Google Deepmind CEO says the memory shortage is creating an AI 'choke point'

The ongoing memory chip shortage is proving to be a formidable challenge for the entire tech industry, including giants like Google. As artificial intelligence companies compete for increasingly scarce memory resources, the situation has escalated to a critical level. Demis Hassabis, CEO of Google DeepMind, spoke with CNBC about the physical limitations that are hindering AI deployment. He noted that demand for Google’s Gemini and other AI models far exceeds supply. "This demand is constraining not just deployment but also research efforts," he explained. Hassabis emphasized that substantial quantities of chips are essential for testing new ideas effectively. The competition for memory chips is fierce, with researchers from various tech powerhouses, including Google, Meta, and OpenAI, all vying for access. According to Mark Zuckerberg, AI researchers prioritize two key resources: minimal oversight and maximum chip availability. Hassabis identified the existence of a "choke point" wherever capacity is limited. Despite having a distinct advantage with their Tensor Processing Units (TPUs), which Google designs for internal use, the company still faces challenges within the competitive memory market. Hassabis pointed out that the overall supply chain remains under considerable strain. The memory chip market is dominated by three key manufacturers: Samsung, Micron, and SK Hynix. These companies are currently struggling to balance the overwhelming demand from AI developers while maintaining their commitments to traditional electronics clients. Furthermore, the type of memory chips preferred by AI companies, specifically high-bandwidth memory (HBM) chips, differs from what standard PC manufacturers require. Looking ahead, Google does not anticipate a decrease in its investments in AI infrastructure and memory chips. During a recent earnings call, the company projected capital expenditures reaching between $175 billion and $185 billion by 2026.