Google DeepMind CEO Demis Hassabis has warned that a critical shortage of memory chips is creating a major bottleneck in the development and deployment of artificial intelligence. Speaking with CNBC, Hassabis highlighted that the global memory market is dominated by just a few suppliers of key components, creating a significant constraint on AI progress. Despite Google’s internal advantages, including its custom-designed Tensor Processing Units (TPUs), Hassabis acknowledged that even the company is feeling the strain. “The whole supply chain is kind of strained,” he said. “We're lucky, because we have our own TPUs, so we have our own chip designs.” However, he emphasized that the challenge isn’t just about chips—it’s about the memory that powers them. AI models like Google’s Gemini require massive amounts of high-bandwidth memory (HBM) to operate efficiently, especially during training and inference. But HBM chips are in extremely high demand from AI hyperscalers such as Google, Meta, OpenAI, and others—driving up prices and creating long lead times. Meanwhile, traditional electronics manufacturers still rely on the same memory suppliers, leading to fierce competition for limited production capacity. Three companies—Samsung, Micron, and SK Hynix—control the vast majority of the global memory chip market. They are struggling to meet the surge in demand from AI companies without sacrificing their existing consumer electronics clients. The situation is further complicated by the fact that AI workloads require a different type of memory than what’s used in standard PCs or smartphones. Hassabis noted that the memory shortage isn’t just a logistical issue—it’s a research bottleneck. “It does constrain a little bit the research,” he said. “You need a lot of chips to be able to experiment on new ideas at a big enough scale that you can actually see if they're going to work.” This limitation affects not only Google but all major AI labs. The demand for computing power is so intense that Meta’s Mark Zuckerberg has reportedly told researchers that the two most important things they need are not just funding, but the fewest number of people reporting to them and the maximum number of chips possible. With AI infrastructure costs soaring, Google has signaled it’s preparing for a massive investment. On its fourth-quarter earnings call, the company projected capital expenditures between $175 billion and $185 billion for 2026—driven largely by its AI ambitions. Even with its own chip designs, Google cannot bypass the fundamental constraints of the global memory supply chain. As Hassabis put it, wherever there’s a capacity limit, there’s a choke point—and for now, that point is memory.