Google Releases 8th-Gen TPUs and Gemini Enterprise Agent Platform
What Happened
Google announced new custom AI silicon and the Gemini Enterprise Agent Platform, integrating hardware, models, and tooling for corporate buyers. The launch features Gemini 3.1 variants and over 200 available models. Google leadership disclosed roughly 3.8 million TPUs and 1.3 million GPUs — about 25% of global AI compute.
My Take
Google is the only frontier player whose moat is physical: chips, fiber, data centers, and a decade of operational scale none of its rivals can replicate quickly. Bundling TPUs with an agent platform is a classic Microsoft-in-the-90s play — make the substrate so cheap and integrated that the model layer commoditizes around it. If you're a CIO, the boring rational choice in 2026 is "default to Gemini and pay the switching cost only when a specific workload demands otherwise."
Read Original Source