Results land on Tuesday after-hours with data centre revenue expected to double and the sector in retreat
Advanced Micro Devices Inc's (NASDAQ:AMD, XETRA:AMD) second-quarter print is the official set-piece since a sharp reversal in sentiment towards semiconductor stocks.
The Philadelphia Semiconductor index has fallen roughly 20% from its late-June peak, wiping more than $1 trillion from the value of chipmakers worldwide, even though it remains up around 90% over 12 months.
The retreat has been driven by doubts about whether hyperscalers can sustain their current rate of spending on artificial intelligence infrastructure, rather than by any deterioration in demand.
Fresh worries about Chinese competition in memory chips, and about the circular financing arrangements linking chip suppliers to their largest customers, have compounded the mood.
AMD shares closed at $476.15 on Friday, having been quoted above $550 a fortnight ago, though they remain up sharply over the past year.
The numbers themselves are expected to be striking.
Analysts forecast earnings of about $1.62 a share on revenue of $11.3 billion, against $0.48 and $7.6 billion in the same quarter last year.
The company had guided to $11.2 billion, give or take $300 million, with a non-GAAP gross margin near 56%.
Data centre revenue is expected to have more than doubled to roughly $6.5 billion from $3.2 billion, with server processor sales growing more than 70%.
The client division, which supplies chips for personal computers, is forecast to bring in around $3 billion, up 20%, while gaming is expected to fall 30% to $781 million.
Rising memory prices are squeezing the PC market, a headwind for both AMD and Intel in the coming quarters.
Attention will focus less on the quarter just gone than on execution of the Helios rack-scale system, which AMD says entered full production last month.
Helios packs 72 of the company's MI455X accelerators alongside its Venice server processors and Pensando networking, priced at roughly $5 million to $5.5 million a rack, and is pitched directly against Nvidia's rival systems.
Lisa Su, chief executive, has claimed the rack delivers 15% more compute performance than Nvidia's Vera Rubin, 50% more high-bandwidth memory and 30% more output per dollar spent.
First customer shipments are due late in the third quarter, meaning meaningful revenue will not arrive until the fourth.
AMD has commitments from OpenAI for up to six gigawatts of capacity and from Anthropic for up to two, alongside deployments at Microsoft.
The constraint is HBM4 memory, the scarcest component in the AI supply chain, and one AMD does not control.
