Queue for Chips: Google Leaves Its Researchers Without Computing Power
Google has built the largest AI infrastructure ecosystem over the past decade. The company launched a successful cloud business, developed its own TPU (Tensor Processing Unit) processors, and began leading the market as a primary alternative to NVIDIA chips. However, the success of this commercial strategy has triggered an internal crisis that Alphabet did not anticipate. This is reported by Ixbt.com reports .
According to Bloomberg, key researchers at Google DeepMind are now forced to wait in strict internal queues for computing resources. Most interestingly, these resources are being sold not to the company's own researchers, but to external clients—giants like Anthropic and Meta. For instance, Google has promised Anthropic access to 5 gigawatts of power and one million new-generation Ironwood chips over 5 years.
Such massive commercial commitments have completely occupied available data center capacity. Internal teams developing the Gemini model family are now joining waiting lists to conduct their experiments. Google DeepMind head Demis Hassabis noted that the shortage is hitting from two sides: first, there is a lack of memory chips from suppliers like Samsung and SK Hynix, and second, the allocation of internal resources is slowing down research speed.
Although Alphabet plans to spend $175–185 billion on infrastructure in 2026, market demand is growing faster than supply. The strategy of creating its own chips has proven economically beneficial, but the company is now unable to simultaneously serve competitors and conduct its own advanced research.
This strict rationing of resources has caused discontent within the scientific community. Over the past year and a half, a number of prominent scientists and engineers have left the DeepMind laboratory, including veteran employee Ioannis Antonoglou, due to the lack of computing power.
