Broadcom fell twelve percent on Thursday. The AI trade did not.
Those are two separate events, and conflating them is how markets manufacture selloffs out of nothing. The numbers Broadcom reported on June 3 were not the numbers of a company facing slowing demand. They were the numbers of a company whose stock had been priced for a miracle and delivered something merely extraordinary, and that gap between expectation and reality — not between reality and the year prior — is what the market sold.
Start with what actually happened. Total revenue came in at $22.19 billion against an estimate of $22.27 billion. The shortfall was $80 million on a $22 billion quarter. That is a 0.36 percent miss, a rounding error by any analytical standard, and it came not from AI semiconductors but from the software segment. On earnings per share, Broadcom beat. On AI revenue, Broadcom beat its own forecast. The company reported AI semiconductor revenue of $10.8 billion, up 143 percent year-over-year, a figure that exceeded management’s prior guidance and represents nearly half of every dollar Broadcom earned in the quarter. There is no reading of those numbers that supports the word disappointment.
The forward guidance is where the picture sharpens further. For the current quarter, Broadcom guided total revenue to $29.4 billion, above the $28.5 billion Wall Street had penciled in. AI semiconductor revenue alone is expected to reach $16 billion in the quarter, a more than 200 percent year-over-year increase. The company reaffirmed its full-year AI semiconductor revenue target at $56 billion, representing roughly 180 percent growth for the fiscal year, and reiterated expectations for AI semiconductor revenue to exceed $100 billion in fiscal 2027. These are not the forward curves of a business slowing down. They are the forward curves of a business accelerating.
The customer picture tells the same story. Broadcom has locked in long-term supply agreements with Google, Anthropic, OpenAI, and Meta for multi-gigawatt AI compute deployments, with significant new rollouts beginning in fiscal 2027. During the earnings call, management disclosed $6 billion in new AI orders from two additional customers. The company has also entered a partnership with Apollo and Blackstone to deploy more than 20 gigawatts of AI compute capacity, with the first tranche valued at $35 billion. These are not the contract structures of an industry approaching saturation. They are the contract structures of an industry in the early innings of a long buildout.
So what did the market actually sell? It sold the absence of a guidance raise that was not promised and was already beyond anything the consensus had previously modeled. In the months before the print, Broadcom’s stock had risen roughly 40 percent year-to-date, a run that embedded the expectation that results would not merely beat but reset the ceiling upward. When Hock Tan confirmed the existing guidance rather than lifting it, the market treated confirmation of extraordinary growth as a failure to achieve something more extraordinary still. That is not a signal about AI demand. That is a signal about how far above reality the price had traveled.
There is one legitimate point inside the noise. Tan indicated a shift in how Broadcom delivers to customers — moving from integrated AI systems toward supplying chips independently, rather than complete solutions. That change in delivery model introduced a degree of uncertainty about near-term implementation timelines for some customers, and uncertainty is always enough to give a jittery market permission to sell what it already wanted to sell. It does not alter the volume of demand. It alters the packaging of supply.
The semiconductor complex that sold off alongside Broadcom — names with no earnings event, no changed guidance, no new information whatsoever — followed not on fundamentals but on sentiment. When a high-profile name drops twelve percent, the path of least resistance for the rest of the sector is down, regardless of what the underlying business has reported. That is what happened to the semiconductor tape on Thursday. It was not a repricing of AI demand. It was the echo of a single disappointed trade.
The demand for the infrastructure required to run AI at scale has not softened. Hyperscaler capital expenditure commitments are at record levels and climbing. The custom silicon pipeline, which Broadcom’s results confirm in granular detail, extends with booked orders and signed contracts into 2028 and beyond. A market that sells twelve percent on 143 percent AI revenue growth and a guide that beat consensus is not reading the business. It is managing its own prior exuberance.
The bar was the problem. The business is fine.