Anthropic deepens compute stack with Google, Broadcom
Anthropic is expanding its infrastructure with a new partnership that adds multiple gigawatts of next-generation TPU capacity from Google and Broadcom, slated to come online starting in 2027. The company says the deal is meant to support explosive Claude demand, future model training, and more resilient delivery across its cloud and hardware footprint, with most of the new capacity based in the United States. The announcement frames compute supply as a strategic moat for frontier AI, not just an operational necessity.
Hot take: this is what AI competition looks like once the model layer matures - the real fight moves to securing power, chips, and supply agreements years ahead of demand.
- –Anthropic is signaling that Claude growth is already large enough to justify long-horizon infrastructure commitments.
- –The Google/Broadcom piece matters because it diversifies Anthropic beyond a single hardware path and reduces dependency risk.
- –This reinforces a broader industry trend: frontier labs are becoming infrastructure buyers as much as software companies.
- –The U.S.-heavy siting also reads like a strategic bet on domestic compute capacity and supply-chain control.
DISCOVERED
4d ago
2026-04-07
PUBLISHED
4d ago
2026-04-07
RELEVANCE
AUTHOR
DIY Smart Code