Broadcom FQ2 26 – Beat, No Raise

A blip and a warning.

• Broadcom disappointed the market in what I think was a blip specific to Broadcom, but also serves as a warning that the market has no patience whatsoever when it comes to missing forecasts.
• FQ2 revenues / Adj-EPS were $22.2bn / $2.44, broadly in line with expectations, but failure to lift guidance for AI revenues put a crimp on expectations.
• Here, FQ3 revenues are expected to be around $29.4bn, which was just ahead of the forecast of $28.3bn but specific commentary around AI caused some concern.
• AI revenues in FQ3 26 are expected to be around $16bn, which is very strong growth but shy of the market’s forecast of $17.2bn.
• In an environment where everyone keeps raising expectations, the market was looking for a beat and raise but only got a beat.
• Broadcom also declined to raise its guidance of AI semiconductor revenue of $100bn in 2027, largely I suspect due to the recent widening of products on offer.
• Here, custom accelerators and CPUs are the fastest-growing portion of the market as more CPUs are required as more GPUs are networked together and models “think” for longer before producing an answer.
• For the last few years, Broadcom has had this market largely to itself with Google, but now everyone else is getting in on the action, and there is increasing uncertainty with regard to how much share Broadcom can hold on to.
• Broadcom’s hesitancy to increase its forecast despite the outlook for its end market improving over the last three months is being seen as a moment of uncertainty.
• Some will argue that the $100bn forecast was already somewhat aggressive and therefore already included the recent upgrades, but I think that the majority of this is the market share uncertainty.
• Hence, I don’t think that this is a reflection on the end market demand picture but more a reflection of market share loss and, to some degree, over-hyped expectations.
• Hence, the situation at the moment remains that compute remains a scarce resource and despite the $750bn being ploughed into it this year, this situation is unlikely to change.
• Part of the reason for this is that compute is still being sold to the users (consumer and enterprise) at a bargain basement price, which is one reason why demand has been so strong.
• This is beginning to change as prices are rising and unlimited usage plans are becoming more scarce or usage is being capped or heavily optimised.
• This is essential because unless the industry can make a decent return on the compute being sold, then there is going to be a large and painful correction.
• This is why the revenue per GW of capacity is so important, as with capex and operating expenses pretty much fixed, the only variable is revenue.
• This needs to rise substantially from the $10bn / GW where it is today, to at least $15bn and more like $20bn – $25bn for the industry to make a decent return.
• Broadcom currently trades on 37x FY2026 PER and 22x FY2027, which is not unreasonable assuming the growth trajectory that has been promised materialises.
• Hence, if this continues to weaken materially, it might be one to have a look at.