
AMD wants to ship substantially more chips in 2027, but chief executive Lisa Su says the company could still use additional capacity. Speaking to reporters in Taipei on October 6, she said demand for central processing units and graphics processors has continued to rise despite supply increases this year. The message is straightforward: in the AI race, designing a competitive chip is only half the job. A manufacturer must also be able to deliver enough of it.
Su said AMD would increase supply again next year and expand its investment in Taiwan’s semiconductor supply chain. She did not put a number on the revised commitment. The $10 billion investment discussed in May is proceeding, according to the Taiwan report, while AMD continues working with chip fabricators, packaging partners and suppliers of the materials needed to assemble finished products.
The demand is not confined to one fashionable AI accelerator. Data centres also need CPUs to coordinate workloads, move information and support the services built around models. That means a customer buying more AI computing capacity may need several kinds of components at once. If any one link in that chain is constrained, a technically ready product can still be difficult to ship in volume.
Su met Taiwanese technology partners during the visit and said she planned to see TSMC and other suppliers. She described Taiwan as critical to AMD, even as she encouraged greater geographical diversity in production. The company is extending some planning discussions from a year or two to as far as three to five years so that wafer, packaging and memory capacity can become available together. High-bandwidth memory, particularly important in AI servers, remains tight across the industry.
This gives fresh context to AMD’s recent data-centre growth. Strong sales make the opportunity clear, but they also raise expectations about how many systems AMD can actually provide. The comparison with Nvidia is not only about which company has the better benchmark result. Cloud providers buying at scale care about delivery dates, complete systems, software support and the ability to expand without repeatedly waiting for scarce components.
AMD says its Helios AI platform began shipping in the third quarter and expects those shipments to rise. That is a company timetable, not yet proof of a particular 2027 production volume. Su gave no specific unit forecast for the planned increase. Investors and customers will therefore be watching whether the suppliers can turn her promise into sustained output. If they can, AMD has a better chance of winning orders from AI buyers who want another source of computing power. If they cannot, demand alone will not close the gap.







