Broadcom hops on AI gravy train with switching chips for hyperscalers

Broadcom announced a big revenue bounce in its first quarter of 16% year-on-year to $8.9 billion, pulling strength from AI networks in data centers at hyperscalers, similar to how Nvidia has benefited with its GPUs used for AI training and inferencing.

Generative AI has created an “urgency and excitement” with  hyperscaler customers who want low latency networks that can scale over Ethernet, CEO Hock Tan said on a call with analysts on Thursday.

“We are seeing… an urgency in hyperscaler customers to secure products with the ability for low latency networks that can scale [with] Ethernet,” he said. “It’s still early innings, but customers don’t want to miss out or be late in this [generative AI] trend.”

In many data center setups, thousands of GPUs for generative AI must run in parallel to grind through huge databases, requiring efficient networks.  New switches in Ethernet are needed that are dedicated to such workloads, and Broadcom is investing in such work, Tan said.

Broadcom’s Tomahawk 5 switch chips are expected to ship in 2024, with double the capacity of the Tomahawk 4 but the same power.  Earlier this week, Broadcom introduced Bailly, a co-packaged optics prototype networking switch.

Tan said revenues from its Ethernet switching in AI in 2022 was $200 million, with an “exponential demand from hyperscalers” to reach $800 million in 2023. “We expect the trend to continue.”  While networking for AI so far is not a large share of its overall revenues, almost the entire Broadcom earnings call was devoted to the AI subject.

Compared to overall  first quarter revenues of nearly $9 billion, all Broadcom networking products were $2.3 billion, up by 20% year-on-year, Tan said.   He said he expects the total to reach $3 billion over all of 2023.

Increased investments in AI which rely on chips used in data centers helped Broadcom, similar to the way Nvidia benefited from the OpenAI ChatGPT phenomenon.  Nvidia makes GPUs,  and 10,000 such chips were used to train the ChatGPT application.

RELATED: Nvidia’s happy to share AI stage with ChatGPT after strong Q4

RELATED: Update: ChatGPT runs 10K Nvidia training GPUs with potential for thousands more

Broadcom sees the second quarter ending April 30 as positive as well as its first quarter, up by 8% over a year ago, reaching about $8.7 billion.

Free cash flow in the quarter was also up by 16%, to $3.9 billion. Even stronger free cash flow is expected in the second quarter, said CFO Kirsten Spear.

First quarter revenues were in line with Wall Street expectations, and earnings per share of $10.33 came out ahead of the consensus estimate of $10.17.  Shares were up about 1% to $598.65 at close.

Tan said he expects the timelines for government reviews of its purchase of VMWare to be extended but with the $61 billion deal closing in 2023.