-old Jensen Huang, have investors hesitating to believe in its clear goal of ongoing world domination.

Futuriom Take: NVIDIA’s ongoing success across datacenter AI infrastructure is meeting a headwind of investor skepticism, particularly related to projections beyond 2027. Still, NVIDIA claims that by increasing the rate of tokens processed by AI, it will continue to support ongoing growth in revenue-generating applications. We’ll see.