You're watching the context. It is time for AI decoded. Welcome to AI decoded. The whole purpose of this program is to lift the lid on artificial intelligence to help you understand what it's all about. And tonight we are going to go under the hood. None of this modern artificial intelligence would be possible without the highly specialized chips. You will have heard of NVIDIA, which alongside Apple and Amazon, is now one of the biggest companies in the world. You might not have heard of Groch. They are the upstarts in the market, the challenger that so far has raised close to a billion dollars in private finance to develop a new and alternative chip to the one NVIDIA supplies. Ai chips are essential if we are to expand AI at scale. But just as important are the data centers that power and train them. They are the digital factories of the future. They need power and lots of it. Saudi Arabia, as the world's biggest oil exporter, has spotted an opportunity. Suddenly, they are building vast desert data centers so big they might one day be capable of reaching half the world's population. And this week in Riyadh, at the AI Global Summit, they signed a deal with Glock, who will supply the chips for their advanced AI systems.

Coming up, we have an exclusive interview with Jonathan Ross, who is the CEO CEO of Glock, who will tell us about that deal that he secured, and alongside me here in the studio, the brains of the operation, Priya Lecarny. He says that here anyway.

Highly artificially intelligent.

Ceo of AI education company, Century Tech. It gets bigger and better that every week, I've noticed. Right, listen, before we go any further, I want to play a little film for our viewers because I want them to understand the difference in the AI chips that the big players like NVIDIA are manufacturing, and this new type that's being developed by grog. So have a look.

This is what a standard AI chip looks like. It's a GPU. You'll notice the complex crossing lines. This design is responsible for the drastic rise in value for its producers, who made the right bet that GPUs could be the engine for AI. Now, this is what a far from standard AI chip looks like. Uniform, clean, and really, really fast, like 18 times faster than its competition. But is this underdog capable of taking a bite out of the hot AI chip market, computational needs were changing fast as self-driving, artificial intelligence, and more were being developed at a rapid pace. A new architecture, more specifically, specialist chips, would be needed to pull that off. Something tailored deeply for its specific task. This is unlike anything on the market. The millisecond you hit enter, this thing is giving you the entirety of its response. To further put this speed in perspective, Groch chips enable LLMs to write a full book in about 100 seconds.

Goodness me. Before we talk to Jonathan, tell me about the chip E ecosystem? What's going on here? What's the context of it?

The reason why this is really exciting and this new technology that Jonathan and his team have developed is really exciting is because there are very few global players in this ecosystem. You have what we call foundries, which are the big manufacturing facilities that manufacture the chips. The biggest foundry for the advanced AI chips is in Taiwan, it's TSMC. Then there are a couple of other foundries as well. Then you have what we call fabulous companies. The fabulous companies are NVIDIA, AMD. They design the chips, but they don't actually manufacture the chips. Then you get companies like Samsung and Intel who essentially have more of the end-to-end ecosystem. They design the chips and they manufacture them. As you'll see, they also start to provide that manufacturing facility to other companies as well. But there's very few players. The market's absolutely massive. But what's interesting, Christian, is NVIDIA has about 80%, let's say, of the market. They also do data centers and other things, but they are the dominant player in the market. So Groch has come in with this technology. They've done this deal in Saudi Arabia with Aramco, and they are, I'm sure Jonathan is going to tell us, vying for some of that market share.

And What's going to be really interesting is what's his strategy?Designer.

And manufacturer?

Well, he'll tell you they're the designers. Actually, they've picked Samsung, and he'll correct me if I'm wrong, but Samsung as their manufacturer, which I do think is really interesting. Why did they choose Samsung? And what was the reason the decision-making process? Was there a geopolitical reason as to why they didn't pick a Taiwanese company, TSMC? We'll talk about the geopolitics of this later because it's really complex and truly fascinating as well at the same time. But that's the ecosystem. Him, but there's huge and huge amounts of money and investment here. Startups can't just go and build a factory. It costs billions and billions of dollars. Tsmc have just opened a factory. They started a factory in Arizona. They haven't finished, completed it because it costs an enormous amount of money, and seeing some issues while they're trying to build that. It's going to be really interesting to hear from Jonathan, what's his strategy and how is he actually going to try and muscle in on some of that NVIDIA market share?

Well, let's bring him in. Jonathan Ross, CEO of grog. Welcome to the program. I should say, congratulations because you've just signed this immense deal with Aramco. Tell us what it is.

We've signed a deal to deploy, starting this year, 20,000 of our chips in the kingdom. Next year, we have an MOU to deploy up to 200,000. Harrison, NVIDIA deployed about 500,000 their end GPUs last year. Now, of course, this year, they've increased it to 2 million, but that still makes us a pretty sizable player.

Right.

I'm guessing that the new chips are not in this communication that we have because your picture is frozen, but we'll press on because we can hear you very well. What is your unique selling point of this chip over the others?

It's B. The biggest selling point is, you remember a dial-up when you used to have to wait a very long time for your answer to arrive and you would see it show up very slowly? With this chip, you get an answer that's almost It's almost like broadband, and it changes the experience, and it makes it much more engaging.

Jonathan, what's your strategy to carve out a share in this space? Because you've got NVIDIA that's going to come out with a Blackwell platform, hopefully in Q4 of this year, and they're claiming 25 times less cost, it's more energy-efficient, and it's a lot faster.

Well, the amazing thing is we built this chip on what's considered fairly a really old technology, 14 nanometer, and you measure it in the size of the things that are on the chip, the transistors and the components. And NVIDIA's latest chips are 4 nanometer, and yet we're actually still faster. Now, you mentioned the Samsung FAB that we're going to be building our next generation chip with. We're actually going to be going to four nanometer as well. And it's like we get to skip three grades in terms of our capabilities when we do that. So already the world's fastest. The other thing is we're much less expensive. A modern day GPU has expensive. They also have a lot of other very complex and expensive parts that go in there that they use to connect up that memory, and we don't have any of that. But also because we're using an older process, 14 nanometer for our current generation, and even by the time our V2 comes around with a 4 nanometer, it's going to be pretty widely available technology. It's going to be underutilized. Whereas right now everyone's trying to use the latest process, and so it's really hard to ramp up the supply.

Just for someone who's not deeply involved in this world as you and Priya are, Priya was explaining to me that you are... These chips are going to be looking primarily at text, but how might they be used in industry in the future? As we move to facial recognition, simulation technology or to driving robot technology, how do you think your second and third versions of this chip might develop?

Actually, our current chip, we just launched what's called Multimodal Model last week, where it takes both text and images in, and it outputs text about the image. So for example, you could upload this image and ask how many How many lights are in the image or how many lamps? And we'll answer that. This is very useful for commerce. For example, you could upload the image of a product and create a description for it. And so that's what some customers of ours are going to be using it for. We also do speech to text. So we're very general. Really, what's different about our chip versus GPUs in terms of what you can use it for is that it's used for something called inference as opposed to training. And the difference between those two when it comes to AI, if you want to become a heart surgeon, you spend many years training. And what GPUs are really good at is training the models. What we're really good at is inference, which is the heart surgeon doing actual heart surgeries, the practice. And you spend a lot of money on the training, but you make your money on the inference.

It's also much more cost-sensitive because every single time you use one of these AIs, whether it's a chatbot or whether it's something else, house, you actually are consuming compute. And so where we operate, if you want to think about AI in terms of the way the industrial revolution happened or transportation, you could think of the model as a car, but it has to run on oil or petrol. And what we're providing is the compute, which is the new oil. You can have the best AI model in the world, but if you don't have enough compute, you can't run it. And one of the astounding things is In about the last couple of months, about four or five months, we've gone from fewer than 10 developers to as of today, we hit 450,000 developers on our platform. It took NVIDIA about seven years to get to 100,000. We got there in six weeks. And a large part of that is because what NVIDIA did, they created some amazing models that people want to use. But because we run them so much faster and because most of the applications are built to use the AI, there's been incredible demand.

And so the reason that we're here in Saudi Arabia is because we need to build out a lot more work and compute, and they have the financial resources and they have the energy.

Jonathan, a question about being in Saudi Arabia in this deal, this MOU, I suppose, the memorandum of understanding that you've got in Saudi Arabia with Aramco. It's intriguing because I know that the US obviously wants its companies to have a huge, addressable market. They want you to do business outside of the US with lots of countries. They want those countries to be cozy up to certainly the US rather than China when it comes to building their AI capabilities. But I'm looking at a parallel example of when Microsoft invested in G42, the UAE company. Washington was very clear that they needed to stop working with Huawei G42. They needed to sell the stock invested in their company that was invested by Chinese investors. You've now got this MOU with Aramco. We know that in Saudi Arabia, they're quite close to the Chinese. They've had deals between the PIF, the Saudi Public Investment Funds. It has an interest in You've had a deal in Lenovo. You've had an investment from Aramco's Prosperity7 in an LLM in China. Have US officials been in touch with you? Do you expect them to be in touch with you and Aramco?

What are you What do you think about this tension that now exists potentially with this deal that you've struck?

We've been in contact with the commerce department since before this deal happened. Just to be clear, we actually are beyond the MOU stage. The project is now budgeted for the initial 20,000 chips. The MOUs for the remaining 200,000 or Chinese companies well in advance of the change in posture from the US government. A big part of the reason why was it was very difficult for tech companies to be successful in China. I happened to start the Google TPU, which is the chip that Google uses for AI. And while I was there, I saw that Google, Meta, and many other companies that were very strong companies had a lot of trouble operating in China. What would happen is as you started to become more successful, things, for example, if you were doing search, your search engine would get slow, and people wouldn't know why. It was the great firewall. It would just be slowed down. And it was very hard to compete, not for competitive reasons, but because the government really didn't want you to win. And we, as a small startup, didn't feel that we were going to be in a position to compete in China.

So we voluntarily to decide not to for purely commercial reasons. Maybe that's part of why we can do this deal. We're very on-side.

Jonathan, we'll have to leave it there. We're up against a break, but thank you very much for coming on and explaining it all to us. Best of luck with the deal that you signed out there. We'll talk about the geo-politics of all this after the break. We're going to speak to Gregory Allen, who used to work in the joint AI Center at the US Department of Defense. You're watching AI Decoded. Welcome During the Cold War, the United States would only supply supercomputers to the Soviet Union if they were used for weather forecasting and not nuclear simulation, and there were permanent foreign monitors deployed to Russia. All data was open for analysis by US intelligence. Those supercomputers, the AI models of the future have both civilian and military implications. What is the subject to the same controls we've just been talking about that? There is a blanket ban on selling restricted chips to China. What does Washington to think about all this? Let's get into it with Gregory Allen. He's a director of the Wadwanic Center for AI and Advanced Technologies at the Center for Strategic and International Studies. He was formerly with the US Department of Defense.

You'll have heard that last conversation, most of it we got, about whether if you're doing a deal with the Saudis, there's a backdoor to the Chinese. How will the US government see a deal like that?

Well, I think the first thing to begin understanding is that leaders in the United States, leaders in Saudi Arabia, leaders in China, all agree that leadership and artificial intelligence technology is foundational to the future of military and economic power. They are all thinking about how they can increase the competitiveness of their economy, and especially in the case of the United States and China, thinking about how they can gain an edge over one another. For the United States, which has the most advanced chip designing companies like NVIDIA, like Roq, based in its homeland, they are seeking to deny China access to the most advanced chips that can be used to train the most advanced AI models. There has been reviews of Chinese military procurement documents, and they are looking for American GPU technology, and that can be used to train hypersonic missiles and run simulations of how they can deliver nuclear weapons, or it can be used for more commercial applications, such as large language models like ChatGPT, which are used by hundreds of millions of people around the world. The point is that this is a dual-use technology, and the United States government reviews exports of chips such as grox around the world under those terms.

Gregory, and there's reports of some of those chips being smuggled into China, right? We also had an arrest recently of an ex-Samsung executive.

Yes. Understandably, China wants access to these chips, whether or not it's legal to get access to these chips. Chinese companies are now engaging smuggling networks to acquire illegally what they can't get legally. Now, that's true of the chip hardware. It is also true of the critical expertise around the semiconductor industry more broadly and AI, specifically. In the case of Samsung, these are two former executives accused, and at real risk of conviction, of transferring critical know-how in advanced semiconductor manufacturing techniques to Chinese companies in violation of intellectual property laws and also economic security laws in South Korea.

It feels like a nuclear arms race.

Yeah. Well, nuclear technology is very explicitly weapons-focused. Really, you have nuclear technology in the energy domain and the weapons domain, and a handful of niche medical cases, but it really is, first and foremost, a military technology. Artificial intelligence is much closer to computing in general, which is it's very useful for running your Microsoft Word application, but it's also useful for running military supercomputers for code-breaking. That's the challenge here.

I think the reason why I'd agree with Christian that it feels like that is because many years ago, nuclear power and the ability to have nuclear capabilities really was a way of asserting your dominance on the global stage. It seems that although that's obviously still before the case, now AI is very much the other area where they're focused on national security and cybersecurity. Gregory, before we finish here, I do want to ask you because China is at the center of this. What, in your view, is the role of TSMC, which we just talked about earlier, being the lead provider of these advanced GPUs. They work with NVIDIA, they work with AMD, amongst many other customers of theirs. What's their role in terms of China's strategy about whether or not there is a future invasion of Taiwan?

Well, you've heard previously in the earlier conversation that NVIDIA has very high market share, north of 80% for the AI chips that are used to train these AI models. But where are those chips made? They're made in Taiwan. Taiwan's market share for logic chips at the most advanced manufacturing nodes is essentially 100%. It's the sole country on Earth that can operate the most advanced semiconductor manufacturing processes. That is a critical dependence, not just of NVIDIA, but of the entire global economy. China, of course, stating that Taiwan is rightfully their territory and stating repeatedly their willingness to use military force if necessary to retake Taiwan, that is an incredible hinge point of global security and the global economy. If a war was to devastate Taiwan's semiconductor industry, which I think is a very real possibility in such a conflict scenario, that would be a global economic catastrophe, and no one would be spared from the consequences of that.

They say it estimated at about 1 trillion global economy damage per year. Per year?

Yeah. Fascinating, isn't it? We talk about what AI can do. We never talk about actually the chips and what is going on around the world to get the best chips and to smuggle them into countries like China. It's extraordinary stuff. Jonathan, thank you very much indeed. That is it for AI decoded this week. Thanks to Jonathan. Thanks also to Gregory and of course to Priya for her expertise as ever. We will do this same time next week. Just a reminder that all the episodes that we've done so far are on the AI decoded playlist on YouTube. We'll see you next week. Thanks for watching.