Source: Wall Street News
Intel CEO Liwu Chen stated that his return target for Intel is "10x in 5 to 10 years," and he is systematically restructuring Intel's technology roadmap around advanced packaging, new semiconductor materials, and next-generation substrate technologies.
In a recent podcast, Chen elaborated on his path to transform Intel: after stabilizing the balance sheet and focusing product lines, he is shifting investment focus to advanced packaging technology EMIB, glass substrate, and new materials such as gallium nitride (GaN), silicon carbide (SiC), indium phosphide (InP), and synthetic diamond, to address challenges as traditional process node scaling nears its physical limits. He also revealed that the explosion of agent AI and inference scenarios is driving a strong rebound in CPU demand, with the CPU to GPU ratio in data center servers evolving from the past one-to-eight towards one-to-four or even lower.
Chen said that in the past 14 months, he has already created about 6x returns for Intel shareholders, but "this is just the beginning." He expects that by 2030 to 2032, the outside world will begin to truly recognize Intel's potential—not limited to the traditional PC client base, but extending into new markets like edge computing, physical AI, and agent AI.
In his view, if Intel's XPU, advanced packaging, and foundry capabilities can be effectively integrated, they will provide customized chip solutions for different workloads. This is the long-term strategic direction he has anchored for the company.
New Materials Key to Breakthrough, Advanced Packaging and Glass Substrate as Focus
Against the backdrop of traditional process node scaling increasingly nearing physical limits, Chen points to material science and advanced packaging as the breakthrough. He said Intel currently has 18A process in mass production, is advancing 14A mass production, and can see the path to 10-nanometer and even 7-nanometer, but "this road will become more and more expensive and difficult."
To address this, Chen has initiated several investments in packaging materials. He invested in the glass substrate company 3DGS, attracted by glass's unique properties as a thermal insulation material. For chip-to-chip interconnects, Intel is pushing the next-generation advanced packaging technology EMIB and has announced advanced packaging manufacturing collaboration projects in India and New Mexico, USA. Intel holds about 1,000 patents in the module space. How to effectively integrate substrates and modules is the core engineering challenge Chen emphasized.
In the direction of new semiconductor materials, Chen stated he has investments in gallium nitride, silicon carbide, and indium phosphide, with some of the investment targets already acquired by large semiconductor companies like ADI. He also invested in a synthetic diamond wafer company, optimistic about diamond's potential as a thermal insulation material in chip packaging. "That's the engineering spirit—you keep hitting bottlenecks, then figure out how to jump over them or go around them," he said.
Foundry Business: Trust First, Yield and Cycle Time as Core Metrics
Intel's wafer foundry business was once seen by the outside world as unsustainable, but Chen chose to stay the course. He said the core logic behind this decision is: advanced domestic manufacturing in the U.S. has strategic value for supply chain security; no large semiconductor company can have its supply chain highly concentrated in one or two geographical regions.
On the execution side, Chen prioritized foundry business metrics around yield, defect density, and cycle time. He emphasized that foundry is fundamentally a business of trust—"before a customer gives you wafers, they must trust you first." If yield fails to meet standards, losing customers due to revenue loss is hard to recover from.
He also stated that Intel and TSMC are partners, not simply competitors, and the industry overall needs more capacity to meet continuously growing demand. He expects that by 2030 to 2032, the true potential of Intel's foundry business will begin to show in the market.
Terafab Collaboration: Building Semiconductor Infrastructure with Elon Musk
Chen revealed that the Terafab project with Elon Musk stemmed from a shared judgment—semiconductor infrastructure development is lagging behind AI demand growth in terms of capacity, production efficiency, and power efficiency. Under this cooperation framework, Musk decided to build his own wafer fab, and Intel will provide technology and process support to help accelerate production. Chen said he holds weekly meetings with Musk's team, and the collaboration is progressing smoothly.
He also mentioned that Musk has unconventional operational thinking, for instance, once discussing whether smoking would be allowed in certain areas of the cleanroom. "I might not go that far, but maybe in certain areas, the key is to keep an open mind."
Biggest Investor Misconception: Intel Still in 'Crawl' Stage, True Potential Emerges After 2030
Responding to market doubts about Intel's transformation progress, Chen invoked his consistent "crawl-walk-run" framework. He said the past few months were still in the "crawl" stage: Intel is quietly building up CPU architecture, GPU architecture, and software architecture teams, aiming to drive leapfrog innovation at "large startup" speed; on the foundry side, the gap with TSMC remains significant, requiring humility to consolidate foundational capabilities like IP and yield.
"My VC instinct tells me—look for 10x return opportunities," Chen said. He referenced his experience at Cadence as a benchmark: from acting CEO to stepping down, creating about 76x to 85x returns for shareholders. He acknowledged that Intel's larger scale makes it harder to replicate, but "achieving a 10x return in 5 to 10 years" is a clear goal he set for himself.
The following is the transcript of the interview:
Host: Welcome back to No Priors. Today Allad and I are joined by Liwu Chen—legendary investor from Walden, former CEO of Cadence, current CEO of Intel. We'll talk about his plan to transform Intel, what it means for the U.S. government to be a major shareholder, how to be a great semiconductor investor, and whether we can make chips within the U.S. Welcome, Liwu.
Why Take on the Intel Challenge?
Host: Let's start with the obvious question. Taking the CEO role at this incredibly important American semiconductor company is a truly tough job. Why did you take it?
Liwu Chen: That's a good question. I'm 66 this year. Many said you should retire, why take the hottest seat in the industry? A few reasons: One, this is an iconic company, extremely important for the entire semiconductor ecosystem and the U.S.; two, after Cadence, I decided to do one more big thing.
Host: A lot has happened in the past year. What surprised you most?
Liwu Chen: The most surprising thing was something I never experienced in any previous job or training—early one morning, President Trump asked me to resign, citing conflict of interest, no exceptions. I first convinced myself: I don't need this job, I'm doing this purely to save Intel. Putting personal emotions aside, I started thinking what I could do for Intel. Fortunately, I managed to get a meeting early Thursday morning, another on Monday. He listened to my case—I was born in Malaysia, grew up in Singapore, graduated from MIT, lived in the U.S. ever since, never left. I shared this, he listened, and gave me a chance to continue. I'm very grateful.
Host: You said this job is to "save Intel." What does winning, Intel thriving, look like in your mind?
Liwu Chen: It's been 14 months now, a lot has happened. First, changing the culture, clarifying accountability, speeding up decision-making. I'm used to startup speed, everything moves at light speed, but Intel had layer upon layer of bureaucratic meetings, which I had to change. Second, listening to customers—to truly satisfy them, you must be humble, willing to listen, face their issues and solve them. Third, from day one, I decided all engineering teams report directly to me. I'm an engineer, I need to personally know what's broken, what needs fixing. Listening to customers, making them satisfied, ensuring we have the right products, simplifying product lines, and setting a clear roadmap and vision for the next five to ten years.
Intel's Ten-Year Vision
Host: What's your vision for Intel ten years from now?
Liwu Chen: My consistent way—whether at Cadence or Intel—is first crawl, stay humble, listen to customers; then walk; finally run. Step by step.
The first step is to fix the balance sheet—frankly, the situation was pretty bad. I'm glad the U.S. government became a major shareholder. I explained to President Trump: look at Japan, look at Singapore, this is infrastructure, the government should provide support.
Second, I'm very grateful to my old friend Jensen Huang—he invested $5 billion in Intel, and I'm happy to have done some valuable work; his $5 billion is now worth $25 billion or more. Also, SoftBank's Masayoshi Son—I served on SoftBank's board—reached out. Through these, we stabilized the balance sheet.
Next is focusing on products, simplifying product lines, listening to customers, launching next-gen leading products. Coincidentally, agent AI and inference CPU demand is extremely strong now, so in a sense, I've hit a good time. The ratio was about one CPU to eight GPUs for training; now I see it shifting to one to four, or even lower. CPUs have become important, I'm pleased.
I've talked to some AI model developers. They said for reinforcement learning and coordinating all agents' speed, CPUs actually perform better. So there's high demand for my CPUs now. After solidifying the data center server product line, another important business is our wafer foundry. It's capital-intensive, not easy. You need the right IP portfolio—like low-power IP for mobile clients; without that you can't serve them. It's a service business, a trust business—if yield fails, customers walk away due to revenue loss. So I'm very focused on yield, defect density, cycle time, ensuring high quality and reliability to serve customers. Ultimately, moving towards full stack, not just the silicon itself—you need software, some customers directly ask 'give me the whole rack,' you need system-level solutions. I'm advancing these things step by step, quietly, while recruiting the best talent I can find. By the way, all hiring I did personally, no headhunters.
Terafab Collaboration with Elon Musk
Host: Another major initiative widely discussed is Terafab and collaboration with Elon Musk. Can you talk about how that came together and how you collaborate?
Liwu Chen: I think we both agree Elon Musk is one of the greatest entrepreneurs of this century. He and I share a judgment: semiconductor infrastructure development hasn't kept pace with AI growth—whether capacity, production efficiency, or power efficiency, there's a gap, we both see it.
Second, I truly enjoy working with him. He's very unconventional, questioning every step, 'why do it the traditional way,' which is refreshing. I like hearing different opinions, then finding the optimal path together, both sides learning a lot. He has a clear vision—his robots, his cars need lots of chips.
Specifically for Terafab, he decided to build his own wafer fab, and we're happy to collaborate, help him move faster, leveraging some of our technology and process—it's a joint project. His team is great, I meet with them weekly, working with him is exhilarating. He mentioned ideas like allowing smoking in cleanrooms, breaking conventions—I might not go that far, but maybe in certain areas, the key is an open mind, we're seriously listening and evaluating.
Changes in Global Semiconductor Supply Chain
Host: Looking at how AI is driving changes in the global semiconductor supply chain from a macro perspective, country by country, what are your observations?
Liwu Chen: AI's impact on the landscape will surpass the internet, and be more profound. AI first makes you more efficient; with many agents helping, many tedious tasks you used to do can be done faster. In semiconductor design, timing optimization and time-to-market can be greatly improved, costs lowered.
AI demand growth faces several bottlenecks: one is power constraints, some countries simply lack enough power; second is helium impact, many don't realize helium significantly affects semiconductors; third is memory shortage, the most urgent now—even if you expand production now, new capacity takes years. CPUs, GPUs are also in short supply, prices rising, costs ultimately passed to clients.
The most impacted companies are those not embracing AI. AI can help companies improve efficiency in almost all functions. Companies should proactively embrace AI, find better ways to use it—whether prediction, design, or various workloads.
Host: The simplest argument against Terafab and Intel's foundry competitiveness is labor costs and feasibility of domestic manufacturing. What's the logic behind your decision to double down on foundry?
Liwu Chen: When deciding whether to bet on foundry or exit, there was lots of noise—too expensive, won't work. But I concluded: this is extremely important for the U.S., for the whole industry.
We've all experienced supply chain challenges. Any large semiconductor company must seriously consider supply chain issues, must have a robust, resilient supply chain, can't rely entirely on one or two geographically concentrated suppliers. More and more will realize domestic U.S. manufacturing is crucial.
Our most advanced process, like 18A (1.4nm class), we're already planning 1nm, 0.7nm. Process nodes get smaller, linewidths thinner than a human hair, extremely complex, any misstep can ruin everything. Because of this, manufacturing precision requirements are higher, becoming more of a bottleneck.
We greatly respect TSMC, we're good partners, and the industry needs more capacity to serve customers, so we decided to grit our teeth and persist—long term it's critical, and where I can create more value for the industry.
Physical Limits and Advanced Packaging
Host: People have long discussed physical limits of chip scaling, linewidths too narrow to shrink further. When do you think we'll truly hit the wall?
Liwu Chen: We have 18A now, advancing 14A mass production, I can see paths to 10nm and 7nm—the road can be traveled, but it gets more expensive, more difficult. That's why we need partners, need close collaboration with substrate suppliers, equipment makers, to jointly push yield and performance.
Another key area becoming a bottleneck is advanced packaging. TSMC has CoWoS, we have a next-gen solution called EMIB; I must ensure it meets customer yield requirements in mass production.
When traditional scaling hits bottlenecks, I start looking for breakthroughs at the material level—GaN, SiC, InP, I have investments in all three. In packaging materials, I started looking at glass—glass is a good thermal insulation material, I invested in a company called 3DGS. Intel has about 1,000 patents in modules; how to integrate substrate and module is a key challenge. We also recently announced advanced packaging manufacturing collaborations in India and New Mexico, USA. Also, I'm looking at synthetic diamond—another excellent thermal insulation material, I invested in a diamond wafer company.
That's the engineering spirit—you keep hitting bottlenecks, then figure out how to jump over or go around them. Having been deeply involved across the full semiconductor lifecycle, from EDA tools to design to manufacturing, I'm now happy to apply that experience to contribute to the industry.
Host: Is there a possibility that process node convergence flattens performance differences between foundries, forming some asymptote?
Liwu Chen: Moore's Law essence is transistor density doubling, but power and cost don't drop proportionally—you can double performance, but area and cost may not decrease equally. Unless you find new materials, new design methods. That's exactly why I'm increasing recruitment of material science talent—it's become core to innovation in this field.
Eighteen years ago when I was still investing in semiconductors, many top VCs were completely uninterested. I remember, after presenting on semiconductors at a partner meeting, half the people left, the other half asked 'do you have any software or services projects,' eventually only one or two stayed out of sympathy. Now, Jensen Huang's Nvidia has a $5.3T market cap, Broadcom and TSMC each around $2T, my good friend Lisa Su's AMD near $800B, Intel around $600B. Semiconductors are hot again, indispensable infrastructure. Fifteen to twenty years ago, almost no VC wanted to invest in semiconductors with me, except big institutions like Samsung, ARM, SoftBank. Now VCs flock, investment enthusiasm is extremely high, I'm very pleased.
Challenges of Semiconductor Investing
Host: You're both a long-term investor and operator. Semiconductor investing faces many difficulties—capital intensive, unpredictable outcomes, deep understanding of workloads needed, high switching risk for customers, cyclical industry... How do you view these risks, and how would you advise others on where to invest in this supply chain?
Liwu Chen: Venture investing entrepreneurship is in my blood, I truly enjoy it. I'm not here to boast, but for some background: I have records of 159 IPOs, 126 M&A exits, over 200 semiconductor investments, 38% in the U.S.
In investment methodology, I always start with a core question: where is the bottleneck, what problem are you solving. For example, I invested in Cradle Semiconductor because interconnect became a bottleneck; I invested in Celestial AI because optical interconnects within clusters are becoming critical—Jensen Huang invested in almost all photonics-related companies, not a coincidence.
At the design level, can AI and ML help reduce complexity, improve design quality—I think EDA is a huge opportunity, several startups are moving this direction, a goldmine. In new materials, GaN, SiC, InP are my investment directions, some already acquired by large companies like ADI. Power management—losses from 40V to 1V conversion are huge—also a bottleneck sector I favor.
My investment framework: Is the problem real? Are customers really struggling? Then very important: who is the first target customer? I lean towards hyperscale customers—they have capability and willingness; if they like your stuff, willing to pay millions over next few years, maybe provide some guarantees, because after landing a big customer you can scale.
Talent is also critical—U.S., Silicon Valley, Austin, and Israel are my focus areas. Israel has very disruptive, innovative entrepreneurs, working extremely hard. Even during wartime, they persist with meetings—sometimes saying 'there's an alert, I need to go to the shelter, internet might be poor, let's switch to voice,' such resilience I deeply admire.
Now besides agent AI, physical AI is the next major frontier; need to seriously look at full stack, which is why I'm still deeply involved in many frontier model-related investments—I'm very bullish on open-source frontier technologies for physical AI, it's a goldmine.
Experience at Cadence
Host: You mentioned AI enabling faster, cheaper, more creative chip design and testing. Based on your Cadence experience, which directions are most fertile? Anything already working?
Liwu Chen: I was at Cadence nearly 15 years. One thing I'm proudest of is finding my own successor, grooming him personally; he's now an excellent CEO, actively embracing AI, bringing agent AI into tools to improve efficiency. Synopsys's Sassine is doing the same, supported by Nvidia's $2B investment, and acquiring Ansys to expand into full-system design.
Large companies are doing it, but there's also room for startups to do more disruptive things, ultimately these companies can IPO or be acquired by the two big companies. It depends on the entrepreneur's vision. My consistent philosophy: if an entrepreneur wants a quick exit, help them achieve it; if they want IPO from day one, help them go that route. As VCs, we support entrepreneurs' dreams, help make them happen.
Scaling and Investment Decisions
Host: The directions you mentioned—material companies, EDA, manufacturing—looking 10 years out, will Intel or future semiconductor companies be unrecognizable due to AI?
Liwu Chen: I think so. Back to capital intensive, unpredictable, cyclical—these need factoring into investment decisions. I usually like to enter very early, build the team; find the right investors who can stick through tough times, not just fair-weather friends; also look for strategic investors, whether adding value in manufacturing, memory, interconnect, other dimensions. I also have growth and hedge fund friends with unique public market perspectives, helping entrepreneurs avoid pitfalls, very valuable.
Frankly, looking back, out of 10 companies I invested in, 9 changed their business plan mid-way because the market shifted. So I prefer entrepreneurs with a team, not just one person. Also need open-mindedness, willing to listen, accept our advice, but ultimately form own judgment—the best outcome isn't 'he told me to do it so I did,' but you gave enough feedback, they derived conclusions you agree with or understand, that's the joy of entrepreneurship.
Looking back in 10 years, winners will be those who can focus on a niche, find the right partners, and scale. Have full-stack solutions, that's important. Large companies can, like Jensen Huang, focus on CUDA and platform, all-out build a platform company, he did it. Startups can also, like Anthropic, OpenAI, change the game elegantly; startups can move at light speed, truly become dominant.
For Intel, I hope it can play such a role—we have XPU, advanced packaging, foundry, if we integrate these, tailor chips for different workloads, that's my direction.
Team Restructuring in the AI Era
Host: The software industry is changing a lot—who to hire, who's good at managing multiple agents. Many now prefer hiring people in their 30s-50s, used to managing teams, that skill transfers directly to managing agents. In the context of hardware or fabs, how do you see team structure and capabilities changing?
Liwu Chen: Back to crawl-walk-run framework. In the 'crawl' stage, I recruited the best semiconductor talent; now I'm thinking what software talent to bring in to build full-stack capabilities; also noticed average team age is 40s to 50s, need to bring in some younger talent, help them understand workloads, understand frontier open-source models.
Interestingly, my son is now my teacher. Every time I visit his home to play with my grandkids, I ask him about AI and machine learning, he knows more than I do. I've learned a lot, then try to turn that into investment decisions and talent acquisition.
Intel used to be a very old-school spreadsheet-dependent company, I'm transforming it into an AI-empowered enterprise—not just in design, but embracing AI across the entire organization, reducing spreadsheet dependency. We need to combine seasoned technical talent with AI tools, not just in sales and marketing, now design is actively embracing AI too.
Industrial Policy and Capital Sources
Host: For capital-intensive businesses, accessing capital has always been a big issue. Industrial policy created companies like TSMC, the most important, but this approach was long frowned upon in U.S. business culture. How do you view this?
Liwu Chen: For capital-intensive businesses and infrastructure projects, capital access is crucial. Now some VCs are willing to put $1B into a single company, unimaginable before. So in early-stage strategy, either get in very early when valuation is reasonable; or do Series A, but now Series A valuations exceed $1B, difficult.
Capital that can help scale, like mutual funds—they're less sensitive to ownership percentage, I welcome such investors. For capital-intensive projects like AI factories, fabs, must seek government funding, sovereign wealth funds, or large infrastructure fund support. Sovereign wealth funds and government capital will become increasingly important.
As a public company, I'm also consciously focusing on more long-term growth-oriented investors, not short-term funds asking 'when will you buy back stock' every quarter—of course shareholder returns are legitimate concern, but I must also build the business, balance is important.
Investors' Biggest Misconception About Intel
Host: What do you think is investors' biggest misconception about Intel now?
Liwu Chen: A few points. First, back to crawl-walk-run: past few months I was still crawling, but people are starting to see potential. On product side, we still have market share in PC client, but must dramatically improve performance—so I'm quietly building CPU architecture, GPU architecture, software architecture teams, preparing for leapfrog leadership, moving fast like a large startup, leveraging better technology for leaps.
In foundry, our gap with TSMC is still big, we must stay humble, focus on building foundations—IP, yield, defect density, cycle time—making foundry more efficient, reliable. It's a trust business, before giving you wafers, customers must trust you first. These things take longer, but I think by 2030 to 2032, people will start seeing how big Intel's real potential is.
PC client is our base, but we're extending to edge, to physical AI and agent AI. Before you provided servers and PCs to humans, now there's a whole new dimension—millions of agents needing compute access, needing software stack access. I think in both agent AI and physical AI directions, Intel has opportunity, the game isn't over.
AI is just beginning. You have Jensen Huang dominating training, you have edge, agent AI, physical AI—a huge opportunity, everyone still has a chance, that's where I'm going all-in. Past 14 months already created 6x returns for shareholders, but just the beginning, plenty of room left.
My VC instinct tells me—look for 10x return opportunities. At Cadence, from acting CEO to retirement, stock from $2.4 rose to bring about 76x returns for shareholders; after executive chairman tenure ended, roughly 85x. Intel is larger, harder to replicate, but my goal is 10x—10x return in 5 to 10 years, as someone with VC in my bones, that's my target.
Where Will Compute Reside?
Host: There's a view that data centers will get bigger, gigawatts just the start, centralization is mainstream. But your described business picture also includes edge and client compute. Where do you think compute ultimately distributes between data centers, edge, client, or entirely determined by application workloads?
Liwu Chen: Current massive AI infrastructure buildout is correct, I see no reason to slow down, because workloads keep growing. Current limiting factors are mainly supply-side—any slowdown comes from supply constraints, not demand side.
But I'm more concerned: after all this infrastructure is built, what applications will run on it? You must find truly scalable applications—like the internet era, Amazon, Netflix emerged, others disappeared or got acquired. AI industry will undergo same process: big growth then consolidation, eventually one or two real winners emerge.
Focusing on applications is key; Netflix is a real application, Amazon is a real application, they won. And certain applications are indeed better suited for edge or client—robotics, defense scenarios, device-side compute choice is critical; assumptions about connectivity, device capabilities determine what you can do. This was somewhat overlooked in the SaaS era.
My investment method: find real problems, find right partners, assess if application market size is sustainable—if you truly believe, double down, triple down. Of course, this includes betting on applications not yet widely deployed.
Host: Thank you so much for being here, this was truly an enjoyment.
Liwu Chen: Thanks for having me.
