AMD Locks In AM5 Through 2029 and Unleashes a Computex Product Cascade

AMD pledges AM5 support through 2029, launches RX 9070 GRE, retro CPUs, and $3,999 AI platform; Q1 revenue up 38% and stock near highs.

AMD used its Computex 2026 keynote in Taipei to deliver a rare combination of long-term reassurance and short-term aggression. The chipmaker formally committed to keeping the Socket AM5 platform viable until at least 2029, promising backward compatibility for current motherboards through the upcoming Zen?6 generation. For desktop users and system builders wrestling with rising memory and SSD costs, that pledge removes a key source of upgrade uncertainty.

The product blitz, however, is anything but cautious. AMD launched the Radeon RX 9070 GRE at $549, a 1440p-focused card with 12 GB of VRAM, 48 compute units and a 220?W power draw that slots below the existing RX 9070 and 9070 XT. The company also rolled out two retro desktop processors: the Ryzen 7 5800X3D 10th Anniversary Edition ($349, available June 25) and the Ryzen 7 7700X3D ($329, arriving July 16). The former celebrates a decade of Socket AM4 with an improved Carbice thermal pad; the latter serves as a cheaper entry point into the AM5 ecosystem, boasting 4.5 GHz boost and 104 MB of cache.

On the AI front, AMD opened pre?orders in June for the Ryzen AI Halo developer platform at $3,999 — $700 below Nvidia’s DGX Spark. Powered by the Ryzen AI Max+ 395 with up to 128 GB of unified memory, the system can run models with as many as 200 billion parameters locally. It sells exclusively through Micro Center and supports both Windows and Linux. AMD dubs such machines “Agent Computers” — devices that plan and execute tasks autonomously. A more potent version, the Ryzen AI Max PRO 400 Series, arrives in the third quarter with speeds up to 192 GB unified memory and 160 GB VRAM, making it, according to AMD, the first x86 client chip capable of handling 300?billion?parameter models. HP and Lenovo are expected to integrate the platform into their systems.

The hardware offensive comes on the back of a strong financial quarter. Revenue for Q1 2026 reached $10.25 billion, up 38% year over year, with the data?center segment growing 57% to $5.8 billion. Second?quarter guidance of around $11.2 billion exceeded consensus estimates. The stock has more than doubled this year, with some tallies putting the year?to?date gain above 130% — and now changes hands near €441, just under its 52?week peak of €444.80. A 4% dip on the session pushed the relative strength index to 29, signalling an oversold condition in technical terms.

Beyond the consumer and developer markets, AMD disclosed that production of its sixth?generation EPYC server chips, code?named “Venice,” has started at TSMC on a 2?nm process. The Instinct MI400 accelerators — armed with 432 GB of HBM4 memory — remain on track for the second half of 2026, directly targeting Nvidia’s next generation. Investors will get more detail on June 2, when CFO Jean Hu addresses the Bank of America Global Technology Conference, a session likely to focus on the Venice ramp curve and AI accelerator demand.

By coupling a multi?year socket promise with aggressive pricing across GPUs, CPUs and AI hardware, AMD is trying to lock in both the upgrade cycle and the developer ecosystem before Nvidia can set the rules for the coming wave of local intelligence.