Starting $AVGO on the Backlog, Not the Hype
Opened Broadcom at $321.00. Also a starter, 0.9% of the portfolio.
This one is about the backlog more than the narrative. AVGO was around $446 in May and $386 in July. At $321 it's roughly 28% off that high while the business numbers went the other way.
Q3 was the quarter that made me move. AI semiconductor revenue more than tripled year over year to $16.7B. Six XPU customers now: Google, Meta, OpenAI, Anthropic, and two others. $73B in contracted AI backlog. Management is guiding to over $100B in AI semi revenue for FY2027.
The specifics behind that number are what make it credible to me rather than a slide. Ironwood TPU v7 shipping in volume to both Google and Anthropic. TPU v8i in production for Google. OpenAI's first custom accelerator already shipped, with 1.3GW planned for 2027 and line of sight to over 5GW in 2028. Hock Tan said on the call that Anthropic is on track to be their largest XPU customer in 2027.
Custom chip contracts lock in for 3 to 5 years because migrating a design is a multi-year engineering project. That's the moat. It's not sentiment, it's switching cost.
The risk I'm watching: Google is building a second supply source, splitting TPU v8 work and bringing MediaTek and Marvell into the picture. Google is Broadcom's biggest custom silicon customer, so that's a genuine threat to the core relationship, not a side issue. Also worth saying out loud that if general purpose GPUs claw back workload share from custom silicon, the 2027 mix gets worse.
Starter size for the same reason as BKNG. I'd rather build into this over time than go in heavy on a stock that's already run hard over three years.
Next earnings December 10.
NFA. DYOR.
