AI Applications Drive Demand for Advanced Memory as HBM4 and HBF Emerge as Inference Mainstays
The rapid expansion of generative AI and inference applications is increasing demand for advanced logic processes and high-bandwidth memory (HBM) in data-center accelerators. The Market Intelligence & Consulting Institute (MIC) at Taiwan’s Institute for Information Industry said memory performance has become a major bottleneck for high-performance computing platforms, affecting AI model processing speeds, power consumption and overall computing efficiency.
MIC’s latest analysis forecasts that HBM4 will enter mass production in 2026, further improving bandwidth and integration with computing chips. In the future, high-bandwidth flash memory (HBF) could work alongside HBM to provide the capacity and high-speed access required for AI inference. Reports disclosed neither an investment amount nor a more specific month for the start of mass production.
All Coverage
1 original reportsThe Backstory
The history behind this eventSurging AI Demand Drives Continued Advances in HBM Technology
Since OpenAI launched ChatGPT in November 2022, the adoption of AIGC and large language models has accelerated. AI training and inference require frequent movement of vast amounts of data, making high-bandwidth memory (HBM) a critical component for increasing throughput and easing data-transfer bottlenecks in GPUs and other high-performance computing chips.
The latest industry developments show that demand for AI memory continues to rise. HBM upgrades will focus on increasing capacity per chip and boosting transfer bandwidth to support larger models and computing workloads. The reports disclosed no investment figures for specific vendors, shipment data or precise publication date, making it impossible to quantify the scale of the current increase in demand.
Subscribe to Mark Radar Weekly
Every Friday, the week's strongest signals in your inbox. Unsubscribe anytime.