AMD's $8.5bn data center quarter makes it a real second source

Chosunbiz reported revenue of $13,377m, up 44.7% year over year and 16.0% from the prior quarter. Operating profit was $2,832m, up 123.0% year over year, and the operating margin was 21% against 14% a year earlier. Net income came to $2,428m, up 95.3%, with GAAP diluted earnings per share of $1.46 against $0.75. On a non-GAAP basis, earnings per share were $2.09, up 74%.

The data center line did the work. Revenue there was $8,546m, up 96.9% year over year and 27.2% quarter over quarter. That is 63.9% of total revenue, against about 47% a year earlier. The segment's operating margin rose to 34% from 25%.

Guidance that makes data center the company

AMD guided fourth-quarter revenue to about $17bn, plus or minus 3%. That would be up about 66% year over year and 27% from the prior quarter. It expects total revenue growth of more than 80% in 2027, with data center above four-fifths of sales.

That is the part that changes the buying decision. A second source at this scale gives hyperscalers and server buyers a credible alternative to Nvidia, and gives AMD pricing power it has not had. What is still unknown is how much of the 2027 target depends on a handful of named customers.

Lisa Su went further on the earnings call. Server revenue should double year over year in 2027, she said. Data center AI revenue will grow faster than the company projected 90 days earlier.

Chief financial officer Jean Hu said data center AI revenue should pass server CPU revenue in the fourth quarter. That crossover is the number to watch. It marks the point where accelerators, not general-purpose processors, set AMD's quarterly rhythm.

Customers, and a disagreement about which chip

Oracle committed to deploying 50,000 Instinct MI450 GPUs starting in 2026, according to Startup Fortune. Startup Fortune called it the largest single hyperscaler bet on AMD silicon to date. Anthropic, Meta, Microsoft, OpenAI and Oracle have begun initial operation of Helios AI server systems. Server makers and design manufacturers are preparing more than 240 platforms for the sixth-generation Epyc chip Venice. First systems are due in the fourth quarter.

The two accounts of the quarter's accelerator shipments do not match.

Chosunbiz reported expanded shipments of the Instinct MI450 series in the quarter. AJP reported that volume shipments of Instinct MI455X accelerators began in the quarter. The company statements quoted in the two reports do not settle which part number led the ramp.

The share AMD has to move

Nvidia still controls more than 90% of the AI accelerator market, according to industry estimates cited by Startup Fortune. The prize is growing. ResearchAndMarkets values the global AI accelerator market at $151.0bn in 2025. It forecasts $443.0bn by 2030, a compound annual growth rate of about 24%. GPUs hold 58% of that market by accelerator type, and data center accelerators 55% by architecture.

That role is contested from another direction. Oracle said on 27 October that it plans to offer Google TPUs on its cloud from 2027. It would become the first hyperscaler to offer them outside Google Cloud.

What happens next

The fourth-quarter guide of about $17bn is the next checkpoint, alongside Venice systems shipping in the quarter. Oracle's MI450 deployments are due to start in 2026 and expand in 2027, according to Startup Fortune.

Politics has entered the buildout as well. AJP reported that Decision Desk HQ projected Democrats would take both the House and the Senate following Tuesday's elections. That adds a constraint alongside memory, packaging and power shortages.