DeepMind's AI talent exodus blamed on chip shortages, a fundamental conflict of interest, and Google bureaucracy

After the departure of several prominent AI researchers at Google DeepMind, sources have revealed background details. According to Semafor, CEO Demis Hassabis has withdrawn from day-to-day business for about a year, handing operational tasks to the new DeepMind chief Koray Kavukcuoglu. He was not pushed out but found management unfulfilling and saw himself more as a visionary scientist. CNBC reports that many researchers are frustrated by restricted access to Google's TPU chips, while Google Cloud sells the same chips to competitors like Anthropic, which are exactly the rivals Google's researchers are supposed to beat. Google plans its computing capacity years in advance for areas like research, product operations, and cloud customers, but priorities can shift short-term. Google's bureaucracy also makes younger companies more attractive. Coincidentally, Google announced today that the 'excited frontier AI laboratory' Mirendil will use over $100 million in TPUs and Nvidia GPUs via Google Cloud as part of a partnership, which is the same computing capacity that DeepMind researchers complain is not internally available.