The Meta-Broadcom announcement contains a number that is easy to overlook and provides more information than the press release could. A gigawatt. About 750,000 American homes could be powered by this amount of electricity. That’s only the initial stage. The floor, not the ceiling.
The market’s response to Meta and Broadcom’s announcement in mid-April that they were extending their custom chip partnership through 2029 was almost comically uneven. In extended trading, Broadcom’s shares increased by roughly 3.5%. Meta hasn’t really moved. It was difficult not to interpret that split as Wall Street subtly acknowledging something it doesn’t always say aloud: the company creating the chips might be more important than the company purchasing them.
For many years, Nvidia dominated the narrative surrounding AI hardware. The company founded by Jensen Huang turned into a toll booth on every machine learning road. However, beneath all that cacophony, a more subdued change is taking place. To avoid Nvidia’s pricing, the wealthy giants Meta, Google, and Amazon have been producing their own silicon. When they want to do it, they keep phoning Broadcom. People tend to undervalue that aspect. The AI race doesn’t have to be won by Broadcom. All it needs to do is continue to sell shovels to all of its operators.
The person to watch in this situation is Hock Tan, the CEO of Broadcom. As part of the agreement, he will leave Meta’s board and take on an advisory position regarding the company’s custom chip strategy. That sounds like corporate housekeeping on paper. In actuality, it seems more intentional—a man eliminating the appearance of conflict while keeping his hands exactly where the work is being done. In the same way that other executives gather press mentions, Tan has spent years compiling these design contracts. Through 2031, Meta, Google, and Anthropic will have a deeper partnership measured in gigawatts. He may have founded the most significant AI company that most people are still unable to adequately describe at a dinner party.
The financials don’t dispute this. With AI chip revenue alone at $8.4 billion, more than twice as much as the previous year, Broadcom reported fiscal first-quarter revenue of $19.3 billion, up 29% from the same period last year. Free cash flow is close to $8.2 billion. a new authorization for a $10 billion buyback. These are not the figures of a business that hopes the AI narrative continues. These are the numbers of a person who has already accumulated on it.
And yet. What keeps seasoned investors up at night is the valuation. a price-to-sales ratio of about 26 compared to a sector median of about 3. AVGO is not inexpensive by any reasonable standards. Bulls claim the premium is justified and point to growth of at least 30%. Opponents argue that even a minor setback could result in a harsh, unsightly correction at multiples this stretched. For a while, both may be correct. Long after the math becomes comfortable, markets have a way of rewarding belief.
The timing is what’s really unclear. The company’s ranking and recommendation systems are already powered by the first Meta chip, the MTIA 300. By 2027, there will be three more generations that are increasingly designed for inference, which is the process by which an AI model truly responds to your query. However, the actual profits from this transaction won’t become apparent until 2027 and later. Depending on which decade you’re thinking of, Wall Street is effectively pricing a future it cannot yet see, which is either visionary or the oldest financial error.
Sitting through all of this gives us the impression that a power structure is subtly rearranging itself. Not by making a big announcement. The majority of people scrolled past a gigawatt commitment and a board reorganization. Tesla also had to deal with its skeptics. Deals that hardly move the stock that signed them can occasionally have the greatest impact.
