AMD chair and chief executive Lisa Su has flown into Taiwan early to fight for the next wave of AI silicon.
Ahead of COMPUTEX 2026, the two big AI chip names have started rattling the supply chain tin. Su arrived at Taipei Songshan Airport by private jet on the afternoon of 20 May, getting in before Nvidia chief executive Jensen Huang.
According to CTEE Su has attended the AI Summit Forum and met TSMC chairman C.C. Wei in person to push for 2nm process capacity and advanced packaging. The aim is to get AMD better placed for the next wave of AI servers and agent AI demand. In other words, Su wants the wafer space before everyone else starts shouting at TSMC.
Agent AI is moving the market beyond the old GPU arms race. AI is shifting from giant model training to reasoning and independent agent execution, which suddenly makes the CPU look less like a spare part.
The industry thinks future AI systems will not simply answer questions. They will break down jobs, plan processes, call outside tools, access data and keep checking results until something useful pops out.
Su has repeatedly said AI has entered the era of “CPU＋GPU collaborative computing”. That sounds like a slogan, but the underlying point is clear enough. She said the data centre ratio of CPUs to GPUs used to sit at about 1:4. With large-scale AI Agent deployment, that could move closer to 1:1.
Agents need task scheduling, data handling, memory management and tool coordination. The CPU is no longer just the bloke holding the GPU’s coat, but part of the core AI machinery. AMD has sharply lifted its CPU market forecast. Su said there are now more than 1 billion daily AI users worldwide, rising to 5 billion by 2030.
With inference and agent demand growing rapidly, AMD believes the server CPU market’s compound annual growth rate will exceed 35 per cent over the next few years. The company once estimated the CPU market would hit US$60 billion by 2030, but has now doubled that to US$120 billion.
Su stressed that CPUs and GPUs do not replace each other. They complement each other, which is handy when you sell both. Supply chain sources say AMD has been pushing Instinct AI GPUs while strengthening EPYC server CPUs and rack-level AI systems. The plan is to take Nvidia on directly in AI infrastructure rather than politely orbiting around it.
As AI server demand continues to heat up, AMD is moving to secure supply of advanced processes and CoWoS packaging. The company clearly does not want another AI chip squeeze turning into a queue-jumping contest.
Analysts think the shift from GPU single-point competition” to “CPU＋GPU＋system-level integration will broaden the benefits for Taiwan’s AI supply chain. That means more than TSMC’s advanced processes and CoWoS packaging.
Packaging and testing players such as ASE Technology and Powertech Technology could benefit. Server ODMs, cooling suppliers, power supply makers, connector firms and high-speed transmission outfits are all sniffing around the next AI infrastructure upgrade cycle.
