At Cisco’s AI event, Lip-Bu Tan, CEO of Intel, sat down with Jeetu Patel to discuss Intel’s recovery, its evolving foundry strategy, and the system-level constraints shaping AI infrastructure. Tan described Intel as an iconic and strategically important U.S. company and said he accepted the CEO role after two years on Intel’s board, despite warnings from peers about the scale and difficulty of the task.
Tan said Intel now operates as two closely linked businesses: a product organization and a manufacturing foundry, each requiring distinct execution models and culture. He confirmed Intel’s intent to run its foundry as a general-purpose platform serving external customers as well as Intel products. On Intel 18A, Tan said yields were weak when he took over, prompting him to bring in outside expertise and work closely with equipment partners. He described 7–8% month-to-month yield improvement as an industry best-practice target and said Intel is now seeing improvements consistent with that cadence, which has led to renewed customer interest. Tan added that Intel remains focused on its next-generation 14A process, with risk production targeted for 2028 and volume production expected in 2029.
On AI infrastructure, Tan identified memory as the most acute bottleneck, saying customers see little relief for several years as AI workloads absorb rapidly growing memory capacity. He said compute demand has accelerated sharply, with deployment cycles shrinking from years to months, placing strain on manufacturing capacity and supply chains. Tan also highlighted thermal limits, power management, optical interconnects, and cluster-level software as critical constraints. He said Intel will continue to develop both CPUs and GPUs, manufacture them internally, and partner with external GPU providers, noting that different AI workloads require different architectures. Tan said he has hired a chief GPU architect and emphasized that software enablement—what he called “software 2.0”—is essential to making heterogeneous AI systems work at scale.
- Intel positions its foundry as a general-purpose platform for internal and external customers
- Intel 18A shows yield improvements approaching industry best-practice monthly targets, according to Tan
- Intel targets 14A risk production in 2028 and volume production in 2029
- Memory availability represents the most immediate bottleneck for AI scale-out
- Intel plans to build CPUs and GPUs, manufacture them internally, and partner with other GPU providers
- Cooling, optical interconnects, packaging, and cluster software now constrain AI systems alongside silicon
- Intel is evaluating new substrates and materials to address power, thermal, and signal integrity limits
- Tan cited glass substrates, diamond-based materials, and compound semiconductors such as gallium nitride as areas of active exploration
“Intel is an iconic company that matters to this industry and to the United States,” Tan said. “We have to execute with discipline, earn customer trust, and invest across manufacturing, packaging, software, and open ecosystems to enable the next wave of AI.”
