The most consequential sentence in Intel’s week came at a dinner in Santa Clara, in answer to a question about customers.
Speaking at the Chinese American Semiconductor Professionals Association’s annual conference on Oct. 24, Intel chief executive Lip-Bu Tan said he had discussed possible cooperation with AMD chair and chief executive Lisa Su. The areas under discussion, according to a recording of the fireside chat obtained by Tom’s Hardware, were custom chips, advanced packaging and foundry work. No agreement, no product and no timetable were disclosed.
That a conversation took place is not a deal. That it took place at all is a measure of how far Intel’s strategy has moved from selling its own processors to renting out the factory that makes them.
The two companies have been x86 rivals for four decades and have spent much of that time in court. They also co-chair an advisory group set up to standardise instruction-set extensions against Arm, a rare piece of cooperation that both sides describe as necessary rather than friendly. A packaging or foundry arrangement would go much further than that, and would put AMD’s silicon inside a plant owned by its principal competitor.
The commercial logic runs through Taiwan. Advanced packaging, the step that assembles a GPU or accelerator from separate dies and stacks memory beside them, has been the tightest link in the AI supply chain for two years. TSMC’s CoWoS lines are booked well into next year, and every customer designing an accelerator has spent the past eighteen months asking for capacity that does not exist. Intel has packaging capacity of its own, in the form of EMIB bridges and Foveros stacking, in plants on American soil.
For AMD, that is the argument. A second source for packaging would shorten the queue, and a US site would reduce exposure to a supply chain that Washington has spent four years trying to relocate. AMD’s processors are made at TSMC and will stay there for the foreseeable future; moving a leading-edge design to Intel’s process would take years of joint engineering and would hand Intel detailed knowledge of a competitor’s roadmap. Packaging is the part of the chain where a customer can buy capacity without surrendering design information.
Tan’s remarks on Intel’s own manufacturing suggested why a customer might listen. He said the 18A process, which pairs gate-all-around transistors with backside power delivery, is in volume production, and that 14A, the 1.4-nanometer-class node, has a 0.5 design kit in customers’ hands with a 0.9 kit due shortly and a 1.0 kit early next year ahead of volume production. He described multiple customers asking Intel to scale their business, a claim the company has made before and has not yet backed with named high-volume orders.
The candid part of the evening was about culture. Tan, who took over in 2025, said Intel had been too proud to modify its processes for customers and that previous management had lectured partners during engineering reviews. He described a list of fifteen fixes the company had made to its client business and a willingness to wait six to eight months for a customer decision that Intel’s sales organisation would once have treated as a delay.
Foundry businesses live or die on that detail. A leading-edge plant costs tens of billions of dollars and needs a handful of large customers running steady volume to pay for itself. Intel has commitments from the US government, which took an equity stake in 2025, and from Nvidia and SoftBank, which invested. None of those is a wafer order at the scale the fab needs. A design win at AMD, even in packaging, would be the first external validation of the foundry from a company that competes with Intel at the top of the market.
There are reasons to be sceptical. AMD and Intel have held talks before, on x86 standardisation and on licensing, and the results have been narrow. Any packaging arrangement would have to be scoped so that AMD’s product plans stay with AMD, and Intel’s foundry teams serve customers who are also its rivals in CPUs, a structural conflict every foundry has faced and none has fully solved. The conversation may also have been exploratory in the way that chief executives’ conversations usually are: a test of whether the other side is interested, conducted without staff work or terms.
The timing is not accidental. Intel needs external customers to justify 14A, whose development cost was the reason some investors argued for exiting manufacturing altogether. AMD needs capacity it cannot get from a single supplier. Both facts point at a deal that would be small in revenue and large in signal, which is often how foundry relationships begin: a packaging order, then a test chip, then, years later, a product.
Until something is announced, the record of a conversation is evidence of intent and nothing more. Tan chose to describe it on stage, in front of an audience of semiconductor professionals, which is where executives put information they want the market to hear without putting it in a filing. The next signal to watch is not another interview. It is whether AMD’s name appears on an Intel packaging order.
For Intel’s shareholders the useful part of the evening was not the AMD reference. It was the admission that the company’s problem with customers has been cultural rather than technical, and that fixing it is being managed as a list of fifteen items. That is a slower story than a partnership with a rival, and a more reliable one.
