Broadcom climbs 6% after Google signals higher AI spending

Broadcom shares rose 6% after Google reported higher capital expenditures for AI in its earnings release.

Google expects to spend up to US$185 billion on capital projects this year, nearly doubling its spending from the previous year.

The company’s AI software runs on its own tensor processing units (TPUs), which Broadcom helps manufacture.

Experts note that such custom chips are typically used by large technology firms, known as hyperscalers, including Google, Microsoft, Amazon, and Meta.

Broadcom also sells Google’s TPU Ironwood rack systems to other AI labs like Anthropic.

While Google uses TPUs, it also relies on Nvidia chips, which rose 2% in after-hours trading.

Analysts suggest that increased AI chip demand benefits both Broadcom and Nvidia, as larger firms continue to develop custom hardware for AI workloads.

🔗 Source: CNBC

🧠 Food for thought

Implications, context, and why it matters.

Broadcom builds custom chips for hyperscalers as a multi-billion dollar business

Broadcom works with Google as its silicon implementation partner. It turns TPU architecture into a manufacturable application-specific integrated circuit (ASIC) layout. It also supplies high-speed serializer/deserializer (SerDes) links. It handles packaging plus post-fabrication testing 1.
Broadcom’s AI-related revenue reached $20 billion in fiscal 2025, up 65% year over year 2.
Broadcom’s custom AI chip work also includes Meta and OpenAI 3.

Custom chips reduce reliance on Nvidia

Hyperscalers pursue custom silicon to reduce supply risk. They also tune AI software to their hardware, which can outperform general-purpose chips 3.
CNBC wrote that Google trained Gemini 3 on Google TPUs rather than Nvidia graphics processing units (GPUs) 4.
Demand for AI chips keeps growing faster than supply 5.
Nvidia relies on its CUDA software platform. It raises the effort required for developers to move to another stack like Google’s XLA-based TPU stack 1.