Nvidia Q2 Operating Profit Up 124% as Memory Shortages Emerge as a Growth Variable

Nvidia posted a sharp jump in second-quarter results, helped by stronger investment in AI infrastructure, rising compute demand centered on agentic AI, and increased compute demand centered on the next-generation Vera Rubin. The article's topic is Nvidia's 124% increase in second-quarter operating profit and memory shortages as a growth variable. Nvidia held its second-quarter fiscal 2027 earnings release and conference call on the 26th (local time). The fiscal 2027 second-quarter period covers May through July.

Nvidia's second-quarter fiscal 2027 revenue came to USD 96.221 billion, up 106% from a year earlier. Operating profit for the same period was USD 63.734 billion, up 124% year over year, and net income was USD 59.688 billion, up 126%. Revenue was about KRW 133 trillion, operating profit was about KRW 88 trillion, and net income was about KRW 82 trillion. The operating margin was about 66.2%.

Nvidia's operating profit and net income rose faster than revenue. The company showed a pattern in which operating profit and net income growth exceeded revenue growth, with both outpacing the increase in sales. This is interpreted as a result of expanding AI infrastructure investment and rising compute demand tied to agentic AI and Vera Rubin.

However, the focus of growth has shifted from securing demand to securing supply capacity. Memory shortages and rising component prices continue. As a result, there are expectations that gross margin will decline going forward, and the pace of expanding the AI infrastructure supply chain is likely to become a variable for growth and profitability.

Nvidia's profitability metrics improved sharply from a year earlier. Non-GAAP operating profit was USD 63.956 billion, up 124% from a year earlier, and net income was USD 53.954 billion, up 118%. EPS was USD 2.22, and gross margin came in at 75%.

GAAP net income included USD 7.771 billion in gains related to equity securities. However, non-GAAP net income, after adjustments for certain items, also rose 18% from the previous quarter, indicating that profit growth in the AI semiconductor core business remained intact.

Jensen Huang, Nvidia's CEO, said AI has reached an inflection point, token productivity and profitability are increasing, and the company has entered a stage in which compute translates into revenue.

The business that drove the results was the data center segment. Data center revenue was USD 89 billion, up 117% year over year and 18% from the previous quarter. Data center revenue accounted for about 92% of total revenue.

The expansion in data center revenue was also strong beyond large cloud providers. Revenue from hyperscalers was USD 48.71 billion, up 13% from the previous quarter, while AI Clouds, Industrial & Enterprise, or ACIE, revenue of USD 40.313 billion rose 25% from the previous quarter and 138% from a year earlier. ACIE revenue accounted for about 45% of data center revenue.

Colette Kress, Nvidia's CFO, cited expanding demand from AI startups, enterprises, and sovereign AI, as well as the buildout of neocloud facilities, as drivers of ACIE growth. Nvidia said GPU capacity at neocloud operators was about 3 GW at the end of 2025 and was expected to expand to more than 8 GW by the end of this year.

Nvidia identified agentic AI as the next demand driver. The scale of compute required was described as 15 to 100 times greater.

Nvidia cited agentic AI as a key basis for future growth in AI infrastructure demand. It said service development is shifting from a model in which people directly use AI to one in which AI reasons, plans, and uses multiple tools on its own, and that this change will greatly increase the required amount of compute. It explained that AI agents do not stop at a single question-and-answer exchange, but instead perform reasoning, planning, and repeated tool use.

Huang said the amount of compute needed for agents is 15 to 100 times greater, depending on the type of problem, compared with human use. He also said the amount of compute required is enormous. Nvidia expects that as AI models advance, the inference compute needed to process a single task will also increase.

Nvidia expects AI infrastructure demand to expand as inference compute rises. Against that backdrop, it is strengthening a strategy of supplying entire data centers as a single system rather than shipping only individual GPUs.

Huang said Nvidia is the only company in the world building an entire AI factory platform. He presented the full-stack system as Nvidia's competitive strength.

Nvidia is pursuing a strategy of handling AI model training through inference with an integrated platform of GPU, CPU, network, and software rather than selling individual GPUs alone. Even as major customers such as Google and Amazon expand their own AI semiconductor development, Nvidia is extending its platform influence by connecting technologies such as NVLink and networking to customers' in-house chips alongside its GPUs.

AWS is a representative example. AWS and Nvidia agreed to deploy an additional 2 million Nvidia GPUs, including Blackwell Ultra and Rubin series products, across AWS's global infrastructure in 2027 and 2028. This 2 million-unit volume is separate from the existing plan to adopt more than 1 million units.

AWS is simultaneously expanding development of its in-house AI accelerator, Trainium, and increasing adoption of Nvidia GPUs. This is expected to support expanded Rubin production, and 20% of data center revenue in the third quarter has been mentioned.

The next-generation Vera Rubin platform has entered the stage of expanded mass production. Nvidia announced that related systems are operating at CoreWeave, Google Cloud, Microsoft Azure, Oracle Cloud Infrastructure, or OCI, and Nebius. Vera Rubin is a next-generation AI platform that combines GPU, Vera CPU, NVLink, and networking.

Nvidia expects Rubin to account for about 20% of data center revenue in the third quarter. While rapid revenue contribution from Rubin after Blackwell is anticipated, whether the gap during the product-generation transition can be minimized is seen as a key variable for sustaining future growth rates.

Nvidia forecast that revenue for fiscal 2028 will increase by about 70% from a year earlier. At the same time, the company’s disclosure of growth rates a year ahead was an unusual move compared with its usual practice of providing quarterly guidance.

Nvidia's growth outlook is being presented on the basis of supply capacity rather than actual demand. The company appears to view supply capacity, rather than a slowdown in AI investment, as the main constraint on future growth.

In this regard, Kress said that based on customer demand forecasts, room for growth next year would exceed 70%. However, she said that after reflecting supply constraints, growth of about 70% is expected next year.

Huang also said that its own demand is far greater than 70%. He then said he was confident that the current supply it has secured is enough to achieve 70% growth.

Accordingly, Nvidia is being asked to expand supply across the entire AI infrastructure stack, including not only GPUs but also memory, power, cooling, and data center capacity. In the process, the burden of securing memory within the supply chain is also increasing.

Nvidia's commitments related to manufacturing, supply, and production capacity rose from USD 119 billion in the first quarter to USD 279 billion, or about KRW 386 trillion, in the second quarter. That was more than double the first-quarter level. Nvidia said the main reason for the increase was memory procurement.

Nvidia's commitments are concentrated in fiscal 2027 through fiscal 2029. The second half of fiscal 2027 accounts for USD 92 billion, fiscal 2028 for USD 87 billion, and fiscal 2029 for USD 88 billion. This concentration of large commitments is linked to the need to secure long-term memory supply on the assumption of expanded AI accelerator shipments. The target inventory includes high-bandwidth memory, or HBM.

To that end, Nvidia is continuing to work with major memory suppliers to secure production capacity. Nvidia and SK Hynix are jointly developing memory for next-generation AI factories, and the partnership takes the form of a multi-year technology collaboration. Kress said a substantial portion of the current memory shortage stems from AI infrastructure buildout, and described the tightening memory supply as a symptom of Nvidia's surging growth demand.

Nvidia is also expanding its influence over memory technology. On the 26th (local time), Nvidia unveiled NVHBM, a next-generation high-bandwidth memory technology for NVLink Fusion. Nvidia plans to establish a structure in which multiple memory vendors can supply NVHBM.

The NVHBM approach integrates the memory controller into the HBM base die. Its defining feature is improved bandwidth and better power efficiency compared with standard HBM4E. Annapurna Labs, Amazon's in-house semiconductor development unit, participated as the first partner.

The memory supply shortage is starting to affect Nvidia's profitability. Nvidia's gross margin was 75% in the second quarter, and guidance for the third quarter was put at around 74%. Nvidia expects gross margin to fall to 71% to 72% in the fourth quarter because of rising memory and component costs. For Nvidia, which has long benefited from faster profit growth than revenue growth thanks to the high prices and market dominance of its AI accelerators, cost pressure is now beginning to show up in earnest.

Going forward, the impact of memory supply and pricing on shipment volume is seen as a variable. Memory supply and pricing are also factors that could affect Nvidia's ability to maintain its high profitability.

Even under this cost pressure, Nvidia guided third-quarter revenue to USD 108 billion, or about KRW 149 trillion. That was above the market estimate of about USD 104.2 billion. However, the third-quarter revenue outlook did not include data center computing revenue from China.

U.S. export restrictions on advanced semiconductors to China are cited as the reason for continued uncertainty in the China business. Even so, Nvidia said it expects quarterly revenue to exceed USD 100 billion even without assuming a recovery in China. That view is based on the expectation that AI infrastructure demand will continue in other regions and customer segments.

Source: IT DAILY · Kim Byeong-ju
Original: https://www.itdaily.kr/news/articleView.html?idxno=241243

References

• https://www.itdaily.kr/news/articleView.html?idxno=241243