Google employees are now competing with Anthropic and Meta for access to Google compute
Google built its reputation as a paradise for ambitious researchers: a place where smart people got massive resources and freedom to experiment.
But in the AI era, the physical infrastructure that powers those breakthroughs is maxed out, and even Google’s own employees are reportedly struggling to get enough computing power.
According to Bloomberg, the bottleneck comes down to hardware. Google’s custom-built AI chips — tensor processing units, or TPUs — are in such high demand that internal researchers say they’re effectively competing for rack space against massive, paying cloud customers like Anthropic and Meta. Frustrated by the bureaucracy of fighting for server time, top engineers are jumping ship to launch their own startups, arguing they can secure more reliable access to infrastructure on the open market than inside the company that actually builds it.
In other words: Google became so successful at selling AI infrastructure that its own researchers now have to justify experimental projects against revenue-generating workloads and a more than $460 billion backlog of paying tenants.
According to Bloomberg, the bottleneck comes down to hardware. Google’s custom-built AI chips — tensor processing units, or TPUs — are in such high demand that internal researchers say they’re effectively competing for rack space against massive, paying cloud customers like Anthropic and Meta. Frustrated by the bureaucracy of fighting for server time, top engineers are jumping ship to launch their own startups, arguing they can secure more reliable access to infrastructure on the open market than inside the company that actually builds it.
In other words: Google became so successful at selling AI infrastructure that its own researchers now have to justify experimental projects against revenue-generating workloads and a more than $460 billion backlog of paying tenants.
