Broadcom rises on $100b AI forecast

Broadcom shares rose almost 3 percent after the company forecast over US$100 billion in AI chip sales next year, signaling potential gains in a market dominated by Nvidia.

Broadcom, a chipmaker and infrastructure software supplier, designs custom processors customers may use as alternatives to more costly Nvidia chips.

Melius Research analysts said Broadcom has visibility for about 10 gigawatts of AI demand in 2027 from clients including Anthropic and Meta, which equals the power needs of more than 8 million US households.

Hock Tan, CEO, said the company has secured capacity for leading-edge wafers and high-bandwidth memory through 2028 amid supply shortages of memory chips and limited capacity at AI processor manufacturer TSMC.

🔗 Source: Reuters

🧠 Food for thought

Implications, context, and why it matters.

Broadcom’s forecast rests on dominance in custom AI chips

Broadcom ties its outlook to a 70% to 80% share of custom AI accelerators, also known as application-specific integrated circuits (ASICs) 1.
Alphabet works with Broadcom to co-design chips, including the Tensor Processing Unit (TPU), rather than relying only on standard parts 1.
This approach helps hyperscalers (the biggest cloud computing companies) lower operating costs. Alphabet’s TPU could cut cost-per-token for AI models by 70% versus earlier versions 2.
Close work with Alphabet, Meta, and Anthropic on specific AI workloads can lock in longer contracts. It also sets Broadcom up in a separate slice of the market instead of a direct fight with Nvidia 1.

The cost of AI growth raises an industry-wide question

Investor unease comes from the same surge in revenue putting pressure on profits.
Broadcom guided to a 76.9% gross margin, down from 79% a year earlier, due to a larger share of lower-margin AI hardware plus rack-scale and custom deals 3.
One analyst said a recent $21 billion Anthropic order may bring little to no margin, since Broadcom may mainly pass through manufacturing costs 3.
That backdrop feeds worries that the AI hardware boom may not last in its current shape. Competition could still “re-commoditize” the market, which would shrink the outsized profits the sector has seen 3.