Nvidia has alleviated concerns about a peak-out in AI investment by posting record earnings for the 13th consecutive quarter. As AI infrastructure investments continue, both Nvidia and major global tech companies are expanding the use of high-bandwidth memory (HBM) in their custom AI chips, leading to a surge in demand for HBM from Samsung Electronics and SK Hynix.
On August 27, Nvidia reported that its revenue for the second quarter of fiscal year 2027 (May to July) reached $96.22 billion, a 106% increase compared to the same period last year. Revenue from data centers, which includes AI accelerators, surged 117% to $89 billion, accounting for 92% of total revenue. Nvidia projected third-quarter revenue to exceed market expectations at $108 billion and forecasted approximately 70% growth for fiscal year 2028.
To meet the soaring demand for AI, Nvidia is accelerating its efforts to secure memory supplies. The company's purchasing agreements related to supply chain and production capacity increased 2.3 times, from $119 billion in the previous quarter to $279 billion. A significant portion of this is focused on securing the memory and production capacity needed for the coming years.
Demand for HBM4 is expanding beyond Nvidia to include major tech companies' custom AI chips. OpenAI plans to deploy its custom AI chip 'Jalapeño', co-developed with Broadcom, for its AI models later this year, with HBM4 confirmed to be supplied by Samsung. As more companies, including Google, develop their own AI chips like the Tensor Processing Unit (TPU), HBM manufacturers are finding opportunities to secure large clients beyond Nvidia.
Samsung is expected to expand its customer base into the application-specific integrated circuit (ASIC) market with HBM4. While SK Hynix has solidified its leading position in the HBM market centered around Nvidia, Samsung is diversifying its supply strategy by collaborating with ASIC companies like Broadcom. SK Hynix has also begun mass production shipments of HBM4 in the second quarter and is collaborating with Nvidia on next-generation memory development to maintain its existing advantage.
With the increase in HBM supply and rising prices, both companies are expected to continue their growth trajectory. The expansion of AI infrastructure has led to ongoing memory supply shortages, enhancing the pricing power of memory manufacturers. The increased share of high-value products like HBM4 is anticipated to raise the average selling price (ASP). Global investment bank Goldman Sachs predicts that next year, the ASP for HBM will rise by 44% from this year, reaching $17 per gigabyte.
* This article has been translated by AI.
Copyright ⓒ Aju Press All rights reserved.