You have a very unique record in my life. Do you know what that is? The speed at which I have developed a friendship with you. I don't think I've developed with anyone else. And it's not a function of me. I think it's a function of just how good you are at doing the kind of amount of things that you do on a regular basis. But thank you for coming here.
Well, thank you for inviting.
Tell us a little bit about Intel. It's a national treasure. Intel has gone through some tough times. You are bringing it back again to life, but there's so much that needs to get done. Walk us through the current state of Intel. Where do you think it is? Where do you think it should go? How should people think about it?
Yes, I think first of all, I joined the board for two years and learned some things going on. A lot of good friends of mine from Intel, and it took me a while to decide to take this role. They say, you know, that's not easy. A lot of my friends told me don't do it. You have a great reputation in the venture investment, you love it, why take on this one? But finally I decided, you know, this iconic company is very important for the industry and also for the United States. So I finally convinced my wife to let me do it one more time. After Cadence, she supported me to do it one more time. And just to update, I think I just passed 10 months, coming to 11 months. And you know, I kind of call it marching off the map. So something that some of the surprises I don't even know, you have to kind of react at the real-time basis and learn from it. So in some way I think it's good to have. And then just to update, we have a very complex company. You have a foundry that's strategically very important, and then the other part is the product. And then, you know, so clearly you need to balance the two.
And do you believe your foundry business becomes a general-purpose foundry that's outside of just your products?
Yes, and so that's the intent. And we have the 18A. When I took over, the yield was quite poor. So I really got all my friends to help, from PDF Solutions to KLA and the equipment, to make sure that we have that 7 to 8% yield improvement per month.
8% yield improvement per month?
That's right, and that will be the best practice. And it took me a while, finally I'm now seeing 7 or 8% yield per month. So in a way, it's very helpful that you open up the kimono and then let people come and help you. So I think one is 18A, we just announced the Penta. And then I can really count on my fab to produce that. And then interesting enough, a couple of customers are knocking on my door now, saying, 'Hey, it seems like your 18A is doing well. We want to be part of that.' So I think we are delighted to see the customers want to see that, but we are laser-focused on the 14A, and that will be, you know, it's very important, it's 1.4. And this is the most advanced, and we're going to be in risk production 2028, and then we'll be in volume production 29. And like any foundry, I learned a lot. You need to have not only the yield improvement, you need to have the variation so they're more predictable, and you also need to have the final product, what the yield looks like and the die spin and all the details. And second part, you need to have all the IP available. You know, if you want to have mobile customers, you got to have all the low-power IP, otherwise you can't serve the customer. So there was something missing in the past, so I really put something in place so that we can have both and then serve the customer. Good news is a couple of customers are very excited. 0.5 PDK this month, we will have that, and the customers can use a test chip to work on us. So I think good news is a couple of customers are engaging heavily, and we're looking forward to serve them. Hopefully, second half of this year, I will see the volume commitment from the customers, which products they want to have, and I can serve them.
On second half, you'll see the volume commitments on the foundry or on your core products?
On the customer, on the products. So in a way, they need to let me know what products and what volume they need.
Then I can. And one thing people will see, because first of all, I'm not going to announce any customer. This is confidentiality. We need to support the customer. And what we're going to do is when you see me starting to put money into substrates, it's a very important material. And then secondly, some of the capex equipment I need to have and scaling it, that means that I have real customers I've committed to it. That's the kind of discipline I have.
So on the fab side, to get to mass scale, what does mass scale look like for you, and what's the time frame that you feel like mass scale could happen where you could be a formidable fab for third-party chips?
Yes. So it's very challenging to operate the two, because one is on the product, you have to drive innovation, drive the product roadmap. The other part is really a service business, culture had to change. And in some way, I kind of tell my team this is a business of grinding. You know, it's not like you have good innovation, good technology, people will come. You have to grind until something that you prove the yield, they can count on you to make the production, and then their whole revenue and profit depend on it. That's a lot of trust and responsibility, and we have to earn it. So I think those are the things. Scaling is important, and that's why I always ask customers, give me the biggest product that you have, the most important product, give me 5%, 10%, 20%, 50%. Let me earn it and let me earn your trust. And then many of them, I worked with them when I was running Cadence, so they trust me. So somewhere, I think back to your point, trust is very important for AI. So I think something that's even more important in foundry is the customer trust.
Now, if you think about the constraints that are there right now, do you think that the energy constraint with power, cooling, all of that is more severe, or right now is manufacturing capacity? Which one limits growth first?
I think in terms of the AI, the biggest challenge I think for a lot of my customers is memory. Memory actually, there's no relief as far as I know. When I talked to the, you know, only three key players, two of them I talked to very frequently, and they told me, 'Lip-bu, there's no relief until 2028.'
Because I think this whole AI is sucking up a lot of memory. You know, our mutual friend Jensen, you know, you talk to him, you know, clearly he needs a lot of memory for his roadmap in the next-generation product. And so I think it's very important to have that memory. If anything is going to slow down, it's going to be the memory. So that I think is one. Secondly, I think clearly, from the compute side, I was very happy to hear that customers are all crying for more products, and I didn't prepare the production enough to meet their requirements. And so I think people are starting to find out that in application, CPU actually is more useful in terms of performance for all the compute requirements. And in fact, a friend of mine mentioned that Moore's Law used to be three to four years, and now it's like three or four months. So the increase of compute is increasing so much. And right now, my biggest challenge is focused on our production and supply chain, make sure we can meet their requirements. Almost every CEO, they call me up, 'Lip-bu, can I have more? I'm your friend, I'm your customer, the most important customer, I want to have more of that.' So I think somehow it's kind of encouraging for me to see that compute has become very important. And then the other part really important to me is beside memory, beside the compute, and then the other part is the thermal. We just talked about the cooling technology. So in terms of some of the high-performance processors, either GPU or CPU, you sometimes have to guide the gigahertz down because of your thermal issue and the power management. So those are the ones that you want to find new ways of cooling. In terms of, you know, liquid cooling, air cooling is not cutting it now. So liquid cooling, microfluidic cooling, immersion cooling, I think those become very important. The other part that you mentioned, I think something that Cisco has is really good, is that interconnect. Yes, it's getting very critical. Used to be more in the copper, you know, like Credo Semiconductor and Astera Labs I invested in, but now moving to optical.
And so optical becomes very critical, and so that's something that is a new wave because the speed and latency become very important. So that's another issue. The part of the constraint is the software. And so if you really want to address some of this, like Moore's Law, you need to really look at the whole full stack. So not just the silicon, you need the cooling, and then you need to have the interconnect. And then the other part is the software. You know, the open source, I'm a very strong believer in it. And then of course, code compatibility is very important. And even more important right now is the cluster of GPU or CPU, and then you become the cluster manager, the software. And right now, you know, sometimes we have problems, you don't even know where to look for it. And now there's a lot of new startups coming to me saying that will be an area that we want to work on, so that we know exactly how to solve some of these clustered issues that you're going to face. Kubernetes is great, but you cannot find the real issue there. So I think those are kind of interesting. Being a venture capitalist and running CEO of Intel, it gives me that kind of perspective of what are the new things coming. And quantum computing is another big area coming up, it's around the corner. And so something that is interesting to see after AI, you know, we talk about agentic AI, the next big wave is physical AI, the next thing is going to be quantum. So I think that's kind of exciting.
And so you talked about open source. Talk to us a little bit about your perspective on open source in the United States. How important is it? What needs to happen?
Yeah, very good questions. You know, I'm very passionate about education, and you know, I'm involved with MIT, CMU, and Berkeley, and others. The key part is, you know, one thing that I worry a lot is the fundamental research. We're starting to not continue. A lot of universities, the best professors are being recruited out to Asia, to Europe. And so I think the fundamental research, some of the public companies like Cisco and ourselves, we have some short-term, midterm requirements. You cannot invest into long-term, 10 years, 20 years from now. And so that part to me is an outcry in terms of fundamental research. The other part is the open source. I'm a big fan of open source. And frankly speaking, a couple of very knowledgeable people told me, 'Lip-bu, we are behind China now.' And so that's something that we have to pay attention. I think DeepSeek is just a wakeup call.
But don't you think, like to some degree, some of the models that are emerging in China are just pure distillations of the models that are here? And so at some point in time, if all you're doing is distilling the models, then the moment they can't really get that much farther ahead, or you don't think that's the case? They'll actually figure out a way to start creating original.
Yeah, very good question. I thought that initially, I thought that they will be falling behind because they don't have the access to the most advanced GPU from Nvidia, most advanced processor from many different players. But you'll be surprised. Recently, I was just trying to recruit some of the top talent in CPU architecture, and I found out that Huawei has 100 CPU architects, top-notch.
And so when I talked to them, 'Why are you going there?' And they said that even though we don't have access to the best tools like EDA tools from Cadence and Synopsys, but we have the poor man's way to do it, and we can do it. And I said, 'What about ASML equipment that you guys don't have?' But they said, 'Well, we are quietly building on it.' And so to me, they are just shortly behind us. And then if you're not careful, they will just leapfrog ahead of us. And what they've done is they've spent the calories on infrastructure optimization because they don't have the right chipset.
And so like if they are at 7 nanometer and we are at 2 nanometer, what they're offsetting that with, and they have unlimited power.
