Intel CEO Lip-Bu Tan used his first in-depth podcast appearance on No Priors to lay out a broad restructuring plan for Intel, combining balance sheet repair, product simplification, foundry discipline, advanced packaging and a renewed push into materials science. Tan said he is 66 years old and many people told him he should retire rather than take on one of the most difficult jobs in the semiconductor industry. His answer was that Intel is an iconic company, central to the semiconductor ecosystem and to the United States, and that after Cadence he wanted to do one more major thing.
Tan described the assignment in direct terms: he said he is doing it to save Intel. He also said Intel has already generated roughly a 6x return for shareholders over the past 14 months, but called that only the beginning. His personal target is to create a 10x return over 5 to 10 years. In the interview, he repeatedly framed Intel’s transformation through what he calls the crawl-walk-run model: first stabilize the basics, then rebuild momentum, and only later move at full speed.
One of the most dramatic moments he recounted involved President Trump. Tan said that early one morning, Trump asked him to resign over alleged conflicts of interest, with no exceptions. Tan said he first set aside personal emotion and reminded himself that he did not need the job, but was doing it to help Intel. He later secured a meeting and explained his background: he was born in Malaysia, grew up in Singapore, graduated from MIT and has lived in the United States ever since. According to Tan, Trump listened to the explanation and gave him the opportunity to continue.
Balance sheet repair, product focus and the first stage of Intel’s rebuild
Tan said the first task was to stabilize Intel’s balance sheet, which he described as being in very poor condition when he took over. He welcomed the U.S. government becoming a major shareholder and said he told President Trump to look at Japan and Singapore, where semiconductor capability is treated as infrastructure and government support is regarded as appropriate. He also thanked Nvidia CEO Jensen Huang, whom he described as an old friend, for investing 5 billion dollars in Intel. Tan said that 5 billion dollars has now grown to 25 billion dollars or more. He also mentioned support from SoftBank founder Masayoshi Son, noting that he had previously served on SoftBank’s board.
Inside the company, Tan said he moved first on culture, accountability and decision speed. He is used to the pace of startups, where everything moves at what he called light speed, but Intel had layers of bureaucratic meetings. He said this had to change. From day one, he decided that all engineering teams would report directly to him. As an engineer by background, he wanted to understand personally where problems were arising and what needed to be corrected.
The product plan begins with listening to customers, improving customer satisfaction, simplifying the product line and setting a clearer roadmap for the next five to ten years. Tan said the timing has helped: agentic AI and inference workloads are driving strong demand for CPUs. In the training era, he said, the ratio between CPUs and GPUs was around one to eight; now he sees that moving toward one to four, or even lower. CPU demand has therefore become stronger. He said AI model developers told him that CPUs can perform better in reinforcement learning steps and in coordinating and scheduling large numbers of agents.
After rebuilding the data center server product line, Tan sees Intel’s foundry business as the next major pillar. He acknowledged that foundry is capital intensive and difficult. It requires the right IP portfolio, such as low-power IP for mobile customers, and it is both a service business and a trust business. If yields are not good enough, customers lose revenue and leave. For that reason, he is focused on yield, defect density and cycle time. He also said the future has to be full-stack: not just silicon, but software and system-level solutions. Some customers, he said, have directly asked Intel to provide an entire rack.
Foundry strategy: trust, yield and a resilient supply chain
Intel’s foundry business has often been questioned, but Tan said he chose to keep investing in it after hearing many arguments against doing so, including that it was too expensive and would not work. His conclusion was that advanced manufacturing in the United States is extremely important for the country and for the industry. Large semiconductor companies, he said, must think seriously about supply chain resilience and cannot rely entirely on one or two geographically concentrated suppliers.
Tan said Intel already has 18A technology, a 1.4-nanometer-class process, and is planning for 1 nanometer and 0.7 nanometer technologies. As process nodes shrink, the line widths become extremely small and the complexity becomes very high. Any single error can undermine the process. This is why manufacturing precision, yield and reliability are becoming more important. Tan emphasized again that foundry is a trust business: customers must trust Intel before handing over wafers.
He also stressed that Intel and TSMC are partners, not merely competitors. Tan said Intel respects TSMC, that the two companies have a good partnership, and that the broader industry needs more capacity to serve customers. The core work for Intel Foundry is to strengthen IP, improve yield, lower defect density, reduce cycle time and become more efficient and reliable. Tan said these tasks will take time, but he expects people to begin seeing Intel’s real foundry potential around 2030 to 2032.
Terafab with Elon Musk and the pressure from AI infrastructure
Tan also discussed Terafab, the project involving Elon Musk. He said he and Musk share the view that semiconductor infrastructure has not kept pace with the growth of AI. The gap is visible in capacity, production efficiency and power efficiency. Musk’s robots and cars require a large number 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 enjoys working with Musk because Musk is unconventional and repeatedly asks why each part of the process must follow the traditional way of doing things. Tan said he likes hearing different views and then finding the best path forward together. He meets with Musk’s team every week and described the team as strong and the cooperation as exciting.
One example Tan gave was Musk raising ideas that break with convention, including whether smoking could be allowed in some parts of a cleanroom. Tan said he would not go that far, but that some areas might be considered differently and that the key is to keep an open mind. He said Intel is listening to and evaluating such ideas seriously rather than rejecting them immediately.
From a wider supply-chain view, Tan said AI’s influence on the overall landscape will exceed that of the internet and be even more far-reaching. AI can improve efficiency across many tasks, including semiconductor design, where timing optimization, time to market and cost can all be improved. At the same time, AI demand is running into constraints. Tan listed power limits in some countries, the impact of helium on the semiconductor industry, memory shortages, and shortages of both CPUs and GPUs. He said costs eventually pass through to clients.
Materials science, EMIB, glass substrates and synthetic diamond
On the question of physical limits in chip scaling, Tan said Intel has 18A, is advancing 14A production, and can see a path to 10 nanometers and 7 nanometers. But he also said that path will become more expensive and more difficult. This is why collaboration with substrate suppliers and equipment companies is needed to improve yield and performance. Traditional node scaling alone is no longer enough.
Advanced packaging is becoming one of the central bottlenecks. Tan said TSMC has CoWoS, while Intel has its next-generation approach called EMIB. His job is to ensure EMIB can meet customer yield requirements in mass production. Intel has also announced advanced packaging manufacturing cooperation projects in India and in the U.S. state of New Mexico.
Tan is also returning to materials science for new routes around physical limits. He said he has invested in gallium nitride, silicon carbide and indium phosphide, and that some of those companies have been acquired by larger semiconductor companies such as ADI. In packaging materials, he has focused on glass and invested in 3DGS, a glass substrate company. He values glass for its properties as a heat-dissipating insulating material.
Intel also has around 1,000 patents in modules, and Tan said integrating substrates and modules is an important engineering challenge. Another direction he highlighted is synthetic diamond. He has invested in a diamond wafer company and sees diamond as another strong insulating material for chip packaging. Tan summarized the engineering mindset this way: engineers constantly run into bottlenecks, then find a way to cross them or go around them.
Tan also said Moore’s Law is fundamentally about doubling transistor density, but power and cost do not fall at the same rate. Performance can double, but area and cost do not automatically improve in the same way unless new materials or new design methods are found. This is why he has increased hiring in materials science. He recalled that 15 to 20 years ago many venture capitalists were not interested in semiconductors. Today, with Nvidia at 5.3 trillion dollars in market value, Broadcom and TSMC each at 2 trillion dollars, Lisa Su’s AMD near 800 billion dollars and Intel near 600 billion dollars, semiconductors have again become essential infrastructure.
Investor lens: bottlenecks, EDA, optical interconnects and AI design tools
Tan also spoke as a long-time investor and operator. He said venture investing and entrepreneurship are in his blood. As background, he cited 159 IPOs and 126 merger-and-acquisition exits, with more than 200 semiconductor investments, 38% of them in the United States. His investment framework starts with one question: where is the bottleneck, and what problem is being solved?
He used Cradle Semiconductor and Celestial AI as examples. Cradle Semiconductor addressed interconnect as a bottleneck, while Celestial AI focuses on optical interconnect, which is becoming increasingly important inside clusters. Tan noted that Jensen Huang has invested in nearly every photonics-related company, and said that is not a coincidence. He also sees major opportunity in EDA, where AI and machine learning can reduce complexity and improve design quality. He described that area as a gold mine.
Power management is another bottleneck he named. Moving from 40 volts to 1 volt creates large losses, and Tan sees that as an important area. In selecting investments, he asks whether the problem is real, whether customers are truly struggling with it, and who the first target customer should be. He prefers hyperscale customers because they have the ability and willingness to pay millions of dollars or provide some kind of support over several years if the product solves a real problem.
Tan also emphasized talent. He named the United States, Silicon Valley, Austin and Israel as areas he watches closely. He said Israeli entrepreneurs are highly disruptive and extremely hardworking, even continuing meetings during wartime and moving to voice calls from shelters when alarms sound. He also said that beyond agentic AI, physical AI is the next major frontier, and that he remains deeply involved in investments around frontier models and open-source frontier technology for physical AI.
Cadence experience, AI teams and Intel’s 2030-to-2032 horizon
Reflecting on Cadence, where he spent nearly 15 years, Tan said one of his proudest achievements was identifying and personally developing his successor, who is now actively embracing AI. Cadence is bringing agentic AI into tools to improve efficiency. Synopsys, led by Sassine, is doing similar work with support from a 2 billion dollar Nvidia investment and has expanded toward full-system design by acquiring Ansys. Tan said large companies are moving in this direction, but startups can still build more disruptive products and then either go public or be acquired.
On scaling companies and making investment decisions, Tan said capital intensity, unpredictability and cyclicality all have to be considered. He usually likes to enter early, help build the team and find investors who will stay through difficult periods, not only during good times. Strategic investors are also useful when they can add value in manufacturing, memory, interconnect or other areas. He said that when he looks back, 9 out of 10 companies he invested in changed their business plans midway because the market changed.
For Intel’s own team in the AI era, Tan said the crawl stage meant recruiting the best semiconductor talent he could find. He is now thinking about the software talent needed to build full-stack capabilities. He also noticed that the average age of some teams is in the 40s and 50s, so he wants to bring in younger people who understand workloads and frontier open-source models. He said his son has become his teacher on AI and machine learning, and that he tries to convert what he learns into investment judgment and recruiting decisions.
Tan said Intel used to be a very old-school spreadsheet-driven company, and he is trying to turn it into an AI-enabled enterprise. That does not only mean AI in design; it means embracing AI across the whole organization and reducing dependence on spreadsheets. He wants to combine senior technical talent with AI tools, not only in sales and marketing but also in design.
On capital sources and industrial policy, Tan said capital access is crucial for capital-intensive businesses and infrastructure projects. Some venture capital firms are now willing to invest 1 billion dollars in a single company, something he said would have been unimaginable before. For AI factories and fabs, he said companies need support from government funds, sovereign wealth funds or large infrastructure funds. As a public-company CEO, he also said he is deliberately focused on longer-term growth-oriented investors rather than short-term investors who ask every quarter when the company will buy back stock.
Asked about the biggest investor misunderstanding of Intel, Tan returned to the crawl-walk-run framework. He said the company has been in the crawl phase over recent months, although people are beginning to see the product potential. The PC client business remains Intel’s base, but performance must improve sharply. He is quietly building CPU architecture, GPU architecture and software architecture teams to prepare for leapfrog progress, moving like a large startup.
Beyond PCs, Tan said Intel is extending toward edge computing, physical AI and agentic AI. The company previously served humans through servers and PCs, but now millions of agents also need access to compute and software stacks. Tan said the opportunity in agentic AI and physical AI remains open. His long-term direction is to integrate Intel’s XPU, advanced packaging and foundry capabilities to build custom chips for different workloads.
Tan ended by comparing his Intel goal with his Cadence record. At Cadence, he said the stock rose from 2.4 dollars and generated roughly a 76x return for shareholders by the time he retired as CEO, and about 85x by the time he finished as executive chairman. Intel is larger and harder to move, he acknowledged, but as a venture capitalist by instinct he looks for 10x opportunities. For Intel, he said, the target is a 10x return within 5 to 10 years.
On where compute will live, Tan said today’s large-scale AI infrastructure buildout is appropriate and he sees no reason for it to slow, because workloads continue to grow. The main constraints are on the supply side rather than the demand side. His larger question is what applications will run on that infrastructure after it is built. He compared the next phase to the internet era, where applications such as Amazon and Netflix emerged while others disappeared or were acquired. In AI, he expects a similar process of growth and consolidation, with one or two true winners emerging in some areas.
Some workloads, Tan said, are better suited to edge or client devices. Robotics and defense are examples where on-device compute matters, and assumptions about connectivity and built-in device capability determine what can be done. His investment approach remains to find a real problem, find the right partners and assess whether the application’s market size is sustainable. If he truly believes in the opportunity, he said he doubles or triples down.