Broadcom sees $22b Q2 revenue, AI chips drive growth

Broadcom forecast fiscal Q2 revenue of about US$22 billion and announced a share buyback of up to US$10 billion, but the guidance drew a tepid market reaction.

The fiscal Q2 ends May 3, with analysts averaging US$20.5 billion and some forecasts above US$22 billion.

Shares were little changed in late trading and down 8.3% in 2026 through the close.

In fiscal Q1, sales rose to US$19.3 billion and adjusted EPS was US$2.05, versus analysts’ estimate of US$2.03.

CEO Hock Tan said Broadcom has increasingly tied its business to AI by making custom chips for firms including OpenAI and Anthropic PBC.

The company shipped the first units of a new processor generation, expected to be adopted by about half a dozen more clients this year.

🔗 Source: Bloomberg

🧠 Food for thought

Implications, context, and why it matters.

Anthropic’s $21 billion Broadcom commitment signals the scale of its AI push

Anthropic PBC committed $21 billion across two orders, one for $10 billion and another for $11 billion. CEO Hock Tan said Broadcom is shipping full “TPU Ironwood racks” (rack-level systems) rather than only chips 1.
Broadcom said it now has five customers for custom AI chips. The newest customer placed a $1 billion order during the quarter, though Broadcom did not name it 1.
Custom silicon brings tradeoffs. Buyers must lock in designs years ahead, which can limit flexibility compared with Nvidia’s general-purpose graphics processing units (GPUs). Broadcom’s fiscal Q2 revenue outlook of about $22 billion still drew a muted market response 2.

Custom AI chips are opening new lanes for Broadcom

Hyperscalers (the biggest cloud computing providers) including Alphabet (Google) and Meta are pairing GPUs with custom chips built for specific workloads, which expands Broadcom’s addressable market 3.
Source material does not back the claim that over 70% of data center revenue now comes from inference. It does describe a shift from training clusters toward large-scale inference (running AI models in production), a change that can favor specialized hardware 3.
OpenAI is taking a mixed approach. Reporting cited by Tom’s Hardware links its custom chip work to Broadcom, yet OpenAI still relies on Nvidia hardware plus the CUDA software ecosystem (Nvidia’s developer platform for programming GPUs) 2.
One source asserts that Saudi Arabia and Japan are building “sovereign AI” clouds using Broadcom’s technology, though other provided materials do not confirm it. Treat it as commentary, not a Broadcom-disclosed revenue stream 3.