What happened
Broadcom reports fiscal third quarter results after the United States market close on 2 September, and Wall Street consensus points to revenue of 29.43 billion dollars with GAAP earnings of 2.55 dollars a share. The single line investors care about is not the total, however. It is the artificial intelligence semiconductor figure, which the company has guided to 16.0 billion dollars, growth of more than 200 percent on the same quarter a year earlier.
That guidance was itself the problem when it was issued. Analysts had modelled roughly 17.2 billion dollars, so the official 16.0 billion number landed below expectations and the shares fell despite what was, in absolute terms, an extraordinary growth rate. It is a reminder that share prices respond to the gap between results and expectations rather than to results alone.
The prior quarter set a high bar. In the three months ended 3 May, Broadcom generated revenue of 22.2 billion dollars, up 48 percent year on year, with adjusted earnings per share of 2.44 dollars, up 54 percent. AI semiconductor revenue within that was 10.8 billion dollars, a rise of 143 percent driven by custom AI accelerators and AI networking equipment.
Broadcom occupies an unusual position. Rather than selling a general purpose graphics processor, it designs custom AI accelerators, chips built to the specification of a single very large customer, alongside the networking silicon that connects thousands of those chips together inside a data centre. Both parts of that business have been running at capacity.
Why it matters
Broadcom has become one of the clearest read outs on whether the AI investment boom is still accelerating. Because its custom chips are ordered many quarters ahead by a handful of hyperscale customers, the order book is a genuine forward indicator rather than a description of the past. When Broadcom raises guidance, it is telling the market that the largest technology companies have committed more capital.
The scale involved is now macroeconomic. Data centre construction has become a visible contributor to United States growth figures, and the associated demand for electricity is reshaping energy planning in several states. A slowdown in chip orders would ripple through construction, power generation and industrial equipment well beyond the technology sector.
There is a concentration risk in the other direction. A small number of AI linked companies now account for a very large share of the total value of the American stock market. That means an index fund which looks diversified across 500 companies is, in practice, heavily exposed to whether a handful of AI capital spending plans continue as scheduled.
The guidance shortfall matters for a subtler reason. If the constraint on 16 billion rather than 17.2 billion is customer demand, that signals the boom is cooling. If it is manufacturing capacity, packaging bottlenecks or supply of high bandwidth memory, that signals demand still exceeds what anyone can build. Management commentary on the call will be scrutinised for exactly that distinction.
Explained simply
If the AI boom is a gold rush, Broadcom is not panning for gold. It is the firm making the custom shovels to order and laying the railway that carries everything back to town.
Training a large AI model requires enormous numbers of specialised chips working together. There are two ways to buy them. You can purchase general purpose accelerators off the shelf, which is fast and flexible but expensive. Or you can commission a chip designed specifically for your own software, which takes longer to develop but is cheaper and more power efficient once running at scale.
The biggest technology companies increasingly choose the second route, and Broadcom is the partner that turns their designs into working silicon. Because these projects take years and involve enormous commitment on both sides, the revenue is unusually visible in advance compared with the rest of the semiconductor industry.
The second half of the business is arguably just as important. A modern AI cluster is tens of thousands of chips that must exchange data constantly at extreme speed. If the network between them is slow, the expensive chips sit idle. Broadcom supplies the switching silicon that prevents that, which is why its networking revenue grows in step with accelerator sales.
The word guidance is central to reading tonight. Guidance is the range management publishes for the coming quarter. Investors treat it as more informative than the results themselves, because the results describe a period that has already ended while the guidance describes the one that has not.
What it means for you
If you hold a global tracker fund, an S and P 500 tracker or a default workplace pension fund, you own Broadcom whether you chose to or not. It sits inside the top ten holdings of most global index funds, so a large move tonight will show up in a pension valuation within a day or two. That is normal and requires no action.
The more useful exercise is to check concentration. Log into your pension and look at the top ten holdings of your default fund. If seven or eight of them are AI linked American technology companies, your portfolio is far less diversified than the words global equity fund suggest. Adding a global small cap or an ex United States fund is a straightforward way to reduce that overlap without abandoning equities.
Resist trading around the result. Broadcom shares regularly move 5 to 10 percent overnight on earnings, and the direction depends on how the numbers compare with expectations that are not published anywhere retail investors can reliably see. Buying after a jump or selling after a fall is the classic way to convert a good long term holding into a poor return.
If you invest monthly through a stocks and shares ISA, this is the argument for regular contributions rather than lump sums. Automatic monthly investing buys more units when prices fall and fewer when they rise, which removes the need to have any opinion at all about a single earnings release.
The bigger picture
The AI capital spending cycle is now three years old and has already outlasted most predictions of its end. The pattern in previous technology build outs, from railways to fibre optics, is that infrastructure spending overshoots real demand, then corrects sharply while the infrastructure itself goes on to be extremely useful. Knowing that pattern exists does not tell anyone when the turn happens.
The signals worth tracking are capital expenditure guidance from the largest cloud companies, the order backlog Broadcom discloses, and whether custom chips continue taking share from general purpose accelerators. Together those describe the health of the whole chain far better than any single share price.
For tonight, the number that will set the tone is the guidance for the following quarter rather than the results just reported. If AI revenue guidance for the next period comes in above 18 billion dollars, the boom is intact. Anything closer to flat would be the first genuine warning sign this cycle has produced.