NVIDIA Sees Memory Bottleneck Lasting Through 2028
NVIDIA, a key player in the global artificial intelligence (AI) industry, expects the memory semiconductor supply bottleneck to persist through early 2028. As demand for AI accelerators surges, securing supplies of high-bandwidth memory (HBM) and other key chips is expected to become a critical fact
NVIDIA, a leading player in the artificial intelligence industry, anticipates a memory semiconductor supply bottleneck to continue through 2028. The surge in demand for AI accelerators has made securing supplies of high-bandwidth memory (HBM) and other critical chips a pivotal factor in maintaining growth. During their earnings conference call on August 27, NVIDIA disclosed that the ongoing memory shortage would continue to hinder their expansion at least until the fiscal year 2028, which spans from February 2027 to January 2028.
Their CFO, Colette Kress, revealed that the company is facing extremely high memory prices, with further increases expected next year.
NVIDIA's aggressive strategies to secure chip supplies have significantly contributed to the current shortage. Their disclosed supplier purchase commitments have more than doubled, soaring from $119 billion in the preceding quarter to $279 billion in the current quarter. This substantial increase is largely attributed to memory supplies. As NVIDIA invests heavily in securing HBM supplies, other major tech and cloud companies are compelled to offer higher prices to secure semiconductors.
Moreover, as memory manufacturers such as Samsung Electronics and SK hynix prioritize higher-margin HBM production, shortages are extending to conventional server DRAM and enterprise solid-state drives (eSSDs). The substantial revenue-generating potential of NVIDIA's latest AI accelerators is also driving up memory prices. Jensen Huang, NVIDIA's CEO, reported that revenue per gigawatt of data center capacity has increased from $18 billion for Hopper to $25 billion for Blackwell and may reach $40 billion for the upcoming Vera Rubin platform.
As the value generated by each server escalates, customers have no alternative but to accept higher prices for HBM, a vital component.
NVIDIA highlighted that its revenue growth would have been considerably higher without supply constraints. CFO Colette Kress further emphasized that the company's 70% revenue growth forecast for the following year is contingent upon the products it can supply, while actual demand growth is nearing 100%. However, NVIDIA expects temporary margin compression due to higher memory costs, which is anticipated to recover by fiscal 2028.
As an unprecedented supplier-controlled market emerges, earnings forecasts for South Korea's leading memory chipmakers are being significantly revised upward. According to financial data provider FnGuide, Samsung Electronics is projected to achieve an operating profit of 394.41 trillion won, roughly $285.73 billion, this year, an 804.6% boost from the previous year.
By 2028, when the memory bottleneck is expected to peak, Samsung is projected to generate record revenue of 1,023.36 trillion won and operating profit of 563.38 trillion won. SK hynix is also poised for rapid growth, with its operating profit expected to surge by 464.0% year-on-year to 266.26 trillion won in 2026, subsequently rising to 414.76 trillion won in 2028.
Both companies are on track to set new records each year from 2026 through 2028, further solidifying their dominance in the global memory market.
Written by urgent.news from BusinessKorea's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.