Dialogue with Intel's Chen Liwu: The End of AI Isn't Just GPU, But Also Power, Materials, and Manufacturing

Video Title: Re-engineering the Semiconductor Supply Chain with Intel CEO Lip Bu Tan

Video Author: No Priors

Compiled by: Peggy

Editor's Note: Against the backdrop of continued investment heating up in AI infrastructure, discussions within the semiconductor industry are shifting from "Is GPU supply sufficient?" to "Can the entire computing and manufacturing system support the next phase of AI expansion?" Over the past two years, the market has focused more on models, computing clusters, and the NVIDIA ecosystem. However, as the long-term growth of AI demand gradually becomes a consensus, a more critical question emerges: If chips, packaging, power, materials, memory, and manufacturing capacity all become bottlenecks simultaneously, what kind of new semiconductor supply chain does the AI industry really need?

This No Priors dialogue invited Intel CEO Lip Bu Tan (Chen Liwu) to discuss Intel's transformation, U.S. domestic manufacturing, the foundry business, AI's renewed pull on CPU demand, and new manufacturing collaborations like TerraFab. Chen Liwu is both a long-term semiconductor investor and an industry operator from Cadence to Intel. Therefore, the value of this conversation lies not in presenting a single company's narrative, but in showcasing how an industry expert rethinks the structure of semiconductors in the AI era.

In this dialogue, Chen Liwu deconstructs "How Intel Revives" into a set of more fundamental structural questions: How to repair the balance sheet, how to refocus product lines, whether advanced manufacturing can return to the U.S., whether AI workloads will redefine CPU value, and how semiconductor investment should focus on real bottlenecks.

First, the Intel problem shifts from "product lag" to "organizational and capital structure reconstruction." Previously, external discussions about Intel often focused on process delays, GPU absence, and insufficient foundry competitiveness. But Chen Liwu first emphasizes not a specific product generation, but the balance sheet, organizational culture, and customer trust. He proposes a path of first "crawling," then "walking," and finally "running": first strengthen the financial foundation, simplify product lines, bring engineering teams closer to the CEO and customers, then gradually rebuild the roadmap. This means Intel's revival isn't achievable through a single product launch but requires a systemic repair of organizational speed, capital patience, and technology direction.

Second, AI's demand for computing structure is becoming more complex. Previously, the AI narrative was almost dominated by GPUs, with training clusters becoming the clearest consensus in capital markets. However, Chen Liwu points out that with the development of Agentic AI, reinforcement learning, multi-agent scheduling, and edge computing, CPUs are becoming important again. The CPU-to-GPU ratio might shift from 1:8 in the training era to 1:4, and even approach 1:1 in some scenarios. This means AI infrastructure won't have only one chip winner; future competition will increasingly revolve around system-level combinations for different workloads: CPU, GPU, NPU, advanced packaging, software stack, and foundry capability will all become part of the same computing network.

Third, semiconductor manufacturing is transforming from a commercial efficiency issue back into a national infrastructure issue. Over the past thirty years, global chip manufacturing became highly specialized driven by efficiency, concentrating advanced manufacturing capabilities in a few regions and companies. But supply chain shocks, AI capacity demand, and geopolitical risks make "relying on only one or two geographic regions and players" increasingly unsustainable. Chen Liwu analogizes the U.S. government's investment in Intel to the early relationship between TSMC and the Taiwanese government, pointing towards a new consensus on industrial policy: for capital-intensive, long-cycle, strategically critical manufacturing systems, governments, sovereign funds, and long-term capital will become key participants again.

Fourth, the logic of semiconductor investment is shifting from "betting on hot sectors" to "finding real bottlenecks." Chen Liwu repeatedly mentions the keyword not as valuation, but as bottleneck: interconnect, photonics, EDA, advanced packaging, power conversion, heat dissipation, new materials, memory, helium, electricity—all could become constraints during AI expansion. Previously, semiconductor investment was avoided by VCs due to high CAPEX, long tape-out cycles, and high customer switching costs; now, as AI demand pushes these bottlenecks to the forefront, semiconductors are becoming a focus for venture capital, strategic capital, and industrial capital. This means truly valuable investments are not simply chasing "AI concepts" but judging which link is becoming the constraint for the next round of expansion.

Fifth, future computing won't exist only in hyperscale data centers. The SaaS and cloud computing era formed a highly centralized computing paradigm, but robotics, defense, home devices, Physical AI, and Agentic AI are making edge and on-device computing important again. Chen Liwu doesn't deny the continued expansion of large AI data centers, but he is more concerned about what applications this infrastructure ultimately serves. In other words, computing capacity building only generates long-term value when combined with sustainable, major applications. This also means the next phase of AI competition is not just about "who builds more data centers," but about "who can connect computing power, chips, and application scenarios into a scalable system."

If this conversation is compressed into one judgment, it is: AI is pushing semiconductors from single-chip competition towards a comprehensive restructuring involving supply chains, capital structures, manufacturing capabilities, and system architecture. In this sense, the subject of this article is not just whether Intel can revive, but whether the computing infrastructure of the AI era needs to be completely redesigned.

The following is the original content (edited for readability):

TL;DR

· The bottleneck for AI is no longer just GPUs; it's the constraints of an industrial system composed of power, memory, packaging, materials, and manufacturing capacity.

· The key to Intel's revival is not a counterattack with a single product but the systemic repair of the balance sheet, engineering culture, customer trust, and product roadmap.

· CPUs are becoming important again not because the GPU narrative is cooling, but because Agentic AI, reinforcement learning, and multi-agent scheduling are creating new computing workload structures.

· Semiconductor foundries are not just a manufacturing business; they are a trust business. Before delivering wafers, customers must first trust that yield, cycle time, and reliability won't destroy their revenue.

· The signal from TerraFab is that AI demand growth is so rapid that leading customers are starting to proactively engage with upstream manufacturing infrastructure rather than passively waiting for chip supply.

· Rebuilding advanced chip manufacturing in the U.S. relies not just on factory subsidies but on the recombination of government capital, long-term capital, industrial customers, and manufacturing capabilities.

· The core of semiconductor investment is not chasing hot concepts but identifying the bottlenecks that truly limit industry expansion, such as interconnect, power consumption, heat dissipation, packaging, and new materials.

· Future AI competition won't happen only in hyperscale data centers; edge devices, robotics, defense, and Physical AI will push computing power back to the application site.

Original Compilation

Host:

Hello everyone, welcome back to No Priors. Today, Elad and I have invited Lip Bu Tan (Chen Liwu). He previously worked at Walden, then served as CEO of Cadence, and is now CEO of Intel. We talked about his plan to transform Intel, the U.S. government becoming a major shareholder of Intel, how to be an excellent semiconductor investor, and whether we can actually manufacture chips in the U.S. Welcome, Lip Bu.

Why Did Lip Bu Tan Take on Intel?

Host:

Lip Bu, great to see you. Let's start with a very direct question: Intel is an extremely important U.S. semiconductor company, but the CEO position is incredibly tough. Why did you still take this job?

Lip Bu Tan:

That's a good question. I'm 66 years old this year. Many people would say you should retire, not take on the hardest job in the industry. I did it for several reasons. First, Intel is an iconic company. It's very important for the semiconductor ecosystem and for the United States. So I decided, after Cadence, to do this one last thing.

Host:

A lot has happened in the past year. What surprised you the most?

Lip Bu Tan:

What surprised me the most was something my previous work experience and training never prepared me for: one morning, very early, President Trump asked me to resign, citing a conflict of interest with no exceptions.

So first, I had to convince myself: one, I don't need this job. I took it purely to save Intel. So I put the personal aspect aside first, then thought about what I could do to help Intel.

The good news is, I scheduled a meeting for Thursday morning and saw him on Monday. He was willing to listen to my explanation. I told him I was born in Malaysia, grew up in Singapore, then went to MIT and have lived long-term in the U.S. I've never lived outside the United States.

I explained all that to him; he listened very carefully and gave me the opportunity. So I'm happy.

Host:

Now you have the chance to really do the work. You said earlier the goal of this job is "to save Intel." In your view, what does Intel winning again, thriving again, specifically mean?

Lip Bu Tan:

I've been in the role for 14 months now. A lot has happened in these 14 months.

First, it's about changing the culture. Clearly, we need a stronger sense of accountability. Second, decisions must be made faster. I'm very accustomed to startup culture: moving at the speed of light, no bureaucracy, no layers of meetings.

So the changes I've driven include: strengthening accountability, listening to customers, making customers happy. Some customers say Lip Bu is humble, willing to listen, willing to solve their problems, trying to satisfy customers.

Also, from day one, I decided to have all engineers report directly to me. I'm an engineer by background, I want to know exactly what's wrong, what needs to be fixed. I want to listen to customers, make them happy, and ensure we have the right products, simplify the product line, while having a clear roadmap and vision for the next five to ten years.

Intel's Vision for the Next Decade: First Save the Balance Sheet, Then Rebuild Products

Host:

What's your vision for Intel for the next ten years?

Lip Bu Tan:

I think there are several things. First, whether at Cadence or Intel, I've always believed: first learn to crawl, then stay humble, listen to customers; step two, start walking; finally, start running, sprinting. That's my culture: step by step.

For me, the first step is to strengthen the balance sheet. Intel's balance sheet is, in a way, pretty bad. So I'm glad to see the U.S. government become a significant shareholder.

As I explained to President Trump, when TSMC started, it also had the Taiwanese government as a shareholder. Look at Japan, Singapore—semiconductors are essentially infrastructure; the U.S. government needs to provide support.

Second, I'm also happy my old friend Jensen Huang (Huang Renxun) invested $5 billion to support me. I'm glad I did at least some things right. His $5 billion investment has now turned into $25 billion, even more.

Also, there's Masayoshi Son from SoftBank. I was on the SoftBank board before; he also extended a hand to help me. So we first strengthen the balance sheet, then focus on products. I significantly simplified the product line, listened to customers, and drove the next generation of leadership products.

In a way, we're also lucky. Now with the rise of Agentic AI, CPUs have become very sought after. Previously in training scenarios, the CPU-to-GPU ratio might have been 1:8; now I see it might become 1:4, even 1:1. I'm glad CPUs are important again.

I've talked to some AI model developers. They say for reinforcement learning, and the speed of scheduling large numbers of agents, CPUs actually have an advantage. So in a way, I'm happy; the market demand for my CPUs is very high right now.

Overall, we need to push on the product side, especially in data center servers. Another part is our wafer foundry business.

Initially, this is a capital-intensive business, not easy. You need to have several conditions. You need the right IP to support customers. For example, if a customer is in mobile-related products, you must have low-power IP. Without those capabilities, you can't serve them.

Foundry is a service business and a trust business. If a customer gives you an order, gives you wafers, but the yield is poor, their revenue gets hurt, they miss opportunities.

So, for us, it's very important to focus on yield, defect density, cycle time, and ensure we can meet customer needs, serve customers with high quality and reliability. These are the things I really focus on.

Ultimately, you also have to go full stack. Not just silicon, but also software. Some customers will ask me directly: Can you give me the whole rack system? That means you have to build the system. So these things, we're quietly advancing step by step, recruiting the best talent wherever possible.

By the way, I do all the hiring personally, no headhunter companies. So sometimes, having a strong network list, knowing who to call, is very helpful.

Host:

You've been in this industry a long time. You were CEO of Cadence before, I remember for about 12 years?

Lip Bu Tan:

13 years.

Host:

13 years. Then served as executive chairman for two more years, so 15 total.

Lip Bu Tan:

I originally agreed to do it for only three months.

Host:

Three months?

Lip Bu Tan:

Yes. So I'm very careful now. Once you say "I'll only do three months," it might turn into 15 years.

What is TerraFab? Why Does Musk Want to Build His Own Fab?

Host:

It seems you still have a long way to go at Intel. Another big project widely discussed is TerraFab and your collaboration with Elon Musk. Could you talk more about how this project came about? What is your involvement? How do you two collaborate?

Lip Bu Tan:

Of course. Elon Musk, I think we all agree, is one of the best entrepreneurs of this century, perhaps even the best. We share a common judgment: semiconductor infrastructure hasn't actually kept up with AI growth speed. You need capacity, productivity, and efficiency. These are the gaps he and I both see: there's truly a missing link here.

Second, I'm happy to collaborate with him. He's unconventional. I call it "non-traditional." He questions every step: why do it the traditional way? In a way, it's very refreshing. I like that. I like working with people who have different perspectives, then figuring out the best path together. Both sides will learn a lot in the process. Clearly, he also has his own vision: his robots, his cars need a lot of silicon.

Host:

Could you explain what TerraFab is? Many might not be familiar.

Lip Bu Tan:

TerraFab is his decision to build his own wafer fab. Meanwhile, we're happy to partner with him, ensuring we can work together to get him into production faster, achieve volume faster, and use some of our technology and processes. This is something we're collaborating on together. His team is excellent; I communicate with them weekly. It's exciting to work with them.

Host:

He's also mentioned some ideas, like wanting to smoke in the cleanroom and such, which are typically considered...

Lip Bu Tan:

Yes, yes, and burgers. I think I wouldn't go that far. Maybe certain areas of the cleanroom could. But the key is to keep an open mind. We'll also listen, see what's feasible.

How is AI Reshaping the Global Semiconductor Supply Chain?

Host:

It's exciting to see you reshaping this company in the U.S.: building the foundry business step by step while also partnering on projects like TerraFab. If you look at the global AI and semiconductor supply chain from a macro perspective, i.e., if you observe how AI is reshaping the supply chain by country in a macro way, you'll see different countries are affected differently.

For example, talk about AI leading to layoffs—I think most of it is exaggerated currently. Many layoffs are just due to over-hiring during the 2020 pandemic. But I see the first to be cut are often outsourcing companies, as businesses prefer to cut external manpower before internal employees. So they cut external customer service, external IT. This impacts countries with large BPO industries more in the short term, like the Philippines, India. They might feel the AI impact more immediately.

Then if you ask how companies in various countries can participate in AI's future in a positive way, it almost needs a country-by-country analysis. Places with cheap energy can do data centers; places capable of training models can train models, but maybe only the U.S. and one or two other places have that capability.

How do you see the global semiconductor supply chain changing? Should certain countries invest more? For instance, Israel has Mellanox, NVIDIA, Intel presence—should it do more in semiconductors? Should the Philippines return to a manufacturing base? How do you think about these issues from a global perspective?

Lip Bu Tan:

That's a good question. Clearly, AI is changing the whole landscape. I think its impact will be bigger than the internet, more profound. AI initially helps you do things more efficiently. Many agents can help you complete tasks that were tedious but necessary, faster. So it can significantly boost efficiency. Even in semiconductor design, AI can boost efficiency, like how fast you can complete design timing-wise; second is cost. So these help companies improve efficiency.

There are also several bottlenecks in AI demand and growth. First, of course, the power constraint everyone knows. Some countries simply don't have enough electricity, so they'll be affected. Second, many don't realize the impact of helium on the semiconductor industry could be very significant. Third, everyone knows memory is severely short now; everyone is scrambling for memory. Even if you want to build fabs to add capacity, it takes years. CPUs, GPUs, same thing; demand will be very high. Prices will also rise because we must pass costs to customers. So these will affect the whole industry's growth.

Overall, I think the companies most impacted are those not embracing AI. Because AI can help companies improve efficiency across all functions. We should embrace AI and find better ways to use it, whether for prediction, design, or various workloads. The potential is huge.

Host:

The simple objection many have about TerraFab or Intel's foundry business being competitive actually centers on one issue: some factors are inside the fab, like the IP, business speed you mentioned; there are also external factors. Elad talked a lot about that too.

One is labor cost and actual manufacturing capability. You invest in the foundry business, clearly believing there's a possibility: we can manufacture locally. Elon believes that too. Could you talk about this? How real is this constraint?

Lip Bu Tan:

You mean the labor constraint?

Host:

Yes.

Lip Bu Tan:

When I decided whether to double down on the foundry business or exit it, there were many voices in the market. You saw it; many said it's too expensive, won't succeed, won't succeed. But I ultimately decided this is very important for the U.S., very important for the whole industry.

We've all experienced supply chain challenges. For any large semiconductor company, you must seriously think about supply chain issues. You need a robust, resilient supply chain; you can't rely on just one or two participants in different geographic regions.

So I think more and more people will realize manufacturing in the U.S. is critical. And the most advanced processes, like our 14A, roughly 1.4 nanometers, we're already planning for 1 nanometer and 0.7 nanometer. Sizes get smaller, even much finer than a hair. So complexity is very high, not easy. If you make a mistake at any step, you start over. So manufacturing must be very precise.

From this perspective, this will increasingly become a bottleneck. We respect TSMC very much; it's a great partner. More importantly, we both need more capacity to serve customers. So we decided to grit our teeth for the long haul. I think long-term, it's critical, and it's a place where I can create more value for the industry.

Host:

People have long discussed that eventually we'll hit some resolution limit, can't shrink further. Line widths become too small to advance. When do you think we'll truly hit that limit?

Lip Bu Tan:

That's a good question. I think now we have 18A, next 14A will go into mass production, and I can still see the path for 10A and 7A. So I think this road can continue. But it will become more expensive, more difficult. That's why we need partners. We can't do it alone. We need to work with material suppliers, equipment vendors to truly improve yield and performance.

Another part becoming a bottleneck is advanced packaging. Everyone knows TSMC's CoWoS. Now we also have a very good next-generation solution called EMIB. I must ensure it can be mass-produced with yield meeting customer requirements.

Now Moore's Law is also starting to lose steam, as you said. So I'm also looking at some new materials, going back to the material level, back to the periodic table. I've invested in three types of materials: gallium nitride, silicon carbide, and indium phosphide, also observing how these new materials can drive the next step.

In packaging, I started investing in glass. Glass is a good thermal insulator, so I invested in a startup called 3DGS. Later I realized, Intel has about a thousand patterns on the module, so how the substrate and module combine is very important.

We just announced a large project with the Indian government for manufacturing in India and New Mexico, U.S. So advanced packaging is very important. I also started looking at synthetic diamonds. It's also a very good insulator. So I invested in Diamond Foundry too. These are directions to watch for the next generation. That is, new materials, new substrate materials, and new design methodologies will drive the industry forward.

As an engineer, you always hit walls. But after hitting a wall, you either find a way to jump over or go around, ultimately getting a better result. As someone who's long invested in semiconductors and participated in building the semiconductor industry, from EDA tools, to design, to manufacturing, having this experience is actually very helpful. Now I can use my own way to make a small contribution to the industry.

The Keys to Semiconductor Investment

Host:

What you just said is interesting: there's always something you can go around, but there are indeed physical limits. When you get to scales like 7 angstroms, you'll hit limits, must find new materials or other detours.

The interesting question is, we've been discussing this for a long time. I remember 20 years ago, people said we'd eventually reach a point where there's no space left on the chip. Will you encounter some kind of asymptote where performance differences between different fabs flatten out?

Lip Bu Tan:

That's a good question. Regarding Moore's Law, in the past we pursued performance doubling while also considering power and cost. You can double performance, but cost and area can't maintain the same advantage. So you must make trade-offs in these areas, unless you find new materials, new design methods, and truly implement them.

I've started hiring more talent in materials science. This is the focus of innovation in our field: how do we continue to advance?

I also remember 18 years ago, I was still investing in semiconductors. Back then, most venture capital firms, including some excellent top-tier VCs who are good friends, were on my meetings. At the beginning of partner meetings, all partners were in the room listening to me talk semiconductors. Halfway through, half would make excuses to leave. The remaining half would ask: Lip Bu, do you have any software service projects? In the end, only two people were left listening out of sympathy.

So history has changed. Now semiconductors are hot again. Look at Jensen Huang's NVIDIA, already a $5.3 trillion market cap company. Broadcom and TSMC are also in the $2 trillion market cap range. Lisa, my good friend at AMD, company cap near $800 billion. And Intel is close to $600 billion.

So in a way, semiconductors are hot again and becoming critical. Fifteen, eighteen, twenty years ago, when I invested in semiconductors, no VCs wanted to co-invest with me, except large companies like Samsung, Arm, SoftBank. Now I see many VCs willing to invest in semiconductors, so I'm happy.

Host:

Given the huge investor interest in this area now, whereas previously it was considered too difficult. You are both a long-term operator and did venture capital long-term at Walden. Generally, people have many concerns about semiconductor investment, I'll list a few: it's capital intensive; tape-out success is very unpredictable; you must deeply understand workloads; another factor is high risk of customer switching suppliers.

We've both been involved in some companies that might have gotten design wins, but scaling orders is still a question. And there's cyclicality: you build heavy asset manufacturing capacity, but demand might change or not in a given year.

How do you view why this industry is hard? Simultaneously, there's now long-term demand growth from different areas, like awareness of supply chain diversification importance, and explosive demand growth on the AI side. You're still an investor, now made the biggest bet of your life to become CEO. How do you think about these different risks? How would you advise others to invest in this supply chain?

I know it's a very big question, but given your experience, I think many might have a "YOLO investment" mentality now: e.g., memory is short, buy memory stocks; but simultaneously unwilling to bear things needing ten-year timelines, like materials science.

Lip Bu Tan: Okay, your question is broad. I'll try to explain.

First, venture capital and entrepreneurship are in my blood; I really enjoy the process. Not to boast, but I've had some good exits. To date, I have 159 IPOs, 126 M&As, including semiconductors. If just looking at semiconductors, I've invested in over 200 companies over the years, 38% in the U.S. So I typically look at micro-trends.

Host:

Just to clarify, that's quite remarkable.

Lip Bu Tan:

Thanks, thanks. I just enjoy building these companies. But more importantly, on the investment side, I first look at: Where is the bottleneck? What problem are you really trying to solve?

For example, I invested in a company called Credo Semiconductor, with labs in Australia. At the time, I saw interconnect was becoming a bottleneck, so decided to support it. I also supported Celestial AI doing optical interconnect. Because inside clusters, interconnect speed is becoming increasingly important, so I think optics will be very important. Look at Jensen Huang; he's invested in almost all photonics-related companies.

Also, I look at what solutions the market needs. For example, as we just talked about design complexity and cost, can we use AI and machine learning to drive better design and better solutions? Now several new startups are entering EDA-related areas trying to improve performance. I think it's a gold mine.

And new materials. We talked about indium phosphide, so I invested in Inphi, later acquired by Marvell. You can also invest in some new materials like gallium nitride and silicon carbide. Some of these companies have started being acquired, including one in power management called Empower, doing well in IVR.

Power management is now a bottleneck. For example, dropping from 40 volts to 1 volt, during conversion lots of power is lost; how to improve power efficiency is key. So power, heat dissipation, these are becoming bottlenecks.

Therefore, I always start from "What problem are we trying to solve?" Is the problem real? Are customers truly pained by it? If yes, then I start investing.

Next step, from day one, lock in the first customer. I usually like the first customer to be a hyperscaler because they have scale. If they like your thing, willing to pay millions over the next few years, even give purchase commitments. This is important because with a large customer, you can scale.

So I always look at some formulas: How do you achieve this? Where do you find talent? Sometimes, finding talent is very important. That's why I'm interested in the U.S., Silicon Valley, Austin. Also, Israel has lots of talent. So I've invested quite a bit in Israel.

Because Israel has many disruptive, innovative entrepreneurs; they work very hard. Even during the war, they still take conference calls. Sometimes they say: Okay, there's an alert, I need to go to the basement, network might be bad, maybe we can only use voice. In a way it's even a bit interesting. I really admire that resilient entrepreneurial spirit.

Overall, I think there are many opportunities, especially in AI. Now besides Agentic AI, physical AI is becoming the next huge frontier. You must look at the problem from a full-stack perspective.

That's also why I'm still deeply involved in many frontier models and some investments I support, because I'm very bullish on open-source frontier technologies for physical AI. I think that's a gold mine.

Host:

You mentioned the opportunity to use AI to make certain parts of chip design and testing faster, cheaper, more creative. Combining your Cadence experience, where do you think the most fertile directions are? Anything already working?

Lip Bu Tan:

I was at Cadence about 15 years, I'm happy. One thing I'm proud of is I could find my successor on the road and groom him. Later he became an excellent CEO. He very much embraces AI, using Agentic AI to improve efficiency.

That's the good side. I think Synopsys is also trying to do these things. They got $2 billion investment from NVIDIA; I think that helps them a lot. He also acquired Ansys to get into whole system design.

Overall, these companies are trying their best. But startups also have opportunities to do more disruptive things, eventually either going public or being acquired by these two companies, or Siemens.

So I think opportunities belong to everyone, depending on the entrepreneur's vision. My philosophy has always been: If the entrepreneur wants to sell the company because it's a faster exit path, no lock-up, no quarterly earnings worry, that's fine. Some entrepreneurs want to IPO from day one.

As VCs, I think we three are all VCs; we support entrepreneurs' dreams and help them achieve them.

Host:

Looking at these different directions you mentioned, including future product development, or AI's impact on the semiconductor industry, there are now companies like Periodic doing materials, Purepoint in EDA and design, other links in the manufacturing chain.

Do you think Intel ten years from now, or future semiconductor companies, will be fundamentally different because of AI compared to today? If so, where are the differences?

Lip Bu Tan:

I think so. First, back to your initial question: capital intensive, unpredictable, cyclical. These must be factored into your investment decisions.

I usually like to enter very early, assemble the team. It's fun. I think you do that too. Second, find the right investors to collaborate with. Not always just looking at brand-name firms; I usually look more at the individual. Who truly understands this field? Most importantly, find partners who can stick through tough times and good times.

Many are happy to work with you in good times, but when the company hits trouble, they leave. I like those who truly stay with the company through difficulties. Some successful companies almost went bankrupt multiple times before taking off. So finding partners willing to do that is very important.

Also, look at strategic investors—can they help the company create value in manufacturing, memory, connectivity, etc. I also have friends in growth stage and hedge funds; I like them because they have different perspectives. They understand public markets, can guide entrepreneurs on paths to avoid. This is very helpful.

Overall, it's very interesting. You realize entrepreneurship, like engineering, is about solving problems. At each step, you find people who can help you solve them. If solved, move to the next frontier.

Frankly, looking back, out of ten companies I've invested in, nine will change their business plan halfway because the market changed. So I like entrepreneurs to be a team, not just one person. Second, they must be open-minded, willing to listen, accept our mentorship.

Finally, they form their own plan, not just do what I say. Better state: you give enough feedback, they draw their own conclusions. As long as you agree with their judgment, even if different from your idea, that's acceptable. That's the interesting part of entrepreneurship. They can move faster.

Back to your question, looking forward ten years from now, what kind of companies will win? Just my personal view: companies that can articulate strategy clearly, laser-focus on a niche, find the right partners, and have scaling ability will win.

In a way, this returns to my point about full stack. You need full-stack solutions. It could be a large company transforming into a large platform. Like Jensen Huang, I admire him. He focused on CUDA, software libraries. He said, I want to be a platform company, and he achieved it.

It could also be startups, like Anthropic, OpenAI, they found paths more elegantly, changed the game. Startups move extremely fast, at light speed, and can also become dominant.

Hopefully, Intel can also play such a role, because we have XPU, NPU, advanced packaging, and foundry. Putting these together, you can build specialized chips for different workloads. I'm moving in that direction.

Host:

That makes sense. Part of my question was to understand where you're going, part was to ask whether this will fundamentally change how you work. Because in the software world, I see very big changes happening now: who you hire, who you want on the team, many people are starting to manage multiple agents.

So now many people I know actually prefer to hire people in their thirties, forties, fifties because they're used to managing teams. They think it transfers directly to managing agents, including understanding how to set up complex tasks, how to do QA, etc.

I wonder, in the physical world, or fab environment, how do you see team structure, capability requirements, or changes with AI overlay? Is it a natural slow evolution, or radical changes in some areas? For example, in materials, is it now enough to just use these three models plus some chemistry knowledge? So I'm curious how you view that future world.

Lip Bu Tan:

Good question. Back to my earlier "crawl, walk, run." In the "crawl" phase, you first recruit the best talent from the semiconductor industry. Now I'm starting to think, to build full stack, what software talent do I need to bring in.

Currently my team's average age is probably around 40s, 50s; I need to bring in some new talent. They understand workloads, frontier models, open source; that's important.

Now my son has become my teacher. Every time he invites me to his house, while playing with the grandchildren, I learn from him about AI and machine learning. He goes deeper than me, so I learn a lot, also trying to understand investments and bring some talent in.

We're changing Intel. It used to be a very old-school, traditional, spreadsheet-dependent company. Now I'm transforming it into an AI-enabled company, using AI in design, making the whole organization embrace AI. So it's less dependent on spreadsheets and manpower.

You must combine excellent talent with the best AI tools, not just for organizational management, not just for sales; now I'm also considering marketing, design, all embracing AI.

Host:

I think for many investors, at least for me, the past few years since founding a company, thinking about different funding sources for capital-intensive companies has been a very educational process.

I did many software investments before. If you say, I need $150 million before reaching some critical scale, then you need some very smart friends with completely different balance sheets.

You've experienced this for a long time. You also have unique experience collaborating with the government as a large stakeholder. How do you view this industrial policy? It once created huge successes like TSMC, one of the world's most important companies. But in U.S. business culture, industrial policy has long been unpopular. How do you think this perception should change now? Where is it applicable?

Lip Bu Tan:

That's a good question. Clearly, for capital-intensive businesses and infrastructure projects, you need access to capital. In a way, for early-stage venture capital, many investments are also becoming capital intensive now. Previously, a VC willing to put $1 billion into a single company was unheard of in VC, but it's happening now.

So in a way, you must adapt. I like to look at things with a bell curve. Either you enter very early, because now Series A can have valuations over $1 billion, so you must be at pre-seed stage, before the company reaches $2, $3 billion valuation. That's very rare today, so you must pick right.

The other part is finding capital that can help the company scale. That's why some mutual funds are starting to enter private markets, joining me to invest early-stage. I welcome them because they are less sensitive to requirements like "must own 20% of the company." There aren't that many 20% stakes to give anymore. So you must find the right investors.

In capital-intensive areas, like AI factories and foundry business, you truly need to leverage government funds, sovereign funds, and some very large capital. Now there are large funds specifically supporting infrastructure; we also hope to leverage some of that capital to ensure we can scale operations.

Overall, governments and sovereign funds have become very important. Simultaneously, as a public company, I also intentionally pay attention to more long-term, growth-oriented investors, because they can help me grow the business, not just focus on short-term capital allocation, asking if you're going to buy back stock. Those questions are fine too, but at the same time, I still must build the business. So balance is very important.

What Investors Most Misunderstand About Intel

Host:

At this point, what do you think investors most misunderstand about Intel?

Lip Bu Tan:

Several things. First, back to "crawl, walk, run." The past four months, I've been crawling. But people are starting to realize its potential. Another very important point: we must truly deliver the best products. For example, in PC client, we still have market share. But we really need to build better performance. So I'm quietly building CPU architecture, GPU architecture, and software architecture teams, enabling us to act faster like a culture of multiple startups and achieve leaps with better technology.

Besides products, there's new energy coming in, like Agentic AI, Physical AI. These are huge markets we can invest in.

On the foundry side, we still have a big gap with TSMC, whether in performance or other aspects. So we must stay humble, build those foundational blocks, like the IP, yield, defect density, cycle time I mentioned earlier, make it more efficient, more reliable. Foundry is a trust business. Customers must trust you first before giving you wafers, relying on you. So these things take longer.

But I think by 2030, 2031, people will start seeing our potential. On the product side, PC client is our foundation. Then we'll move into edge computing, into Physical AI and Agentic AI.

Previously, we provided servers and PCs mainly for humans. Now you'll see another dimension: millions of agents, they also need computing power, need to access software stacks. So I think this is an area we have a chance to participate in.

The game isn't over. We can still play in Agentic AI and Physical AI. That's the direction I'm heading.

AI is just beginning. The training part is led by Jensen; edge computing, agents in Agentic AI, and Physical AI, I think are huge opportunities. Everyone has a chance. So that's the direction I want to pursue.

I hope investors understand, although in the past 14 months, we've created 6x return for shareholders, this is just the beginning. We have a lot of room.

Host:

From here it's venture capital-style returns.

Lip Bu Tan:

Yes. I'm always looking for 10x opportunities. As someone with a VC heart, you always look for 10x.

At Cadence, when I was CEO, starting from the interim CEO price of $2.42, by the time I stepped down as executive chairman, created about 85x return for shareholders. Close to 76x, even 85x.

Doing that at Intel is hard because the base is bigger. So I said, okay, then look at 10x. If we can do 10x in five, ten years, I think that's a good return. With a VC heart, that's my goal.

Will Computing Power Always Stay in Data Centers?

Host:

Wishing you success in this very large mission on an already large base. There's an implicit judgment behind your description: where workloads will run. Some say we'll only build larger and larger data centers, 1 gigawatt is just the beginning. Centralized operation, including centralized inference, will dominate efficiency-wise.

But others consider the edge, client-side. Do you believe there will be some equilibrium state for future computing? Or is it solely determined by the workloads themselves? What's your view?

Lip Bu Tan:

That's a very good question. Right now AI infrastructure is being built at massive scale; I think that's correct. I don't think it will slow down because workloads are increasing significantly.

Host:

We're supply-constrained now.

Lip Bu Tan:

Right, supply-constrained. So if anything slows development, it's supply constraints.

But on the other hand, I always look at what all this infrastructure building ultimately serves—what solution, what application. I'm more concerned about the application. If you can identify a huge application, or a few applications combined are meaningful enough, and focus around it, then not everyone building will win. Some will win big, some will slowly fail, or plateau.

Just like the internet era. You saw some companies become huge, like Amazon, Netflix; some became marginalized, disappeared, or got acquired. So for me, the thinking is the same. What's truly important to watch is: what application do they want to serve? How big is that application? Is it sustainable? Is it too crowded?

If too crowded, maybe only one or two remain; others consolidate. So the industry will experience big growth, then start consolidating, maybe one or two become real winners. We've seen this movie before, so I'm not surprised.

Focus on the application. Netflix is an application, Amazon is a real application. In my view, they are winners.

Host:

But you're assuming some of these applications will be better served via client or edge computing than purely relying on data centers?

Lip Bu Tan:

Exactly.

Host:

I also invest in some robotics and defense companies myself, so I know on-device computing is a very important choice. For example, if you have robots at home in the future, what compute you assume at home, what connectivity, determines what you can do. I think in the SaaS era, this was somewhat forgotten.

Lip Bu Tan:

Yes. My investment logic is: find the real problem that needs solving. Second, find players you can partner with. Third, look at the application. How big is it? Is it sustainable? If it's truly big, and you believe in it, then double down, triple down.

Host:

But you also include betting on applications not yet widely deployed.

Lip Bu Tan:

Yes.

Host:

Amazing. Thank you so much for joining today, great to talk.

Lip Bu Tan:

Thank you very much.