AMD pairs cloud and on-device AI with new EPYC and Ryzen chips

AMD framed its strategy around combining cloud computing with on-device computing, and said the approach unifies CPUs, GPUs, memory, networking and software.[1] The company cited security and privacy protection, lower latency and usefulness where network connectivity is poor as advantages of running AI locally.[1] That makes the announcement relevant to teams weighing where AI workloads should run, though the cited benefits are AMD’s stated rationale rather than comparative test results in the evidence. AMD also cited a Goldman Sachs projection that token consumption will grow 24-fold by 2030.[1]

AMD said it began shipping 6th-Gen EPYC processors in the third quarter and identified Meta as an initial customer, with plans to expand supply from the fourth quarter.[1] It divided the EPYC line into SP7, SP8, VeniceX and Verano families.[1] AMD announced general availability of SP7 processors on Nov. 10, 2026, and said SP7-based platforms were shipping from leading server OEMs and ODMs.[3] AMD also said consolidating 1,000 two-socket Intel Xeon servers into 82 EPYC servers can reduce server operating costs.[1] These are AMD’s claims, not independent validation in the evidence.

Naver Cloud said it operates about 35,000 servers powered by AMD EPYC processors within a fleet of approximately 250,000 servers across 13 data centers in South Korea.[2] It said EPYC reduced rack-space requirements by 20% to 25% compared with Intel CPUs.[2] Naver Cloud’s measurements compared an Intel dual-socket, 144-core platform consuming 800 watts with an AMD EPYC 9825 single-socket platform consuming about 700 watts.[2] The comparison gives infrastructure teams reported figures to examine, but the evidence does not specify broader benchmark conditions.

AMD said models with 9 billion parameters can now achieve performance comparable to earlier 120-billion-parameter models.[1] It also said models of up to 300 billion parameters can run at 4-bit quantization using up to 192GB of unified memory.[1] Clebby unveiled Uniform, a document-processing engine built on Ryzen AI Max+ 395; the company said Uniform processes document text and structure directly, while its Cova engine handles images and tables.[1] Clebby said it is conducting proofs of concept with offices of education and design firms and pursuing an on-premise AI appliance combining Ryzen AI Max hardware with Uniform software.[1]