
Last 4 earnings calls · through Q2 FY2026
The bull case rests on server CPUs: agentic AI work is pulling general-purpose compute demand well beyond what management forecast a year ago.
Helios, the rack-scale system that combines Instinct GPUs with EPYC CPUs, is in production with named commitments from the largest AI labs.
A broader set of large AI customers means accelerator revenue no longer depends on winning a single deployment or a single buyer.
Free cash flow has more than doubled while gross margin improves, so the capacity expansion is funded from operations rather than from debt.
The bear case is that a few named AI labs account for most committed accelerator volume, so a delay at one of them would move the whole segment.
Rising memory and component costs are already cutting into consumer PC and gaming demand, and management has lowered its second-half outlook.
Console chip revenue is shrinking as the current console cycle ages, and the next generation does not arrive until 2027.
The accelerator business carries lower margins than the rest of AMD, so faster growth in it dilutes profitability even as revenue climbs.
In short: Data center revenue is doubling on AI and server CPU demand, while gaming and PC demand are weakening.
Data center revenue more than doubled year over year to $6.7 billion and supplied 58% of segment revenue, so AMD's results now track AI infrastructure spending more than the PC cycle.
Management now expects the server CPU market to grow more than 50% a year to about $220 billion by 2030, up from the 18% annual growth it modeled in November, which underpins the raised CPU outlook.
Client and gaming revenue rose only 6% year over year as semi-custom console sales fell, and the decline continues for several quarters until Microsoft's next Xbox launches in 2027.
Management first said memory prices could affect consumer demand and now plans for a softer second-half PC market outright, which caps client revenue even as Ryzen keeps gaining share.
In short: Free cash flow is strong, while accelerator mix holds gross margin below the corporate average.
Free cash flow was $8.4 billion over the last twelve months, or 20.3% of revenue, versus $4.0 billion a year earlier, which funds factory prepayments and memory commitments without adding debt.
The reported gross margin figure is flattered by comparison with a year-ago quarter that carried $800 million of China-related inventory charges, so the underlying improvement is smaller than the reported gain.
Management says the accelerator ramp is below the corporate average and ramps hardest in the fourth quarter, so doubled data center AI revenue dilutes blended margin even as gross profit dollars grow.
Reported operating margin was 17.3% in the second quarter against the 27% non-GAAP margin management cites, largely because stock-based pay and acquisition costs are excluded.
In short: AMD is a core supplier to the largest AI labs, while those few buyers and custom chips limit pricing power.
OpenAI, Meta and Anthropic have each committed multi-gigawatt Instinct deployments, giving AMD named multiyear volume instead of one-off orders from a single customer.
Management now describes ROCm as production-ready with day-zero support for leading models, after years of saying more work was needed, and that software credibility is what lets Instinct win inference workloads.
Management names only three frontier labs as the large committed customers, so a delay or a reallocation at one of them would materially change the 2027 data center AI ramp.
Hyperscalers are building their own accelerators and ARM server processors, which limits AMD's pricing power even while total AI compute demand grows.