AMD Results Place AI Infrastructure Spending Boom Under Renewed Scrutiny

Advanced Micro Devices Inc is set to report its second quarter results on Tuesday evening amidst a sector-wide retreat that has raised questions about the sustainability of hyperscaler spending on artificial intelligence infrastructure. The Philadelphia Semiconductor index has declined approximately 20 per cent from its late June peak, erasing more than one trillion dollars from the global valuation of chipmakers, despite remaining up roughly 90 per cent over the preceding 12 months.

The sell-off reflects mounting concerns over whether hyperscalers can maintain their current pace of AI infrastructure investment rather than any fundamental weakening in demand. Heightened anxiety surrounding Chinese competition in memory chips and the circular financing arrangements linking chip suppliers to their largest customers has intensified the negative sentiment. AMD shares closed at 476.15 dollars on Friday, down from above 550 dollars a fortnight earlier, though still substantially higher year on year.

Market consensus anticipates earnings of approximately 1.62 dollars per share on revenue of 11.3 billion dollars, compared with 0.48 dollars and 7.6 billion dollars respectively in the same quarter last year. The company had previously guided towards 11.2 billion dollars in revenue, subject to a variance of 300 million dollars either way, alongside a non-GAAP gross margin approaching 56 per cent.

Data centre revenue is projected to have more than doubled to roughly 6.5 billion dollars from 3.2 billion dollars, with server processor sales forecast to have grown in excess of 70 per cent. The client division, which supplies chips for personal computers, is expected to generate around 3 billion dollars in revenue, representing an increase of 20 per cent, whilst gaming is anticipated to fall 30 per cent to 781 million dollars. Rising memory prices are constraining the PC market, presenting a headwind for both AMD and Intel in the quarters ahead.

Investor attention will centre less on the historical quarter and more on the execution of the Helios rack-scale system, which AMD announced entered full production last month. Helios integrates 72 of the company’s MI455X accelerators alongside its Venice server processors and Pensando networking capabilities, with each rack priced between approximately 5 million and 5.5 million dollars. The product is positioned as a direct challenge to Nvidia’s competing systems.

Chief executive Lisa Su has asserted that the rack delivers 15 per cent more compute performance than Nvidia’s Vera Rubin, 50 per cent additional high-bandwidth memory and 30 per cent greater output per dollar spent. Initial customer shipments are scheduled for late in the third quarter, which implies that material revenue will not materialise until the fourth quarter. AMD has secured commitments from OpenAI for up to six gigawatts of capacity and from Anthropic for up to two gigawatts, in addition to deployments at Microsoft.

The primary constraint remains HBM4 memory, the scarcest component in the AI supply chain and one over which AMD exercises no direct control. This bottleneck represents a significant variable in the company’s ability to capitalise on the substantial pipeline of customer commitments.