NVIDIA (NASDAQ: NVDA) released its earnings report for Q2 of its fiscal year 2025 on Wednesday, and the result was pretty close to what I expected: continued dominance of the enterprise datacenter AI space, especially in the hyperscaler market, along with plenty of other areas of strength to show that NVIDIA is more than a one-trick pony. In commentary on CNBC, Yahoo! Finance, and CNBC again this week, I said that I expected the NVIDIA train to keep on rolling as it has been for another 12 months; the actual numbers from Q2 reinforce my view—with a couple of caveats.

Here’s an overview of the numbers.

Performance versus Expectations

• Q2 FY2025 revenue: a company-record $30.0 billion, up 122% year-over-year
• Q2 FY2025 diluted EPS: $0.67, up 168% YoY

Other Key Numbers

• Q2 FY2025 operating income: $18.6 billion, up 174% YoY
• Q2 FY2025 net income: $16.6 billion, up 168% YoY
• Q2 FY2025 revenue by category
  • Data Center: a record $26.3 billion, up 154% YoY
  • Gaming and AI PC: $2.9 billion, up 16% YoY
  • Professional Visualization: $454 million, up 20% YoY
  • Automotive and Robotics: $346 million, up 37% YoY

Analyst Notes

The question for NVIDIA’s earnings this quarter wasn’t ever “Is it going to be good?,” but rather “How good is it going to be?” The answer is very good indeed, with a few small points of caution.

All of the hyperscalers (AWS, Microsoft, Google, Meta) have serious FOMO about the generative AI opportunity. They fully understand the hype but are also ready to keep spending stacks of CapEx to keep up with each other—and their customers’ interest—in AI. And they’re not stopping anytime soon. Look at that over-the-top $26.3 billion in revenue from NVIDIA’s Data Center business: cloud service providers accounted for about 45% of that number, according to a comment by CFO Collette Kress during the earnings call.

One of the key growth areas for NVIDIA is with tier-2 CSPs and enterprises that have on-prem or colocated datacenters. Based on all my conversations with NVIDIA’s OEM customers, this is a significant area of opportunity for the company beyond what it’s doing with the hyperscalers. Something like 80% of all enterprise data lives on-premises—not in the cloud estates of the hyperscalers. Smart enterprises want to harness this data to fuel generative AI for targeted use cases, and NVIDIA is happy to help them with that. The earnings report didn’t distinctly call this out, but the supporting figures we are seeing in the financials of NVIDIA partners Dell Technologies and HPE are promising.

As I keep reminding anyone who will listen, NVIDIA isn’t just GPUs, and it isn’t even just chips. The company builds and sells entire systems—full datacenter trays and racks of equipment—with eye-watering margins. It also partners to sell even more types of systems, as we see with its partnership with Dell to build AI factories and its partnership with HPE to get enterprises rolling with private cloud AI. It also wins deals and gets even more embedded with customers thanks to its prowess in software. I thought all of that was reflected well in the Q2 growth numbers for the system-level business. Datacenter compute was up 162% YoY and networking was up 114% YoY.

The Potential Trouble with Blackwell

There has been concern about the production timetable for the forthcoming Blackwell GPUs and related platform components that are succeeding the current generation of Hopper high-end chips. Rumors swelled, but we hadn’t heard anything until the earnings call. A short delay in coming to market wouldn’t mean much—customers can make do with other NVIDIA parts already available—but it would be a problem if Blackwell were delayed by many months.

The earnings report clarified that Blackwell needed a mask change to achieve improved yields. CFO Kress clarified that “Blackwell production ramp is scheduled to begin in the fourth quarter and continue into fiscal 2026. In the fourth quarter, we expect to ship several billion dollars in Blackwell revenue. Hopper demand is strong, and shipments are expected to increase in the second half of fiscal 2025.” It’s hard to say what the delay means as the company had never given Blackwell details before.

NVIDIA’s Pressures

• The real question is: When do enough companies start seeing enough real business benefits to justify the extraordinary CapEx being plowed into GenAI? We’re already hearing about great internal AI-driven efficiencies from Cisco, Dell, IBM, Salesforce, and Walmart. But those are hellaciously smart companies with the deepest of pockets, so of course they’re going to find the benefits sooner than later. The broader business impact of AI doesn’t need to happen today or even this year, but at some point all that CapEx must be rationalized by tangible efficiencies or revenue gains for companies outside these giants. Otherwise, this GenAI bubble could burst—which would of course affect NVIDIA even though its offerings are diversified to prevent the worst damage. Think of it this way: Cisco was the NVIDIA of the dot-com era, and we saw the Cisco bubble burst as the downstream benefits of the Internet were delayed.
• Meanwhile, any company in any market with a position as dominant as NVIDIA’s is going to have a target on its back for competitors to aim at. AMD is close to parity with some of its GPUs, and it’s coming on strong—and quickly. Intel has had its ups and downs, but it is still a major force in the semiconductor industry and has its sights squarely set on AI. I would not be surprised to see AMD begin to change the dynamics of this market between now and the end of the year, and for Intel to do the same in 2025 and 2026. The biggest direct competitors are still the AI custom silicon coming from AWS (Tranium/Inferentia), Azure (Maia), Meta (MTIA) and Google (Trillium, a.k.a. TPU).
• Broadcom and Marvell, among others, are positioned to make lots of money on the AI boom as they serve the need for custom AI hyperscaler silicon. At the moment, however, there’s so much money to go around that they’re not cutting into NVIDIA’s sales in any significant way. But the ecosystem will do what it tends to do, bringing in smart new players like this to create more choice and mitigate risk.

Looking Forward

Barring a meltdown in the broader economy, which likely won’t happen, I see another 12 to 18 months of growth along its current lines for NVIDIA. Nobody outside the company itself likes to see NVIDIA with 95% of market share. But AMD might take away 5% of that as soon as the fourth quarter, and I wouldn’t be surprised if AMD, Intel, the hyperscalers, and other savvy players knocked NVIDIA down to, say, 75% sometime in 2025. But even then the company should be making loads of money.

Absolutely, we need to see the downstream business benefits of AI at some point, but that won’t really matter for a while yet. Right now we’re still building the foundation, and the hyperscalers will keep spending with NVIDIA to support this. Until the day—maybe years from now—when we see investors demanding reduced hyperscaler CapEx, the beat-beat-raise goes on for NVIDIA.

Recent MI&S Research on NVIDIA

Mistral NeMo: Analyzing Nvidia’s Broad Model Support — By Matt Kimball and Patrick Moorhead — August 5, 2024

Nvidia Doubles Down On AI And Taiwan At Computex 2024 — By Anshel Sag and Patrick Moorhead — July 30, 2024

Nvidia GTC 2024 Wrapup: Blackwell, MediaTek, Omniverse, And Vision Pro — By Anshel Sag and Patrick Moorhead — April 23, 2024

RESEARCH NOTE: NVIDIA GTC Networking and Mobile Telecom Announcements — By Will Townsend — April 4, 2024