Google just dropped a number that's reshaping the AI infrastructure landscape - the company's planning to spend up to $185 billion on capital expenditures this year, nearly double last year's outlay. The announcement sent Broadcom shares climbing 6% in after-hours trading Wednesday, while Nvidia gained 2%. The market reaction reveals something crucial about the AI hardware race: it's not winner-takes-all, and Google's custom chip strategy is making Broadcom a massive beneficiary.
Google just rewrote the playbook on AI infrastructure spending, and Wall Street's scrambling to figure out who wins. The company's bombshell earnings announcement Wednesday revealed plans to spend as much as $185 billion on capital expenditures this year - a staggering near-doubling from 2025 levels that's reshaping the semiconductor landscape in real time.
Broadcom shares surged 6% in extended trading, while Nvidia climbed 2%. But the real story isn't just the rising tide lifting all boats. It's about how Google's massive bet on custom silicon is creating a parallel universe to Nvidia's GPU dominance, with Broadcom sitting at the center of it.
"That is an incredible number. We are laughing because that number is so good for the Google cohort," Ben Reitzes, Melius Research head of technology research, told CNBC's Closing Bell Overtime following the release. The "Google cohort" he's referring to includes a tight circle of suppliers building the infrastructure for Google's AI ambitions.
Here's what most people miss: much of Google's cutting-edge AI work doesn't run on industry-standard Nvidia chips at all. The company's state-of-the-art Gemini 3 model was trained on tensor processing units, Google's proprietary chips designed specifically for machine learning workloads. And Broadcom is the critical partner helping Google design and manufacture these TPUs.
Broadcom has quietly built a burgeoning custom chip business focused on application-specific integrated circuits, or ASICs, which some experts believe can be more efficient than general-purpose GPUs for certain AI tasks. In December, Broadcom it would sell Google's TPU Ironwood rack systems to Anthropic, the AI lab behind Claude. The move essentially turned Google's internal infrastructure into a product line.
