Fiscal Q2 2027 earnings

Sales hit 133 trillion won, up 106% year on year

13 consecutive record quarters

AI investment boom seen lasting through 2028

Fiscal 2028 sales forecast to grow 70%

'Growth rate would be far higher without supply constraints'

Supply commitments for memory and other inputs reach 385 trillion won

Nvidia CEO Jensen Huang explains the architecture of Vera Rubin at the Computex trade show in Taipei, Taiwan, in June. Analysts say Nvidia still has room to rise even as semiconductor stocks have been falling in recent sessions. [Reuters]
Nvidia CEO Jensen Huang explains the architecture of Vera Rubin at the Computex trade show in Taipei, Taiwan, in June. Analysts say Nvidia still has room to rise even as semiconductor stocks have been falling in recent sessions. [Reuters]

Nvidia, the world's top AI chip maker, delivered its 13th consecutive earnings surprise, reaffirming that demand for AI infrastructure remains strong — and projecting that the investment boom will continue at least through 2028.

The company also warned that a memory chip bottleneck, where supply is failing to keep pace with surging demand, will persist through the end of 2028. Nvidia's supply commitments to secure memory and other inputs more than doubled in a single quarter.

The results suggest that Samsung Electronics and SK hynix, which dominate the global memory market, will maintain their pricing power and bargaining leverage — and continue to see earnings improvement — for the foreseeable future.

Nvidia said Wednesday (local time) that revenue for the second quarter of fiscal year 2027 (May–July) reached $96.22 billion (132.7 trillion won), more than doubling from the same period a year earlier and setting a new all-time quarterly record for the 13th straight time. The result beat the $92.17 billion consensus estimate compiled by the London Stock Exchange Group (LSEG) by more than $4 billion. Earnings per share came in at $2.22, topping Wall Street's forecast of $2.10.

Revenue from the data center segment — Nvidia's core business — reached $89 billion, surpassing the market estimate of $85.8 billion. Sales to hyperscalers operating large-scale data centers totaled $48.7 billion, while revenue from cloud and enterprise customers outside that group came to $40.3 billion. Both segments more than doubled year on year, a sign that demand for Nvidia's AI accelerators is spreading rapidly beyond a handful of big tech companies.

In a press release, Nvidia said it plans to supply at least tens of billions of dollars' worth of GPUs — about 2 million units — to Amazon Web Services (AWS) between fiscal years 2027 and 2028.

CEO Jensen Huang identified government-led AI infrastructure buildouts, alongside hyperscalers, as a key driver of future AI market growth. "Most people are only looking at hyperscalers, but that is only half of total demand," Huang said. "We also need to pay attention to the enterprise and government AI market, which makes up the other half."

Nvidia projected strong growth continuing at least through 2028. Chief Financial Officer Colette Kress said on the post-earnings conference call that she expects fiscal year 2028 revenue to grow about 70%. "Even at our current scale, demand is growing faster, and customer order outlooks show that the pace of growth will double next year," she said.

Constrained memory supply is acting as a brake on even higher growth. Kress explained the basis for the 70% revenue growth estimate, saying that actual AI-related demand is approaching 100%, but the 70% figure reflects what the company can actually supply. "If there were no supply constraints, the growth rate would have been far higher," she said, adding that the memory bottleneck is expected to last through the end of 2028.

Huang echoed that view. "AI has now reached an inflection point and is doing genuinely useful things," he said. "This year we grew 100% year on year. Without supply constraints, it would be far higher." He noted that while Nvidia has already secured substantial supply, more semiconductor production capacity is needed to meet the explosive growth in demand.

Nvidia's long-term purchase and supply commitments to lock in memory and other production capacity also expanded sharply. Commitments for the quarter totaled $279 billion, up 2.3 times from $119 billion the previous quarter. A significant portion of the increase is believed to be related to securing memory supply.

Jensen Huang (right), Lee Jae-yong (left) and Euisun Chung at a meeting dubbed the 'kanbu summit.' [Yonhap]
Jensen Huang (right), Lee Jae-yong (left) and Euisun Chung at a meeting dubbed the 'kanbu summit.' [Yonhap]

Market observers have long noted that the three major memory makers are securing production capacity by collecting advance payments from customers as part of long-term supply agreements (LTAs). Nvidia's sharp increase in supply commitments shows that such preemptive investment to lock in memory supply is in fact accelerating significantly.

Memory prices are also climbing steeply. "There is fierce price competition in the memory segment. Price increases have far exceeded initial expectations and are forecast to rise further next year," Kress said, while also noting that "the memory supply shortage is a result of the demand surge that is driving our growth."

Still, surging memory prices are weighing on Nvidia's profitability. The company forecast that its gross margin, which stood at around 74% this quarter, will slip to 71–72% in the fourth quarter of fiscal year 2027, which runs through January. Rising memory costs are cited as the primary factor.

Nvidia expects to partially recover its margins from fiscal year 2028 onward, in part by raising product prices, with gross margins projected to climb back to the 72–73% range. The company appears to believe that, as long as AI chip demand continues to outpace supply, it can pass higher input costs on to customers.

Nvidia also plans to begin mass shipments of the Groq 3 LPX by the end of this quarter. The Groq 3 LPX is a system bundling 256 language processing units (LPUs) optimized for inference workloads. The Groq 3 LPU is manufactured entirely on Samsung Electronics' foundry 4-nanometer process. Vera Rubin, Nvidia's next-generation AI platform using HBM4 from both Samsung Electronics and SK hynix, has already entered full-scale mass production.

Nvidia's share price briefly fell in after-hours trading immediately after the earnings release, as concerns over rising memory costs and a dimming profitability outlook initially dominated sentiment.

Sentiment recovered quickly, however, after Nvidia projected 70% revenue growth in fiscal year 2028 on the conference call and emphasized that AI demand continues to outstrip supply capacity. Nvidia's shares subsequently rose as much as 4.4% in after-hours trading.


go@heraldcorp.com
yckim6452@heraldcorp.com