Memory shortage to worsen in 2027, ease only by H2 2028—Silicon Motion CEO

Wallace Kou, president and CEO of Silicon Motion, has issued a stark forecast for the global memory market over the coming years. He predicts that the current memory shortage will worsen throughout 2027, with relief unlikely to arrive until the second half of 2028 at the earliest. Beyond that timeline, Kou expects memory prices to remain elevated for the next two to three years, underscoring the depth and duration of the supply-demand imbalance. The shortage is being driven by AI inference—the computational step in which trained machine-learning models process user inputs and generate outputs—which demands substantial memory capacity and throughput. As companies race to deploy and scale AI systems, the appetite for memory chips has outpaced manufacturers' ability to expand supply, creating a bottleneck that Kou's forecast suggests will take years to resolve.

The forecast from Wallace Kou, a leading figure in memory controller design, reflects structural strain in the semiconductor supply chain driven by accelerating AI adoption. As companies scale inference workloads—the computationally intensive process of running pre-trained AI models on user queries—demand for memory capacity and bandwidth has exceeded manufacturing capacity. The CEO's timeline suggests this imbalance will persist well into 2028, implying that memory will remain a binding constraint on AI infrastructure expansion rather than a temporary bottleneck. The expectation of elevated prices over two to three years signals that supply recovery will lag demand, creating persistent cost headwinds for cloud providers, semiconductor manufacturers, and enterprises building AI systems.