Intel CEO Lip-Bu Tan used his first extended podcast interview on No Priors to describe how he is reshaping Intel’s technology roadmap and operating model. Tan said Intel has already created roughly a 6x return for shareholders over the past 14 months, but added that “this is just the beginning.” His personal target is a 10x return over five to ten years. In his view, Intel’s future is not limited to its traditional PC client base. He pointed to edge computing, physical AI and agentic AI as markets that can expand the company’s role beyond servers and personal computers for human users.
Tan, who is 66, said he accepted the Intel CEO role because the company is iconic and remains highly important to the semiconductor ecosystem and to the United States. After Cadence, he said, he wanted to do one more major thing. He also described an unexpected episode in which President Donald Trump asked him to resign early one morning, citing a conflict of interest. Tan said he first set aside his personal emotions and reminded himself that he did not need the job and was doing it to save Intel. He later secured meetings, explained that he was born in Malaysia, grew up in Singapore, graduated from MIT and had lived in the United States since then, and was given the opportunity to continue.
From “crawl-walk-run” to a ten-year Intel plan
Tan framed Intel’s turnaround with the same “crawl-walk-run” approach he has used in other roles. The first step, he said, is to stay humble and listen to customers. Only after that can the company walk and eventually run. Since taking the job, he has focused on changing Intel’s culture, clarifying accountability and speeding up decision-making. He contrasted the pace of startups, where work moves at “light speed,” with Intel’s layers of meetings and bureaucracy, which he said he needed to change.
One of Tan’s first decisions was to have all engineering teams report directly to him. As an engineer, he said, he wanted to know personally where the problems were and what needed to be corrected. The operating agenda he laid out includes listening to customers, making customers satisfied, ensuring the right products are in place, simplifying the product line, and building a clear roadmap and vision for the next five to ten years.
On the balance sheet, Tan said Intel’s situation was “quite bad” when he arrived. He said he was glad the U.S. government became a major shareholder and told President Trump that, as in Japan and Singapore, semiconductors should be treated as infrastructure that deserves government support. Tan also thanked Nvidia CEO Jensen Huang, saying Huang invested $5 billion in Intel and that the stake has since grown to $25 billion or more. SoftBank’s Masayoshi Son, with whom Tan had worked when he served on SoftBank’s board, also helped. Tan said these moves stabilized Intel’s balance sheet.
Product focus is the next layer of the plan. Tan said demand for CPUs has become extremely strong because of agentic AI and inference workloads. During the training-heavy phase, he said, the ratio of CPUs to GPUs in data center servers was around one to eight; now he sees it moving toward one to four, or even lower. He said AI model developers have told him that CPUs perform better in reinforcement learning and in coordinating and scheduling many agents. After strengthening the data center server product line, Tan said the foundry business becomes another major pillar, followed by a full-stack model that includes software and system-level solutions. Some customers, he noted, have asked him directly to “give me the whole rack.”
Foundry execution: trust, yield and cycle time
Intel’s foundry business has faced outside doubts, but Tan said he chose to stay committed. His reasoning is that advanced manufacturing in the United States is strategically important for supply-chain security and for the broader industry. Semiconductor companies have all experienced supply-chain challenges, he said, and no major company should depend completely on one or two geographically concentrated suppliers.
Tan emphasized that Intel respects TSMC and sees it as a strong partner, not only as a competitor. The industry needs more capacity to serve customers, he said, which is why Intel decided to keep pushing. Still, he acknowledged that Intel remains far behind TSMC in foundry and must be humble. The priorities he identified are IP, yield, defect density and cycle time. Foundry, in his words, is both a service business and a trust business: “Customers have to trust you before they give you their wafers.” If yield is not good enough, customers suffer revenue losses and leave, making the relationship difficult to recover.
Tan tied the foundry roadmap to advanced process technology. He said Intel has 18A, described as a 1.4-nanometer-class process, and is planning for 1 nanometer and 0.7 nanometer. As nodes become smaller and lines become thinner than a human hair, precision requirements increase and a single mistake can undermine the entire process. Tan said these manufacturing demands are becoming a major bottleneck, and that Intel’s real potential in foundry should begin to show in the 2030 to 2032 period.
Terafab and the semiconductor infrastructure gap
The interview also covered the Terafab project with Elon Musk. Tan said he and Musk share a view that semiconductor infrastructure has not kept pace with AI growth. The gap, as he described it, exists in capacity, production efficiency and power efficiency. Musk’s robots and cars need large volumes of chips, and under the Terafab framework Musk decided to build his own wafer fab while Intel provides technology and process support to help accelerate production.
Tan said he meets with Musk’s team every week and described the collaboration as energizing. He praised Musk’s willingness to challenge conventions and ask why each step must be done in the traditional way. According to Tan, this approach creates a process in which both sides can learn and search for a better path. He also mentioned Musk’s unconventional operating ideas, including a discussion about whether smoking could be allowed in certain areas of a cleanroom. Tan said he would not go that far, but that some areas could be evaluated; the key, he said, is to keep an open mind.
Materials science, EMIB, glass substrates and synthetic diamond
Tan placed materials science and advanced packaging at the center of Intel’s attempt to push beyond the physical limits of conventional scaling. He said Intel has 18A, is advancing 14A toward mass production, and can see a path to 10 nanometers and 7 nanometers, but that this path will become increasingly expensive and difficult. Because of that, Intel needs partners and close collaboration with substrate suppliers and equipment makers to improve yield and performance.
Advanced packaging is one of the bottlenecks Tan highlighted. TSMC has CoWoS, while Intel is promoting its next-generation EMIB approach. Tan said he must ensure EMIB reaches customer-required yield levels in volume production. Intel has also announced advanced packaging manufacturing cooperation projects in India and in New Mexico in the United States. In modules, Intel has roughly 1,000 patents, and Tan identified the integration of substrates and modules as a core engineering task.
When conventional scaling approaches its limits, Tan said he goes back to materials. He has invested in gallium nitride, silicon carbide and indium phosphide companies, and some of those investments have already been acquired by larger semiconductor companies such as ADI. In packaging materials, he has focused on glass and invested in a company called 3DGS because of glass’s properties as a heat-dissipating insulating material. He has also invested in a synthetic diamond wafer company, seeing diamond as another strong insulating material for chip packaging. “That is the spirit of engineers — you keep running into bottlenecks, and then you find ways to cross them or go around them,” he said.
Tan also discussed whether process-node convergence could narrow the performance gap among foundries. He answered by returning to Moore’s Law: transistor density can double, but power and cost do not necessarily fall at the same rate. Performance can rise, but area and cost do not automatically improve unless new materials and new design methods are found. That is why, he said, he has increased hiring in materials science, which he sees as central to innovation in the field.
Semiconductor investing, Cadence experience and AI-era teams
Tan drew on his long record as an investor and operator. He said venture capital and entrepreneurship are in his blood, and cited 159 IPOs, 126 merger-and-acquisition exits and more than 200 semiconductor investments, 38% of them in the United States. His investment framework starts with a simple question: where is the bottleneck, and what problem is the company solving? He mentioned Cradle Semiconductor as an investment tied to interconnect bottlenecks, and Celestial AI as one tied to the growing importance of optical interconnects inside clusters. He also said Nvidia’s Jensen Huang has invested in almost every photonics-related company, and that this is not a coincidence.
In design, Tan sees major opportunity in EDA if AI and machine learning can reduce complexity and improve design quality. He also named gallium nitride, silicon carbide and indium phosphide as materials areas where he has invested, with some companies acquired by ADI and other large semiconductor players. Power management is another bottleneck he cited, especially the large losses involved in converting from 40V to 1V. For every investment, he asks whether the problem is real, whether customers are truly struggling with it, and who the first target customer is. He prefers hyperscale customers because they have the capability and willingness to pay millions over several years if they like a technology.
Tan connected this investing view to his nearly 15 years at Cadence. One of the achievements he said he is most proud of is finding and personally developing his successor, who is now actively embracing AI and bringing agentic AI into tools to improve efficiency. He said Synopsys’s Sassine is doing the same, with support from Nvidia’s $2 billion investment and through the acquisition of Ansys to expand into full-system design. Large companies are moving in this direction, but Tan said startups can still create more disruptive products, eventually going public or being acquired by the two large EDA companies.
On team structure, Tan again returned to the crawl-walk-run model. In the crawl phase, he has recruited top semiconductor talent. Now he is thinking about what kind of software talent Intel needs to build full-stack capabilities. He also noted that the average age of parts of the team is in the 40s and 50s, and that he needs younger people who understand workloads and frontier open-source models. Tan said his son has become his teacher on AI and machine learning; when he visits his son’s home to play with his grandchildren, he asks questions and converts what he learns into investment and hiring judgment.
Tan described Intel as historically old-school and spreadsheet-dependent, and said he is trying to turn it into an AI-enabled enterprise. That does not mean using AI only in design, but embracing it across the organization and reducing reliance on spreadsheets. He said Intel needs to combine senior technical talent with AI tools, not only in sales and marketing but also in design.
Capital sources, investor misunderstandings and where compute will live
For capital-intensive businesses and infrastructure projects, Tan said access to capital is essential. Some venture firms are now willing to put $1 billion into a single company, something he said was previously unimaginable. For AI factories and foundries, he said companies must seek support from government funding, sovereign wealth funds or large infrastructure funds. As a public company CEO, he is also deliberately focusing on long-term growth-oriented investors rather than short-term investors who ask every quarter about share buybacks, while acknowledging that shareholder returns are a reasonable concern.
Asked about investors’ biggest misunderstanding of Intel, Tan said the company is still in the crawl phase, even though people are starting to see the opportunity. In PC clients, Intel still has share, but must greatly improve performance. He said he is quietly building CPU architecture, GPU architecture and software architecture teams to prepare for a leap, moving like a large startup. In foundry, he said Intel must keep building the basics: IP, yield, defect density and cycle time. These things take longer, but he believes the 2030 to 2032 period is when the market will begin to see Intel’s true potential.
Tan said Intel’s PC client business remains the base, but the company is expanding toward the edge, physical AI and agentic AI. In the past, servers and PCs were provided for humans; now millions of agents will need access to compute and software stacks. Training is led by Jensen Huang and Nvidia, while the edge, agentic AI and physical AI create additional dimensions. Tan said the game is not over and that Intel has opportunity in both agentic AI and physical AI.
He returned to the investor mindset behind his 10x goal. In Cadence, he said, the stock rose from $2.40 during his period from acting CEO through retirement, creating roughly a 76x return for shareholders; by the end of his executive chairman period, the return was around 85x. Intel is much larger and harder to replicate, he acknowledged, but as someone who remains a VC at heart, his target is a 10x return over five to ten years.
On the distribution of compute, Tan said large-scale AI infrastructure buildout is correct and that he sees no reason for it to slow, because workloads continue to grow. Current limits are mainly on the supply side rather than the demand side. But he said the key question is what applications will run on the infrastructure once it is built. He compared this to the internet era, when Amazon and Netflix emerged as real applications while others disappeared or were acquired. Some workloads, including robotics and defense, are better suited for edge or client-side compute, where assumptions about connectivity and device capability determine what can be done. For Intel, Tan said the direction is to integrate XPU, advanced packaging and foundry capabilities to create custom chips for different workloads.