Nvidia earnings recap: The AI party rages on as the chip-maker warns 'extreme' memory costs will push down margins

See you next quarter!

"Is our goal to put as much compute on the plot of land? Is our goal to put more compute into one gigawatt or less?" Nvidia's CEO says. "Obviously, we would like that, you know, the speed of light answer. The perfect answer is actually infinity per gigawatt."

"And so, if we could literally get a trillion dollars of compute into one gigawatt and one piece of land power shell, it would be a fantastic outcome," Huang says. "The answer is directionally, in that direction."

The final question is about scaling compute and Nvidia's general chip ambitions.

AI is already shifting from tools prompted by people to agents that run continuously in the background, Nvidia CEO Jensen Huang says. He's still confident that it will drive Nvidia's massive profits.

Huang says he doesn't see open models as a big risk to the proprietary AI models. "Both closed and open models are going to succeed," he says. "And they're both simultaneously driving our sales." Huang stresses that open models, the leading of which are made by Chinese-based companies, run on Nvidia.

There is at least one exception to his argument. Earlier on Wednesday, Z.ai revealed that it was behind the mysterious "Ox-Alpha" model. In announcing the release of GLM-5.3-Flash, Z. ai says it ran the open-weight model on Chinese chips.

Huang says Nvidia's work with hyperscalers like Microsoft Azure and AWS only represents "half the picture."

Huang says Nvidia's work with smaller, more specialized firms is important, too. "They really need an entire factory platform built for them," Huang says. "We add just a tremendous amount of value."

Nvidia's expanding work with suppliers, power generators, and data center developers gives the company greater visibility into the future of the AI boom, Huang says.

The company is sharing its first-ever full-year revenue guidance so its customers and suppliers — all committing significant funds to the AI build-out — can use the same expectations, the CEO says.

CEO Jensen Huang kicks off the Q&A section with analysts. He's answering a question about Nvidia's confidence heading into the next quarter.

Nvidia's CFO says the cost of memory has surpassed Nvidia's expectations: "The magnitude of the price increase has exceeded our prior expectations and are headed even higher into next year," Colette Kress says.

She stresses that "tighter memory supply is a symptom of the same demand surge that's driving our own growth." Nvidia expects supply to remain a bottleneck "at least through the end of fiscal year 28."

Kress says Nvidia isn't including revenue from China in its next-quarter expectations because of "ongoing geopolitical uncertainty." She says less than 1% of Nvidia's total data center revenue in Hopper 200 products went to customers in China in the last quarter.

The company is still expecting revenue in its next quarter to race past $100 billion, even without any revenue from China.

CFO Kress says Nvidia is not making "loans" for data center buildouts. Instead, she says the world's largest company by market cap is coming out ahead in the revenue-share deals it's striking.

"We get paid twice, once on the hardware sale and again through the share of rental revenue," she says. She adds that "independent capital still underwrites every deal on its merits."

Nvidia stock was down after hours immediately following the earnings release — now, nearly 20 minutes into the analyst call, the stock is trading up 4%.

CFO Colette Kress says Nvidia continues to "see demand for approximately 20 billion in total server CPUs." Increasingly, these chips are being used to power AI agents. She says that the company's outlook "is for CPU revenue to more than double" in fiscal 2028, "positioning us as one of the world's leading server CPU suppliers."

We know that Nvidia has sold billions of dollars' worth of chips to AI labs like OpenAI and Anthropic. Now, it sees more opportunity beyond the massive AI companies.

"Beyond the hyperscalers lies a massive market anxious to adopt AI," Kress says in her opening remarks, "customers with no interest in designing their own custom silicon."

CFO Colette Kress announces "an expansion" of Nvidia's partnership with Amazon Web Services.

"AWS is deploying an additional 2 million GPUs starting this quarter," Kress says. Kress adds that Amazon will adopt Nvidia's physical stack to power its warehouse robots.

"Closed and open models alike, adoption is skyrocketing," Nvidia's CFO says. "Nvidia runs the leading closed models."

CFO Colette Kress says Nvidia expects revenue to grow by 70% in fiscal 2028, adding that "this is a supply-constrained outlook."

"Hyperscale revenue of 49 billion grew 13% sequentially, driven by sustained strength in Blackwell, reinforcing that more compute drives more revenue as new GPU capacity comes online," CFO Colette Kress says. "Our hyperscale customers delivered strong financial results in the quarter with accelerating revenue growth and expanding margins."

Nvidia CEO Jensen Huang and CFO Colette Kress are on the line and will begin with prepared remarks followed by a live Q&A with analysts.

"Even extraordinary growth can fail to satisfy investors as scrutiny of AI spending and its financing intensifies," writes Emarketer analyst Jacob Bourne.

"Nvidia still faces a changing competitive landscape, with some of its largest customers developing their own chips while startups target parts of the AI compute market," Bourne says. "Yet Nvidia continues to deepen its relevance across the AI stack, expanding beyond training into inference, models, networking, and broader AI infrastructure. Its ability to adapt as AI workloads and competitive dynamics evolve provides a strong argument for the durability of its position."

The chipmaker writes in its 8-K that it's stepping in to help customers secure land, power, and data center capacity. It's another sign that the company remains a key financier in the AI boom — something that has some investors skittish amid concerns of circular financing.

• Revenue: $96.2 billion vs. $92.29 billion estimate
• Adjusted EPS: $2.22 vs. $2.09 estimate
• Data center revenue: $89.0 billion vs. $85.83 billion estimate
• Adjusted gross margin: 75% vs. 75% estimate

Shares were down at about 3:35 p.m. as investors braced for the results. The stock dipped 1% to $210.56, slightly higher than the intraday low of $209.47. Option pricing implies a 5% move in either direction after earnings, or about a $280 billion swing in market cap.

A pattern has held for four straight quarters of earnings: Nvidia beats expectations, and the stock falls anyway. Chris Beauchamp, chief market analyst at IG, said Nvidia has become a victim of its own success. "Beating expectations isn't enough and hasn't been for a while," he said. "What matters now is the guidance for Q3 and beyond," he added, noting that investors want signs that demand is broadening beyond the hyperscalers.

Nvidia is still the chip king, but with new market dynamics at play, it needs to reassure investors that it has plenty of room to run. "That kind of guidance would really be needed to trigger a rerating in valuation," Beauchamp added.

Some of Nvidia's biggest customers are struggling to out-earn their AI spending. Google reported a negative free cash flow of $5.9 billion in Q2, while Amazon's free cash flow fell to a $7.6 billion outflow for the 12 months ending in June. Meta's free cash flow plunged 91% from the previous year. Given that these companies buy Nvidia's chips for internal use and to rent them to their own customers, having less cash is potentially a problem for Nvidia.

On top of that, Google and Amazon are firing up their own competing chip businesses. Google said it delivered its first TPUs (Tensor Processing Units) to customer data centers in Q2, and Amazon said it is also in discussions to sell its Trainium chips directly. Nvidia said in November that it was "delighted" by Google's success — but are Nvidia's investors?

BofA says Nvidia is a top pick in semiconductors, and pointed to five things investors should focus on heading into the earnings call:

1. The launch and supply of Vera Rubin, and if there are changes to Nvidia's forecast that it can generate more than $1 trillion in revenue from 2025 to 2027
2. Clarity on vendor financing and backstops in Nvidia's deal with OpenAI, and when cash can shift to buybacks
3. Whether Nvidia can maintain gross margins despite memory cost inflation
4. How fast demand is growing among hyperscalers and the accelerated-computing infrastructure and enterprise-related firms Nvidia works with
5. If open-source AI models become more popular relative to closed models, and the implications that could have on Nvidia's addressable market.

The bank reiterated its "buy" rating and its $350 price target for the stock, implying 64% upside from current levels.