SK Hynix hits 100 trillion won revenue as AI memory demand spikes

Record-breaking quarterly results

SK Hynix reported a massive financial milestone in its third quarter, with revenue hitting 104.6 trillion won, which is approximately $78 billion. This marks the first time the South Korean memory giant has surpassed the 100 trillion won revenue threshold in a single quarter. The company's growth was fueled by a significant surge in demand for memory components used in artificial intelligence.

The company's profitability also reached extraordinary levels, with operating profit jumping 632% year-over-year to 83.3 trillion won. This performance resulted in an 80% operating margin, meaning the company earned about 80 won of operating profit for every 100 won of sales. These figures significantly exceeded the 77.5 trillion won operating profit that analysts at Bloomberg had forecasted.

Despite the surge in core chip sales, the company's net income of 39.8 trillion won was lower than the previous quarter. This was due to 32.6 trillion won in net non-operating expenses, which included a 22.9 trillion won investment valuation loss and an 11.6 trillion won currency loss. These specific hits depressed the final net profit even as the primary business of selling memory chips thrived.

The AI memory supply crunch

The primary driver behind these numbers is a severe imbalance between the demand for AI memory and the current global supply. SK Hynix reported that customer volume requests are currently more than double what the company is able to supply at this time. This shortage is particularly acute for DRAM, the standard memory used in phones and PCs, and HBM, which is the ultra-fast stacked memory used to feed AI models.

The company expects this supply constraint to persist for a long time, with projections extending through 2030. CFO Kim Woo-hyun explained that the complexity of leading-edge processes and the long time required to build new production facilities make it difficult to scale quickly. Because of these constraints, some customers are already looking for ways to stretch their limited chip supplies by adjusting HBM specifications.

To manage this scarcity, SK Hynix has moved to secure its future production through long-term agreements with 14 key customers. These contracts, which typically last five years, already reserve more than half of the company's planned production for the years 2027 and 2028. These deals include specific price bands and safeguards like deposits to ensure a steady flow of revenue and a clear allocation of scarce wafers.

Broadening AI infrastructure spend

While some market skeptics have raised concerns about whether AI investment will lead to commercial success, SK Hynix leadership remains optimistic. CEO Kwak Noh-Jung pointed to the fact that major cloud customers are reporting year-over-year revenue growth between 30% and 80%. Furthermore, these customers are maintaining order backlogs that are two to five times larger than previous levels, suggesting a sustained appetite for infrastructure.

The company also notes that AI spending is moving beyond just the largest tech companies, often called hyperscalers. It is now expanding into non-tech firms, sovereign projects, and on-premise deployments. This diversification helps reduce the risk that a single buyer slowing down their spending would significantly impact the company's overall growth trajectory.

However, the environment for building this infrastructure is becoming more difficult as financing tightens. Elon Musk recently noted that debt markets have largely closed to firms that do not have significant non-AI revenues. This suggests that while the demand for the chips themselves is high, the companies building the data centers may face more hurdles in securing the capital needed to fund those projects.

Future production and outlook

Looking toward the next few years, SK Hynix anticipates a major increase in investment activity starting in 2027. This is expected to coincide with the first phase of its Yongin complex reaching full-scale operation. This facility will be a critical component in addressing the long-term supply issues that are currently squeezing the market for high-performance memory chips.

The company expects a shift in the broader market as well, specifically regarding the demand from PCs and smartphones. Because of higher memory prices and existing supply constraints, SK Hynix predicts that demand from these traditional devices will likely soften. However, the company believes that the massive boom in AI infrastructure will more than compensate for any decline in the consumer electronics sector.

For users and developers, these trends mean that the cost of high-performance computing will remain elevated for the foreseeable future. The long lead times for new fabrication plants mean that the supply of DRAM and HBM will stay tight. This creates a situation where the underlying costs for cloud-based AI services and on-premise AI deployments will remain high and potentially volatile as the industry tries to keep up with the current demand.