DENVER — Intel Chief Executive Lip-Bu Tan said the industry must break Taiwan Semiconductor Manufacturing Co.'s grip on the global foundry market, arguing that orders should be spread across suppliers to meet rising demand for artificial intelligence chips. With Intel staking its future on a foundry revival, Tan also said surging demand for central processing units will keep supply tight.
Tan appeared on the keynote stage at Splunk's .conf26 conference in Denver, Colorado, on the 14th. Asked by Cisco President and Chief Product Officer Jeetu Patel about the importance of Intel's fabs, Tan said the ability to handle manufacturing as well as design, along with advanced packaging, is critical. "That is why we are in it," he said, describing the company as an iconic player in industry both in the United States and worldwide. Intel is an AI infrastructure partner of Splunk, an AI data management platform company, and of Cisco, Splunk's parent.
Fabs are a difficult business because they require the best process technology, Tan said, adding that yield, defects, cycle time and variability must all be tightly managed to make output predictable. Customer requirements differ, he said, and a foundry must hold the right intellectual property and support electronic design automation to serve them. Advanced packaging matters most of all, he added: putting CPUs, memory, input-output devices and silicon together in a single package is difficult and requires substantial expertise.
Tan was explicit about his intent to break TSMC's dominance in global AI chip production. "Everything is moving more and more toward a system approach and packaging, and that is the future," he said. "So I think relying 95% on one company, especially one based in Taiwan, is very risky." The remarks pointed to U.S. companies including Nvidia, AMD and Apple, which route nearly all of their AI chip production to TSMC, and signaled Intel's intent to capture foundry volume in line with the Trump administration's priorities. In an interview before the election, President Donald Trump criticized TSMC, saying chip companies are very wealthy and that they took 95% of the U.S. business, which now sits in Taiwan.
TSMC accounted for more than 70% of global foundry revenue in the first quarter, according to research firm Counterpoint, far ahead of Samsung Electronics (005930), China's SMIC and Intel. Tan said the business is not easy, requires a careful approach and demands heavy capital spending. "The good news is that things have improved over the past 18 months," he said. Intel struggled with yields on its 14A process, a 1.4-nanometer-class node, but agreed in April to supply 14A process technology to the Terafab project led by Tesla and SpaceX. Terafab's first product is due in mid-2028.
In June, Intel hired Lee Seok-hee, former chief executive of SK hynix (000660), as executive vice president overseeing advanced packaging and back-end technology development and manufacturing in its foundry unit. The move brought in a former Intel executive to strengthen back-end operations as the company wrestled with yield problems. Industry watchers have speculated that SK hynix and Intel could cooperate on producing base logic dies for high-bandwidth memory.
Tan also expressed confidence in EMIB, or Embedded Multi-die Interconnect Bridge, Intel's advanced packaging technology. EMIB connects chips with silicon bridges so that multiple dies operate as one. TSMC uses CoWoS, or Chip on Wafer on Substrate, which places an interposer wafer on a substrate and mounts chips on top. Intel is positioning EMIB as its answer to CoWoS. "One of the biggest challenges is the substrate," Tan said, noting that Japanese and Taiwanese companies are locking up ample supply through prepayments.
Tan emphasized demand for CPUs, Intel's core product, and predicted that the current shortage will persist as inference-focused AI agents multiply into the trillions. While graphics processing units are central to training, CPUs handle orchestration and are seen as the key AI chip of the inference era. CPUs coordinate large numbers of GPUs working in tandem while managing applications.
"CPU demand is so high that we can only serve 50% of our customers," Tan said. "Many CEOs call me and I tell them I am sorry we cannot produce enough." He added: "There are countless agents, and going forward millions and trillions of agents will need resources. We need enough compute power and security to handle that."
