Recent reports suggest SK hynix's memory inventory has fallen to a critically low four-week level, signaling a major power shift to suppliers in the semiconductor market.
This situation didn't happen overnight; it's the direct result of the explosive demand for AI. The core of the issue lies in a production bottleneck created by prioritizing a specific type of memory. First, the AI industry's growth requires immense amounts of high-performance HBM (High-Bandwidth Memory). Giants like Google and Nvidia are racing to secure it, with Google DeepMind's CEO even calling memory a near-term 'choke point' for AI development.
Consequently, memory manufacturers like SK hynix are dedicating more of their production capacity to HBM, which is far more profitable than conventional memory. Second, HBM is incredibly complex to produce, consuming substantial wafer and advanced packaging resources. Third, with limited new factory (cleanroom) space available in the short term, this focus on HBM directly reduces the capacity for producing the DRAM and NAND chips used in everyday devices like PCs and smartphones. This creates a supply squeeze for the entire memory market.
The market is already feeling the impact. TrendForce, a market research firm, recently issued its most aggressive forecast ever, predicting that conventional DRAM contract prices could surge by 90-95% in the first quarter of 2026 alone. This aligns with SK hynix's own statements about its capacity being completely sold out and its plans to accelerate the construction of new fabs to meet overwhelming demand.
While the specific 'four weeks' inventory figure is from secondary reporting and not officially confirmed by SK hynix, the surrounding evidence is overwhelming. The combination of record price hikes, sold-out capacity, and urgent fab expansions all point to the same conclusion: the memory market is exceptionally tight. This seller's market is expected to persist through at least the first half of 2026, keeping prices high for everyone.
- HBM (High-Bandwidth Memory): A high-performance type of memory essential for AI processors, which processes large amounts of data very quickly.
- DRAM (Dynamic Random-Access Memory): The main memory used in most PCs and servers to store data that the processor is actively using.
- NAND Flash: A type of storage memory used in SSDs, smartphones, and USB drives for long-term data storage.