[Edaily Choi Oh-hyun SOYEON KIM Reporter] As NVIDIA achieved record-high revenue for the 13th consecutive quarter, the company once again emphasized its collaboration with major memory manufacturers such as SamsungElectronics and SK hynix. This reaffirms that securing memory is a key factor determining AI chip production amid the explosive growth in investment in artificial intelligence (AI) infrastructure. With competition intensifying among global tech giants and others vying to secure memory, SamsungElectronics and SK hynix are expected to strengthen their position as leading suppliers.

According to industry sources on the 27th, NVIDIA posted revenue of $96.221 billion (approximately 133 trillion won) for the second quarter of fiscal year 2027 (May–July). This represents a 106% increase year-over-year and an 18% increase quarter-over-quarter, setting a new all-time record.

Notably, $89 billion (approximately 123 trillion won)—92% of total revenue—came from the data center segment, where demand for AI infrastructure is concentrated. This figure represents a 117% jump from a year ago. NVIDIA also projected third-quarter revenue to be around $108 billion (approximately 149 trillion won), exceeding market expectations. This indicates that investment in AI infrastructure shows no signs of slowing down.

Notably, this earnings report drew attention as the funds allocated to securing the supply chain more than doubled compared to the previous quarter. NVIDIA’s supply commitments for future product production surged 2.3-fold in just one quarter, rising from $119 billion (approximately 164 trillion won) last quarter to $279 billion (approximately 385 trillion won) as of the end of July. Of these commitments, $92 billion (approximately 127 trillion won) is concentrated in the second half of fiscal year 2027, while $87 billion (approximately 120 trillion won) and $88 billion (approximately 122 trillion won) are concentrated in the following two fiscal years, respectively. NVIDIA explained that these commitments are “primarily related to memory procurement.”

This means that even NVIDIA—a “super-powerful player” in the AI market—is entering into large-scale, long-term agreements to secure a stable supply of memory. As the bottleneck in the AI industry shifts to memory supply, the bargaining power of SamsungElectronics and SK hynix is expected to grow even further. Jensen Huang, CEO of NVIDIA, said during a conference call that day, “Saramin are coming to understand that it was quite a smart move for us to collaborate with memory suppliers from the early stages of the supply chain.”

Analysts suggest that the sharp rise in prices caused by the memory shortage has also impacted NVIDIA’s profitability. While NVIDIA’s non-GAAP gross margin for the second quarter remained unchanged from the previous quarter at 75.0%, it is projected to decline to 74.0% (±0.5%) in the third quarter. Even as revenue more than doubled over the past year, memory procurement costs have emerged as a new variable affecting profitability. Colette Kress, NVIDIA’s Chief Financial Officer (CFO), noted that memory prices have risen significantly more than expected and predicted they would rise further next year.

Market research firm TrendForce projected that HBM manufacturers will maintain their pricing power through next year. Analysts suggest that even if HBM bit shipments increase by 50–60% next year compared to this year, it will be difficult to keep pace with the rate of demand growth. Furthermore, as the expansion of HBM production encroaches on general DRAM production capacity, it could exacerbate supply shortages across the entire memory market, including server DRAM. This implies that the supplier advantage that began with HBM could spread to general-purpose DRAM as well.

The increased pricing power of memory manufacturers is expected to be reflected in the earnings outlooks for SamsungElectronics and SK hynix. According to the securities industry consensus, SamsungElectronics’ annual operating profit is projected to rise by approximately 39%, from 391 trillion won this year to 543 trillion won next year. SK hynix’s operating profit is also expected to increase by 47%, from 266 trillion won this year to 391 trillion won next year.

SamsungElectronics and SK hynix are expanding their production capacity to meet demand. SK hynix confirmed a total investment of 54.3 trillion won this month for projects including the Y2 fab at the Yongin Semiconductor Cluster and the construction of the M17 facility in Cheongju. Of this amount, 35.2 trillion won will be allocated to the Y2 fab alone, which is slated to serve as a production hub for next-generation DRAM, such as HBM. SamsungElectronics is also expanding its production capacity for 1c-nanometer DRAM and HBM4. SamsungElectronics has begun mass production and shipments of HBM4 and plans to increase the share of HBM4 in its HBM revenue to over 60% in the second half of the year.

The AI memory boom, which had been concentrated on HBM, is now spreading to SOCAMM and other technologies. This is because the market is expanding as AMD and Qualcomm—following NVIDIA—are turning to SOCAMM, a memory expansion solution based on low-power DRAM (LPDDR). SOCAMM refers to modules that reconfigure low-power memory—previously used in mobile devices such as smartphones—to suit server environments. It is a product primarily used in next-generation AI servers and is often referred to as the “second HBM.”

As the applications for OCAMM—developed at NVIDIA’s request—expand across the global Big Tech sector, it is expected to emerge as a new source of memory demand for SamsungElectronics and SK hynix. OCAMM2 is considered a memory solution optimized for inference servers.

SoCam 2 has already been optimized for NVIDIA’s next-generation artificial intelligence (AI) platform, “Vera Rubin,” and has entered mass production. In April, AMD announced plans to integrate SoCam 2 into “Verano,” its next-generation EPYC central processing unit (CPU) for servers, and Qualcomm is also expected to adopt SoCam 2 for its next-generation AI accelerators and solutions. In addition, the Joint Electron Device Engineering Council (JEDEC) is pushing for the standardization of LPDDR6-based SoCAM, making it highly likely that the scope of SoCAM applications will expand even further in the future.