Intel CEO Lip-Bu Tan on the Comeback of American Chip Industry

Summary

• Michael Marks frames the macro case in his intro: the world bought over $790B of semiconductors last year, up more than 25% in a single year, meaning the industry’s trillion-dollar-by-2030 goal will land roughly four years early. In his closing, Marks adds that American chip companies took 53% of the global market last year — the highest US share since 1984. He calls Intel “the only firm on US soil able to compete in making leading-edge semiconductors at scale,” with Washington, Nvidia, and SoftBank now investors.
• Tan’s contrarian arc is the episode’s spine: 20 years ago, pitching semiconductors emptied VC partner meetings, and a co-investor once challenged him to “name me 1 semiconductor company with more than $1 trillion in market cap.” His answer today writes itself — “the most valuable company is a semiconductor company” — and the man whose investors asked “are you nuts?” about tripling down on a “sunset industry” now runs Intel.
• Intel’s strategy is deliberate vertical integration: “the product, advanced packaging, and foundry tie together. You create more value for the customer.” Tan wants a full stack — best CPU, GPU, and system architects plus software — aimed beyond PC client at “agentic AI, the edge, and also physical AI.” Having watched Intel miss mobile, cloud, and AI: “From here on, I said, ‘I’m not going to miss any of the big waves.’”
• A memory re-entry is being openly teased. Tan “used to say, ‘Don’t invest in memory because it’s kind of a commodity business,’” but now sees CPU-memory stacking and new memory architectures as “one of my pet projects.” He just hired his friend Shauki Lee, whom he said used to run SK Hynix: “You kind of know something that I’m thinking about. We’re not ready to unfold it.”
• Tan’s roughly 550-company investing record and startup network are part of Intel’s radar: SambaNova’s first investment, according to Marks, was $2M in 2017 at “something like” a $12M valuation, while Tan is in the process of closing a Series F round between $800M and $1B; Credo sits at about a $50B market cap; and Astera Labs was described through a four-person team “from Instart from TI” at about $72B. The model going forward: stay plugged into VCs and university labs, “spin out some of the startups. Eventually, I can spin them in and then acquire them” — building a bigger Intel platform across photonics, microfluidic cooling, EMIB-T packaging, glass, and artificial-diamond insulators.
• The Cadence playbook — a stock at $2.42 that rose 3,000% over 15 years, per Marks — transfers directly: humility, obsessive listening, and less-than-24-hour customer response. A competitor’s diagnosis of his edge: “My customers treat me as a vendor, but the same customers treat you as a partner.” Tan says it’s “the same playbook at Intel… but a little bit more complex,” running both products and foundry.
• On competing with his own investors — Nvidia, the US government, and SoftBank — Tan refuses the frame: “I don’t view them as competitors… the market is big enough.” And on horizon: “I’m not the short-term guy. I’m looking at 10 years, 15 years from now. At my age, I can easily retire, but I decided this is so important I want to step in and make an impact.”

Deep dive

1. From emptied conference rooms to the most valuable company on earth

• Tan’s origin story runs through failure: MIT nuclear engineering was cut short when Three Mile Island ended the career path (“there are no new nuclear power plants, there’s no future career for you”), leading to the only résumé he says he ever wrote and a free ticket to San Francisco in his early 20s.
• The defining anecdote of his contrarian era: pitching semiconductors to a tier-one VC partnership 20 years ago, “half of them found an excuse to leave the room… politely,” while the two who stayed asked whether he had any software-service startups instead. Even some of his investors asked, “Are you nuts? It’s a sunset industry?”
• The vindication, in his own words: a co-investor once demanded that he “name me 1 semiconductor company with more than $1 trillion in market cap” while Tan argued not to shut down the semiconductor part of the business. “I’m so happy right now that question is gone.” Marks’ closing framing adds that US chip companies took 53% of the global market last year, the highest share since 1984: “Being right early is lonely work. It’s a lot less lonely now.”

2. The early-stage record: SambaNova, Cerebras, and the connectivity trade

• The AI compute thesis, dated nine years back, “almost 10 years ago”: knowing from S3-era graphics investing that “the GPU is very power-hungry,” Tan believed generative AI would be the future and much bigger than the training side. He backed two companies and technologies: Cerebras at Series A for its difficult wafer-scale approach, and SambaNova’s RDU dataflow architecture for better performance and lower power.
• The SambaNova specifics Marks supplies: their first investment there, led by Tan, was $2M in 2017 at “something like” a $12M valuation. Tan says he funded the company from Series A through the B, C, and D rounds, and now F, bringing in investors each time. He is in the process of closing a Series F round “somewhere between $800 million and $1 billion.”
• The connectivity portfolio as proof of pattern recognition: Credo is now about a $50B market cap; Tan also names Astera Labs, describing a four-person team “from Instart from TI” at about a $72B market cap as he told it. He cites photonics exits as well: Elara AI sold to Marvell and DustPhotonics sold to Credo. Marks: “This is when investing is really fun.” He calls the numbers “nuts.”
• His global, team-first filter across close to 550 investments: he invests in Israel, India, China, and Taiwan, and says nine out of 10 companies change their business plan as the market changes. So he backs a team — “not just one individual” — that can build a world-class company with the right culture.

3. The Cadence turnaround playbook — and why it transfers to Intel

• The reluctant-CEO comedy is real history: Tan said no 10 times to joining the Cadence board, 12 times to becoming interim CEO, and agreed to do it for three months while the stock was down to $2.42. Those three months became 15 years; Marks says the stock rose 3,000%.
• The mechanism was listening at industrial scale: an all-hands invitation for ideas produced 300 emails per day, and Tan answered every one. A customer once demanded a refund and said it no longer wanted the product; Tan says he won those customers over one by one. Another later marveled, “I just complained to you, and less than 24 hours later, people came to the office and fixed it.”
• The line that carries the whole philosophy, from a competitor’s mouth: “My customers treat me as a vendor, but the same customers treat you as a partner.” Partners share road maps and customer feedback; Tan says that when a company becomes large, he can serve as a customer’s window into competitors, industry trends, and change.
• Asked if it is the same playbook at Intel: “Yes, very much so, but a little bit more complex” — more people, employees, products, and a foundry business, with success required on both sides. The culture change remains the same: listen, be humble, learn from customers, and show vulnerability so people will help.

4. Intel’s strategy: vertical integration, full stack, and never missing a wave again

• Why he took the job when he “could easily retire”: Intel is an iconic company, important to the industry, and important to the United States. The strategic bet is that “the product, advanced packaging, and foundry tie together. You create more value for the customer.”
• The product agenda is a full-stack focus — “the best CPU architect… and the GPU architect and the system architect and software” — aimed beyond PC client at “agentic AI, the edge, and also physical AI.” On servers, Tan acknowledges that “over the years, we’ve made a lot of mistakes” and says he now needs to correct them by looking at requirements over the next five or 10 years.
• The frontier map runs through high-speed connectivity and optical links; cooling that progresses from air to liquid to microfluidic; and packaging from EMIB-T to glass, with artificial diamond as an insulator. Tan’s way to stay current is to work with startups, VCs, and professors at Stanford, MIT, and Berkeley: “Develop them in-house if you can. If not, spin out some of the startups. Eventually, I can spin them in and then acquire them.”
• The through-line, stated in the episode: “In the past, Intel missed the big waves, like mobile, then cloud and AI. From here on, I said, ‘I’m not going to miss any of the big waves.’”

5. The memory tease, coopetition, and the 15-year clock

• Marks’ back-to-the-future question — Intel was formed in 1968 as a memory company, and he says friend Sanjay Perot’s memory business is “over a trillion, 1.2, I believe” — draws Tan’s most newsworthy answer: CPU demand is huge, and his day is filled with CEO calls asking for more CPUs. CPU and memory can be stacked together, and Tan sees new memory architectures emerging.
• The reversal is stated plainly: “I used to say, ‘Don’t invest in memory because it’s kind of a commodity business.’ But now it has become different.” One of Tan’s pet projects is exploring new memory architectures. He also says he hired his friend Shauki Lee, who used to run SK Hynix: “You kind of know something that I’m thinking about. We’re not ready to unfold it.”
• On the strangeness of Nvidia, the US government, and other partners being connected to Intel while also competing with it, Tan dissolves the tension rather than engaging it: “I don’t view them as competitors. I view the market as big enough. We just find how to create a bigger market for all of us.”
• His timeline, delivered to his board: “I’m not the short-term guy. I’m looking at 10 or 15 years from now. How are you going to build a bigger platform?” It echoes the 15 years he spent at Cadence and frames the long-term Intel journey.