Google's $185B AI Bet Lifts Broadcom and Nvidia

Google just told the market it could spend as much as $185 billion on capital expenditures this year, nearly double last year. Stocks reacted like someone handed out free GPUs: Broadcom jumped about 6% after hours, Nvidia rose ~2%. Simple math - more datacenters, more chips, more revenue for the supply chain.**AI finds religion:**Google is doubling down on custom hardware (TPUs) instead of relying only on off-the-shelf Nvidia silicon. Broadcom is in the middle of that build, helping design and ship the TPU racks Google uses for big models.

• $185B: Google’s potential capex haul for 2026, focused on AI datacenters.
• Broadcom’s role: partner on TPU systems and seller of TPU Ironwood racks (Anthropic bought some). Broadcom calls its custom efforts "XPUs."
• Market reality: custom ASICs make sense only for hyperscalers - the handful of firms with billions to burn and billions to train.**Is it time to freak out?**Not unless you think every company will build its own chips. For Broadcom, this is a very big deal - custom chip work is higher-margin and sticky, and hyperscalers need partners that add IP and get silicon made.

For Nvidia, this isn’t a cold shoulder. Google still uses Nvidia chips alongside TPUs, and a boom in datacenter buildouts generally fattens demand for accelerators, networking gear, and everything in between. The short version: Broadcom gets a rarer, lucrative slice of the custom-chip pie, Nvidia keeps selling the mainstream slices, and the hyperscalers win the compute arms race.

If you sell to big cloud builders, start practicing your elevator pitch and dust off those NDAs. This is where the real money lives.