I Was Wrong. Broadcom Is Far From Topping Out.

I keep hitting the buy button on Broadcom (NASDAQ:AVGO | AVGO Price Prediction), and I am here to admit my earlier bearish stance was wrong. A few months ago I was convinced the valuation had run too far ahead of the fundamentals. Then the rally paused, the multiple breathed, and the business kept doing what it has been doing for eleven straight quarters. Now the stock is moving in earnest again, and I would rather be early than embarrassed.

The thesis is simple. The data center build-out is lasting longer than even the optimists priced in last year. Hyperscalers keep raising their commitments. Anthropic, the customer everyone supposedly worried about, is ballooning toward positive cash flow and chronically short on compute. On the most recent call, Hock Tan said Anthropic went from “1 gigawatt of TPU compute” in 2026 to demand “in excess of 3 gigawatts” in 2027. That is one customer tripling its order book inside twelve months.

I’ll buy, even if it’s “overpriced”

Data centers are showing no sign of slowing down, so there’s no reason to be bearish on the demand.

AI semiconductor revenue ran from $4.40 billion in Q2 FY25 to $5.20 billion, then $6.20 billion, then $8.40 billion in Q1 FY26, the most recent quarter up 106% year over year. Q2 guidance is $10.7 billion. Tan now has “line of sight to achieve AI revenue from chips, just chips, in excess of $100 billion in 2027”, with supply secured through 2028. When a CEO locks up wafers, substrates, and high-bandwidth memory that far out, he is telling you the order book is real.

Then there is the cash machine underneath the AI story. Q1 FY26 generated $8.01 billion of free cash flow, 41% of revenue. Broadcom returned $10.9 billion to shareholders in the quarter, $7.8 billion in buybacks and $3.1 billion in dividends. The board authorized a fresh $10 billion repurchase program through December 31, 2026 and raised the dividend 10% in December, the 15th consecutive annual hike. Adjusted EBITDA margin sits at 68%. Compounders throwing off cash like that rarely stay still for long.

Note that there was a stock split after Q2 2024. Broadcom now has six custom XPU customers, including a fresh OpenAI engagement landing in volume in 2027 at over 1 gigawatt of compute capacity. The lead over any customer-owned tooling effort runs 12 to 18 months, and the SerDes and advanced packaging stack is hard to replicate quickly. Networking is becoming its own engine, growing 60% in Q1 and on track for 40% of AI revenue in Q2.

I wouldn’t go knee deep, though

Now the honest risk. Customer concentration is real. Six hyperscalers carry the AI story, and the OpenAI revenue miss report on April 28 reminded everyone what a single bad headline does to the whole AI complex. The forward P/E sits at 38, which leaves little room for a stumble. I have made peace with that because the contracts are multi-year, the supply is locked through 2028, and the cash flow is already on the page.

You should also note that there are better options out there. If you are going to buy an AI hardware stock, my first choice is still Nvidia (NASDAQ:NVDA). You can buy Broadcom to add a bit more diversification, but AVGO stock is still not cheap enough for it to be a bargain.

What keeps the buy button live is the gap between what management has signed and what the market will underwrite. The crowd on Polymarket gives Broadcom a 0.3% chance of being the largest company in the world by May 31. Fair enough. My own bet is smaller and longer. A business compounding AI revenue at triple digits, returning $10.9 billion a quarter to owners, and booked solid through 2028 is worth owning for the decade, and that is why I keep buying.