AMD’s third quarter was its best on record by the reported numbers. The more important disclosure was a sentence about which business will be larger next quarter.
The quarter
AMD reported fiscal third quarter 2026 revenue of USD 13.377 billion, up 45 per cent year on year, a quarterly record and about 2.3 per cent above analyst expectations.
Non-GAAP earnings per share rose 74 per cent to USD 2.09, more than 6 per cent above consensus. Non-GAAP operating income reached USD 3.923 billion, up 75 per cent, lifting the operating margin to 29 per cent from 24 per cent a year earlier and from 27 per cent in the prior quarter.
Data centre revenue was USD 8.546 billion, up 97 per cent year on year and 27 per cent sequentially, driven by EPYC server processors and the ramp of the Instinct MI450 series. The segment is now about 64 per cent of company revenue.
Data centre operating income was USD 2.882 billion, up 168 per cent, with segment margin rising from 25 per cent to 34 per cent.
The older businesses moved the other way. Client and gaming combined produced USD 3.7 billion, down 8 per cent, on weaker PC demand and a declining console cycle. Embedded contributed USD 1.1 billion, up 29 per cent, a bright spot too small to offset the decline.
The structural statement
Guidance came in at roughly USD 17 billion for the fourth quarter, plus or minus 3 per cent, implying about 66 per cent year on year growth and 27 per cent sequentially, well above the roughly USD 16.2 billion analysts had modelled.
Chief executive Lisa Su and finance chief Jean Hu then said the fourth quarter will be the first in which AMD’s data centre AI revenue exceeds its server CPU revenue.
Su also raised the longer view. AMD now expects 2027 revenue growth above 80 per cent, with server revenue doubling and data centre AI revenue more than tripling, both higher than the guidance given 90 days earlier.
Second reinvention, different rules
AMD has changed its business mix before. Before the Zen architecture arrived in 2017, the company was a follower in PC processors. Zen lifted its server CPU revenue share from low single digits to above 40 per cent.
That recovery happened inside a mature market with a fixed framework. AMD took share in x86 servers, which meant it changed its rank rather than the rules.
This transition is different. Moving from CPU vendor to AI compute supplier has taken under three years, from MI300 volume production to AI accelerator revenue passing server CPU revenue. The competitor set has also changed. Intel is no longer the primary reference point. Nvidia holds the software ecosystem, and the large cloud customers have both the compute budgets and the engineering capacity to design their own accelerators.
Supply chain, customer bargaining and the basis of competition have all been rewritten at once, and the risk is less bounded than it was in server CPUs.
What the market is now pricing
As data centre takes the majority of revenue, the client and gaming businesses move to the edge of the story. AMD’s multiple will track AI capex sentiment rather than PC cycles or x86 share.
That cuts both ways. The guidance assumes the supply chain does not break, orders convert on schedule and the software ecosystem gap narrows steadily. Each of those is a condition rather than a fact.
For suppliers and competitors watching from Asia, the relevant read is that accelerator demand is now large enough to reorganise a legacy semiconductor company’s income statement within a single year. The capacity and packaging constraints that follow sit mostly outside the United States.
Editor’s note: This is an adapted translation of the original 36Kr report. It has been trimmed and restructured for readability for an international business audience. The full original (in Chinese) is at https://36kr.com/p/4060589261854592.