Memory chip prices have soared and become scarce due to the AI buildout, driven by two core needs: compute and memory. Compute hardware, such as Nvidia’s graphics processing units (GPUs), provides processing power for AI, and demand from Big Tech hyperscalers and major AI labs has made these chips coveted resources, leading to a GPU shortage and boosting rivals AMD and Intel. Google offers its Tensor Processing Units (TPUs) to select firms, including Meta, with which it announced a multibillion-dollar pact in February. Memory hardware stores the vast data AI models require, and advanced AI needs high bandwidth memory (HBM), a specialized form of dynamic random-access memory (DRAM) that stacks layers of DRAM on an AI GPU. Without enough memory, AI can hit a “memory wall,” where computing power increases but memory bandwidth lags, leaving setups idle. AI hyperscalers are set to spend $750 billion this year and $1 trillion in 2027, according to Goldman Sachs.