NVIDIA Touts Ongoing AI Growth in Q3 Report
NVIDIA (Nasdaq: NVDA) reported another quarter of strong earnings last night, while management faced questions about the ramp-up of its new AI chips and services.
The numbers were indeed eye-popping. Sales for the quarter were $35.1 billion, up 17% sequentially and 94% year-over-year (y/y). Datacenter sales of $30.8 billion accounted for nearly 90% of that revenue and were up 17% sequentially and 112% y/y. Adjusted earnings per share was $.81, up 19% sequentially and 103% y/y. Adjusted net income was an astounding $20 billion, up 18% sequentially and 100% y/y.
These are great numbers that beat analyst estimates on all fronts. But NVDA shares were volatile in early trading today and were down $2.50 to $143 (-2%) in in midday trading. The picture now shifts to what can happen in 2025. Investors have already bid up NVIDIA shares on the excitement about the growth in AI infrastructure, but now they are looking at concerns about how sustainable that might be in 2025. Growth has slowed: Q1 sales increased 262% y/y and Q2 sales rose 122% y/y. The law of large numbers makes scaling new heights more difficult.
One of the issues that caught investors' eyes was the projection for 70% revenue growth in the fourth quarter, about $400 million above consensus, according to a research note from Sebastien Naji, an analyst with William Blair.
Much of the growth in sales of NVIDIA’s massive GPUs has been down to a limited number of hyperscaler public cloud providers, which have established their so-called AI factories. While they continue to place orders, these companies are also looking for alternatives from AMD and others companies—and they are developing chips of their own, as evidenced by AWS's Trainium and Inferentia chips, Microsoft's Azure Maia AI Accelerator, and Google's Trillium TPU.
Management met the issue head on last night, saying that new technologies for growing AI models are emerging and inference demand is rising. CEO Jensen Huang said growth is coming in AI from many angles. And for NVIDIA’s emerging GPU series, Blackwell, demand is exceeding supply.
Blackwell Questions
Blackwell dominated the earnings call. Analysts questioned delays in production of the chips and voiced concerns that getting the product out the door will shave a few points off gross margin. The reported 75% gross margin dropped a fraction of a point last quarter, and management acknowledged it could go as low as 71% in coming months, as Blackwell production remains complex and costly.
Still, management was bullish on Blackwell, saying it is in full production after an engineering change that seems to have held things up a bit. According to CFO Colette Kress, 13,000 of the chips were sampled to customers during the quarter, and OpenAI, Microsoft, SoftBank, and Oracle have implementations in the works.
Other Areas of Datacenter Growth
Meanwhile, NVIDIA’s other datacenter products are selling well. The Hopper series, Blackwell’s predecessor, has taken root in the market, with hyperscalers AWS, Microsoft, and CoreWeave now offering NVIDIA H200-powered cloud instances and Google Cloud and Oracle reportedly planning similar instances.
How will Hopper overlap with Blackwell? It’s too soon to tell, Kress said. Sales of Hopper will continue through the next quarter, though Blackwell demand is growing as well, so it remains a wait-and-see situation.
Blair's Naji said that Hopper demand will continue to underpin the growth while the market shifts to Blackwell:
"Looking ahead, with Blackwell expected to continue ramping up into the first half of fiscal 2026, Hopper demand should remain robust, still representing the majority of data center revenue until likely the second half of fiscal 2026. We note that despite some temporary headwinds to profit margins as Blackwell is ramping up, non-GAAP operating margins should remain resilient in the mid-60% range, underpinning the sustained strong demand for Nvidia chips, the increased complexity of its system-level solutions, and the still limited competition from other vendors."
NVIDIA’s ongoing strength in datacenter environments was evident this quarter in several projects, including the establishment of a sovereign AI network in Denmark; a Blackwell-based supercomputer in the works by SoftBank in Japan; and the xAI Colossus supercomputer in Memphis, Tenn., which features 100,000 H100 GPUs and is supported by NVIDIA’s Spectrum-X Ethernet.
NVIDIA Taps into Telecom
NVIDIA has expanded its AI involvement in areas outside the datacenter. Within the latest quarter, for instance, the company has introduced the NVIDIA AI Aerial platform for telecom service providers, which provides artificial intelligence for use in mobile networks.
NVIDIA is also working with a number of providers, including T-Mobile, as well as equipment suppliers Nokia and Ericsson to promote the adoption of what it calls AI-RAN, which adds mobile radio access network (RAN) technology as a layer of the AI stack. This tack is aimed at reducing complexity and cost while increasing the functionality and monetization of services on mobile networks.
Futuriom Take: While the rollout of Blackwell chips may nibble at margins for a couple of quarters, NVIDIA scored another strong quarter and is showing no signs of slowing down its momentum in the AI markets.
