✦ Cloud, TPUs & AI InfrastructureNarrow moat

Google Cloud spent a decade as the also-ran — a distant third behind AWS and Azure, roughly an eighth of the market by the share trackers³ — and then the AI cycle handed it the one card the leaders do not hold. Revenue growth accelerated to 82 percent while the backlog swelled to $514 billion¹, the fastest growth of any hyperscaler by a wide margin.

The card is the TPU. Google has designed its own AI silicon since 2015, and for years that read as an internal cost saving. In October 2025 it became a product: Anthropic contracted for up to one million TPUs and more than a gigawatt of capacity, a deal worth tens of billions² — the first proof that a frontier lab will bet its training runs on non-Nvidia silicon at scale. Gemini itself trains on TPUs; the sales pitch is simply Google's own infrastructure, rented out.

The honest caveats: an eighth of the market is still an eighth, AWS and Azure hold the enterprise relationships, and every hyperscaler now builds custom chips — the TPU's edge is a head start, not a patent. But a third-place cloud growing at twice the leaders' pace, holding half a trillion in contracted work and the only proven Nvidia alternative an outsider can buy, is a different asset than the one the market filed away years ago. Watch the backlog: it nearly doubled in three months, and it converts to revenue on a schedule.