The demand for memory has surged as AI applications proliferate. Analysts expected Micron to meet this rising need. However, the context length for AI models is growing at a rate of 30x annually. This rapid increase has caught Micron off guard.

Micron’s recent efforts include the release of a 245TB SSD, which significantly reduces storage footprint by over 80%. The company also developed its HBM4 product, boasting more than double the bandwidth of HBM3e. Yet, these advancements have not alleviated supply shortages across the entire memory hierarchy from HBM to SSDs.

Experts point out that AI’s demand for memory is driven by the need for KV cache. This has become critical for efficient AI inference. Jeremy Werner, a prominent industry voice, stated, “If you can’t store context, compute balloons at a squared rate.” He emphasized that training and inference use memory in fundamentally different ways.

The inference era of AI is just beginning, and many applications are yet to be deployed. Werner highlighted that the industry is not building enough fabs worldwide to keep up with this demand. He remarked, “I really believe we are only scratching the surface of the changes AI will bring.” This sentiment reflects broader concerns within the industry about future capacity.

The challenges faced by Micron are not isolated. Other major players like Nvidia and Intel are also grappling with similar issues related to GPU utilization and data center demands. The competition in this space is intensifying as companies race to enhance their capabilities.

As Micron continues its expansion efforts, it remains uncertain whether these new fabs will be sufficient to meet future demands. The entire landscape of AI applications relies on advancements in memory technology, making this a pivotal moment for all stakeholders involved.

The post Scratch: Micron’s Memory Production Struggles Amid AI Demand Growth appeared first on 1News.