Broadcom Built Something Nearly Perfect So I Keep Buying
Hock Tan engineered a semiconductor company that profits whether hyperscalers bet on GPUs or custom silicon, and the demand already booked stretches years into the future. Here is why that structure keeps pulling me back to the buy button.
I keep hitting the buy button on Broadcom (NASDAQ:AVGO | AVGO Price Prediction) because Hock Tan has built the rarest thing in semiconductors: a company that wins whether the AI buildout runs on GPUs or on custom silicon. Every rack needs interconnects. Every cluster needs Ethernet fabric. Every hyperscaler eventually needs a private cloud that runs the workloads. Broadcom sells all three, and I am not clever enough to bet against that setup.
Why the Buy Button Stays Active
Start with what Q2 fiscal 2026 actually showed. Revenue hit $22.2 billion, up 48% year-on-year, with AI semiconductor revenue at $10.8 billion, up 143% year-on-year. Networking alone drove almost 40% of Q2 AI revenue. That is the interconnect story in one line: whether a hyperscaler is buying GPUs or XPUs, the fabric holding those clusters together comes from Broadcom’s Tomahawk 6, Jericho 4, and 1.6 terabit DSPs. Tan told the call, “Networking is key to building scalable XPU and GPU clusters”, and “We have at least one generation of technology and product leadership.”
Then look at the forward book. Q2 AI bookings came in at over $30 billion, compared with $10.8 billion shipped during the quarter. Management now sees visibility all the way to 2028, with reiterated fiscal 2027 AI semiconductor guidance in excess of $100 billion. Tan’s phrase for it: “Demand for XPUs and networking is simply insatiable.”
Cash, Dividends, and a Software Moat People Ignore
The financial engine backs the growth story. Q2 free cash flow was $10.3 billion, or 46% of revenue. Broadcom has raised its dividend for 15 consecutive years, with the most recent bump a 10% increase in December 2025 to $0.65 per quarter.
The overlooked piece is VMware. Infrastructure software revenue was $7.2 billion, up 9% year-on-year, with ARR growth of 17% year-over-year and operating margin around 79%. VCF 9.1 now supports heterogeneous compute across AMD, Intel, Nvidia GPUs and CPUs. That is a subscription annuity attached to the same customers buying the silicon.
Why Not the Obvious Alternatives
Readers will ask why I do not just own NVIDIA (NASDAQ:NVDA) or Advanced Micro Devices (NASDAQ:AMD). I own some NVIDIA. What I like about Broadcom is that its custom XPU relationships with Google, Meta, OpenAI, and Anthropic pay off exactly when hyperscalers try to diversify away from Nvidia GPUs, while its networking silicon still sells into every GPU rack Nvidia ships. Broadcom’s forward P/E of 19x also gives me a valuation I can live with. As for Marvell Technology (NASDAQ:MRVL), it plays in the same custom silicon lane, but Broadcom’s Q2 AI revenue of $10.8 billion operates at a scale Marvell has not reached.
Risk Worth Watching Closely
Customer concentration is real. A handful of hyperscalers drive most of the AI silicon revenue, and if two of them push in wafer orders or shift more work to Nvidia GPUs, the growth curve bends. What keeps me buying anyway is that Broadcom sits inside the networking layer of those same GPU clusters, and Q2 non-AI semiconductor bookings of over $6 billion point to the cyclical business finally coming back.
What Keeps Me Adding
Broadcom is compounding cash flow at 46% of revenue, raising the dividend every year, and has booked demand it cannot ship until 2028. That is why my buy button stays warm.
Contact [email protected] for any questions or corrections.







