Google said this week that Gemini crossed one billion monthly users faster than any product in the company's history, according to Ars Technica's August 12 report. ChatGPT hit the same billion-user mark around the same time, per The Verge, so both now sit at roughly the same audience size. But the number that matters is what's underneath, not the user count. OpenAI serves ChatGPT off Nvidia GPUs rented through Microsoft Azure. Google serves Gemini largely off TPUs (Tensor Processing Units, chips Google designs in-house rather than buying from Nvidia) running in its own data centers. That means every Gemini query answered is a query Nvidia doesn't get paid for and Google doesn't have to queue behind anyone else's GPU allocation to serve.

The billion-user headline reads as a horse race between two chatbots. The real story is a supply chain question: who owns the silicon under a billion daily queries. Nvidia's H100 and B200 chips are the bottleneck every other AI lab (OpenAI, Anthropic, Meta) has to buy around, at prices and lead times Nvidia sets. Google doesn't have that constraint because it started building its own TPU fleet in 2015, long before this demand existed. That's why Google can absorb a billion-user product without asking Nvidia for anything, while OpenAI's growth to the same scale runs through Azure's GPU capacity and Microsoft's own build-out schedule. Watch Google's next TPU generation announcement, expected alongside its Cloud Next event, for the number that will tell you how much of that headroom is actually available to rent to everyone else.