Silicon Valley’s ongoing race for artificial intelligence (AI) dominance has hit a massive bottleneck: Computing power. Even tech giants like Google and Facebook-parent company Meta are also feeling its heat. A report has claimed that Google has officially placed strict limits on Meta’s access to its premier Gemini AI models after Mark Zuckerberg’s social media giant tried to buy up more computing capacity than Google could physically provide.Citing people familiar with the matter, The Financial Times reports that Google informed Meta that it does not have the server space or computing capacity to fulfill Meta’s massive orders. The squeeze is a direct result of the explosive global demand for “inference workloads” – the computing power required to actually enable AI chatbots, coding assistants and automated agents to quickly respond to tasks assigned to them.Even Google’s multi-billion-dollar network of data centers is buckling under the weight of the AI boom.“Obviously, we are compute-constrained in the near term. Our Cloud revenue would have been higher if we were able to meet the demand,” Google CEO Sundar Pichai acknowledged at the company’s first-quarter earnings in April, noting that Google Cloud’s backlog of signed-but-undelivered contracts recently doubled to a staggering $460 billionTo survive the hardware drought, Google recently signed a massive $920 million-a-month deal to lease extra computing capacity from Elon Musk’s SpaceX network. Rival AI lab Anthropic reportedly signed a similar emergency capacity deal with SpaceX last month.
Meta forced into delays and token rations
Google’s computing cap may have already disrupted and delayed Meta’s highly anticipated plans to release its newest AI technology to outside developers. Unlike Google or Microsoft, Meta does not own a commercial cloud computing business. While Zuckerberg has committed to spending a $600 billion on US data centers by 2028 to build what he calls a “personal superintelligence,” the company currently remains dependent on its rivals.Meta had been quietly using Google’s Gemini to run its own internal operations, which include automated customer service chatbots and AI systems that root out scams and harmful content, because Gemini outperformed Meta’s own open-source “Llama” models, the report said.Because of Google’s capacity limits, Meta has been forced to ration its usage, strictly ordering its engineers to be hyper-efficient with “tokens” (the digital units used to measure AI processing). Earlier this month, Meta even delayed plans to release its newest AI model to developers.