Nvidia Chief Executive Jensen Huang said on the 26th that artificial intelligence has reached an inflection point, pushing back against growing skepticism in the market about the durability of the AI boom.
Memory Shortage Persists: "Supply Constraints Continue"
Huang made the remarks as the company reported results for the second quarter of its fiscal year, covering May through July. "Tokens are productive and profitable, and compute has now become revenue," he said. "This time last year, there was only one AI lab driving the infrastructure buildout. Now we are in a golden age with new labs and startups pouring out." He added that Vera Rubin, which has entered full-scale production, "was developed for exactly this moment."
Huang stressed on a conference call the same day that the AI expansion remains firm. Asked by a brokerage analyst about the basis for Nvidia's own estimate of 70% revenue growth next year, he said actual demand related to AI is approaching 100%, but that the 70% figure was set based on the volume the company is able to supply. "Without supply constraints, the growth rate would have been much higher," he said.
Surging memory prices are indeed weighing on Nvidia. Costs to secure memory have risen sharply as demand jumps for high-bandwidth memory (HBM) and DRAM used in AI servers. Nvidia said memory bottlenecks and price increases will continue, and on that basis revised its margin outlook for next year to a range of 73.5% to 74%. That is slightly below market expectations in the high 74% range. The company also plans to raise product prices in response to higher costs, starting in the first quarter of its next fiscal year.
Huang also pointed to the recent advance of open AI models, led by Chinese developers, as a factor that could drive Nvidia's revenue. "The world needs both closed models and open models," he said. "The strength of Nvidia's technology is that it is the most versatile. Because most AI models are built on Nvidia, our position is very solid."
He also predicted that the outlook for government AI demand and sovereign AI projects will brighten further. "Most people are only looking at the hyperscalers, but that is only half of the demand," Huang said. "We should also pay attention to the other half, the enterprise and government AI market." Nvidia's sovereign AI business grew 35% from the previous quarter in the second quarter and more than tripled from a year earlier.
70% Revenue Growth Expected Next Year
Nvidia disclosed on the 26th that second-quarter revenue for the May-July period rose 106% from a year earlier to $96.22 billion, or about 133.2 trillion won. That exceeded the market consensus of $92.17 billion compiled by the London Stock Exchange Group (LSEG) by more than $4 billion, marking the company's 13th consecutive quarter of record revenue.
By segment, the data center division, which includes AI chip sales, posted $89 billion, up 117% from a year earlier and 18% from the prior quarter, accounting for 92% of total revenue.
Sales to hyperscalers, the operators of large-scale data centers, doubled to $48.7 billion from $24.2 billion a year earlier. Excluding those customers, sales to cloud providers and enterprises rose about 2.4 times to $40.3 billion from $16.9 billion, suggesting the customer base is gradually diversifying.
The edge computing segment, which covers chips for PCs, game consoles and vehicles, grew 27% from a year earlier to $7.2 billion. Earnings per share came in at $2.22, above Wall Street's estimate of $2.10. Gross margin was 75% and operating margin 66.5%.
Nvidia guided for the strong performance to continue into the third quarter of its fiscal year, with revenue rising further to $108 billion, or about 149.5 trillion won. That is above analysts' forecast of $104 billion.
Chief Financial Officer Colette Kress said on the conference call that the company expects revenue to grow about 70% in fiscal 2028, covering October 2027 through September 2028.
Nvidia shares closed 1.59% lower in regular trading, then jumped 4.4% in after-hours trading, hovering around $218 as of 5:15 p.m. Eastern time.
Amazon to Buy 2 Million More Nvidia GPUs
Nvidia also said it will supply an additional 2 million graphics processing units (GPUs) to Amazon Web Services' global infrastructure between 2027 and 2028. In a press release, the company said it is significantly expanding the strategic partnership to meet surging global demand for AI infrastructure.
Amazon had previously said it would install 1 million Nvidia chips in AWS data centers starting this year, and the new volume comes on top of that.
The chips covered by the agreement include Nvidia's Blackwell Ultra, Rubin and Rubin Ultra GPUs. Amazon and Nvidia did not disclose the financial terms, but with list prices for these products starting in the tens of thousands of dollars per unit, the total is estimated to reach at least tens of billions of dollars. Discounts offered to large customers should be taken into account.
Nvidia GPUs are a core element of the AI boom, underpinning the development and operation of most large language models (LLMs).
Amazon, Microsoft and Google, all Nvidia customers, are at the same time developing their own chips to compete with the company. These hyperscalers believe they can build hardware optimized for their own data centers and pass efficiency gains on to customers. Amazon has invested heavily in recent years in Annapurna Labs, its in-house chip development unit, which produces central processing units (CPUs) as well as AI accelerators designed to challenge Nvidia GPUs. Amazon has said it will use Nvidia networking technology in the next version of its own AI chip, Trainium.
Circular Financing Concerns Called "Strategic Investment"
Kress said on the conference call that the funding support at the center of recent debate over circular financing amounts to strategic investment. Leading AI companies such as OpenAI are constrained in their growth by a shortage of computing resources, so supporting them with capital and infrastructure to grow the AI ecosystem also benefits Nvidia, she said.
