Broadcom delivered strong AI chip growth, but the market wanted a bigger signal. That is the real lesson for anyone watching the next phase of AI infrastructure spending.
Broadcom did not report a weak quarter. It reported the kind of numbers most companies would happily put on the front page of an investor deck, then watched its shares fall because the AI chip story was not quite strong enough for Wall Street's current expectations.
The company said fiscal second-quarter revenue rose 48 percent from a year earlier to $22.19 billion, while AI semiconductor revenue grew 143 percent to $10.8 billion. Adjusted EBITDA reached $15.24 billion, equal to 69 percent of revenue. Those are large numbers by any normal standard. The problem is that AI chip investors are no longer using a normal standard.
Broadcom guided for $16 billion in AI chip revenue in the current quarter, slightly below the $16.36 billion expected by analysts polled by Visible Alpha, according to Reuters. It also left its longer-term target of more than $100 billion in 2027 AI chip revenue unchanged, even as Chief Executive Hock Tan said the company now expects to ship more than 10 gigawatts of AI chip capacity next year.
This is what happens when a stock becomes a test of belief. Broadcom has spent the past year becoming one of the clearest ways to invest in custom AI silicon outside Nvidia. Its relationships with major cloud companies, including Alphabet's Google and Meta, have made it a central supplier in chips designed for specific AI workloads rather than general-purpose graphics processors.
That position matters because the AI infrastructure market is no longer just about buying more Nvidia GPUs. Large cloud companies want chips tailored to their own models, software stacks and cost targets. Google's TPUs are the best-known example. Meta has been building its own MTIA chips. OpenAI has also been widely linked with custom chip plans. Broadcom sits behind many of these efforts, supplying the custom accelerator and networking expertise that turns a hyperscaler design ambition into something that can be manufactured and deployed.
So the issue was not whether demand exists. It clearly does. The issue was timing, scale and whether the curve is steep enough to justify the expectations already built into the stock. When AI revenue is already growing at triple-digit rates, a guidance number that is merely strong can still feel disappointing.
Custom silicon is not moving at software speed
There is a practical point here that investors sometimes ignore. Custom AI chips do not arrive just because a cloud company wants them. They require design work, manufacturing slots, advanced packaging, memory supply, networking capacity and data center power. Every one of those pieces can affect the pace at which revenue appears on Broadcom's income statement.
That makes the current reaction more subtle than a simple demand scare. Broadcom said it is very comfortable with supply for 2026 and 2027, which suggests the company is not warning of a breakdown in availability. But the market was looking for evidence that custom silicon programs were moving even faster than expected. When Tan held the 2027 revenue outlook steady, investors read it as a ceiling for now rather than a floor.
This distinction matters for other chipmakers trying to enter the hyperscaler supply chain. Marvell Technology, for example, has been pushing deeper into custom silicon and recently said its custom chip business should exceed $10 billion in revenue in 2029. That is a serious target, but Broadcom's report shows how little patience investors may have for long development timelines once AI expectations are priced in.
Nvidia remains the benchmark. Its GPUs are still the standard for training and many inference workloads, and its roadmap gives customers a ready-made platform that keeps improving. Custom ASICs can reduce cost and improve efficiency for specific workloads, but they are not a quick replacement. They are a second track, and that second track has to prove it can scale reliably.
The bigger AI spending question
Broadcom's report also says something about the AI capital expenditure cycle. Big Tech companies are expected to spend heavily on AI infrastructure this year, with estimates above $700 billion across the industry, up sharply from 2025. That spending does not flow evenly to every supplier. Some goes to Nvidia GPUs. Some goes to networking. Some goes to power, cooling, buildings and memory. Some arrives later through custom silicon programs that need years to mature.
For founders and investors, the lesson is useful. AI infrastructure is not one market moving in one straight line. It is a chain of markets, and each part has its own bottlenecks. A startup selling software into AI workloads may see demand immediately. A chip company building custom accelerators may wait quarters before design wins become revenue. A data center operator may be constrained by power before it is constrained by customers.
Broadcom is still in a strong position. Its semiconductor solutions segment reached $15.01 billion in quarterly revenue, up 79 percent from a year earlier, and its infrastructure software business added $7.18 billion. The company also guided for total third-quarter revenue of about $29.4 billion, ahead of broader analyst expectations compiled by LSEG. This was not a company losing its place in the AI race.
But the reaction shows that AI chip stocks are now being judged against a different standard. Investors do not just want proof that revenue is growing. They want proof that the buildout is accelerating faster than the last forecast, and they want that proof every quarter.
The next thing to watch is whether Broadcom turns its 2027 custom silicon pipeline into visible upward revisions later this year. If it does, the selloff may look like a short-term reset. If it does not, the market may start asking a harder question: whether the custom AI chip boom is real, but slower and more uneven than the stock market wanted to believe.
Also read: Nvidia brings local AI computing into the Windows PC fight • Putin is pulling Russia's AI race closer to the Kremlin • Alphabet is turning the AI race into a capital markets test
