AI Competition explained in 10 minutes

0:00 In order to talk about how the US and China are competing for AI, let's frame it in these layers. To create AI applications, you need large language models. And to innovate large language models, you need massive server infrastructure. And that infrastructure depends on a supply of advanced chips. And supply requires semiconductor production. And production is only possible with advanced design. And practically at every layer of this, you have varying levels of government intervention between the US and China in order to compete for AI. So let's start with this question. If the competition between us and China is this intense, why don't I really see it? Let's start at the application layer. Most of us are in the application layer where we create and use applications like Chacha PT, Cloud Code, Notebook LM, and Manis. And these applications leverage large language models. And contrary to popular

0:46 opinion, competition at the application layer is actually higher than any of the layers above. And that's because of a concept called barrier to entry. In other words, anybody can create AI applications for practically free. But not everyone can compete with LLM providers like OpenAI, Enthropic Meta, and Deepseek. For example, Meta's Lama 3.1 model, which is an LLM that's 405 billion parameters in size, was released in July 2024. In order to train this massive model, it required 38 septilian math operation, which in the AI industry, this math operation is called flops, which stands for floatingoint operations. So, in order to do 38 septilian operations to train the Llama 3.1 model, you're going to need a lot of computing. Precisely, you need about 4,486

1:31 years just to train Llama 3.1 with one H100 GPU, which is considered the highest grade GPU for training LLMs. And for reference, the H100 GPU costs around 25 to $40,000 a piece. So, how do we shrink down the 4,486 to something more tangible? Thanks to transformer architecture, we're able to run most pre-training in parallel, basically by throwing more money at it. Meta used 16,000 H100 units just to train this model, which cost them around $400 to $640 million in equipment just to train Llama 3.1 model. So, a simple math should tell us that Meta's investment of 400 to $640 million to scale up to 16,000H100 units to run 38 Septilian flops in parallel should take them around 3 months rather than 4,486

2:17 years. So, as you can see, the barrier to entry is quite high, meaning most of us probably won't be starting a company to compete against companies like Meta, OpenAI, and Anthropic. But China has a lot of companies that can compete at this level like Alibaba, Deepseek, BU, Moonshot, and Bite Dance. And as we saw with the Llama 3.1 model, training requires a lot of computing. In other words, for Chinese companies to compete at the LLM layers, they need access to as many GPUs as possible. For example, if Meta only had half of what they actually had, so 8,000 instead of 16,000 H100, the training time for the Llama 3.1 would have been around 6 months as opposed to three. And in the AI industry, a lot can happen during that 3 months that they would have taken them to train the model longer. And this is where we start to go from the LLM layer

3:03 to the infrastructure layer. Because in order to innovate LLM, you need the infrastructure to support the innovation of LLM. So if you put yourself in the shoes of Chinese companies where you have to compete with companies like Meta to innovate the next generation LLM, it's in your best interest to put your hands on as many GPUs as possible. And that's exactly what we're seeing from these companies. Meta's infrastructure is estimated to be about 350,000 H100 GPUs. OpenAI is planning to build Stargate, which is supposed to run up to 2 million GPUs when it's completed. And XAI released tens of billions of dollars on their Colossus facility in Memphis that can run up to 350,000 GPUs. China is a bit of a different story. Unlike the US counterparts like Meta and OpenAI, China doesn't have the same level of infrastructure with advanced chips like H100. And this is largely due

3:51 to heavy restrictions that the US has put on China. And in order for Chinese competitors like Alibaba, Deepseek, Moonshot, Bite Dance to compete against US counterparts like OpenAI, Anthropic, Google, and XAI, you need GPUs in the magnitude of hundreds of thousands if not millions of them. In October 2022, the US banned Nvidia's export of H100 and A100. In October 2023, US banned the H100 chips. And in January 2025, they further restricted H20 until recently the US government agreed to share Nvidia's 15% of their revenue to Chinese sales. So as you can see in the infrastructure layer, China is faced with this barrier to entry, which means their infrastructure doesn't as easily support innovation in the LLM layer as the US does. Which brings us to the next

4:36 layer, which is the supply layer. Although China has many companies that are willing to compete with the US in the LLM layer, their supply of chips is being throttled by the US leading to a weaker infrastructure. So the natural question is this. If China doesn't have access to these chips, how are they able to compete with the US? Deepseek, for example, used the inferior GPU, the H800 card that were bought previously to the band to train the Deepseek V3 model. And even with the lower grade GPU with lesser capacity, Deepseek released a V3 model on December 26, 2024. And this has completely shocked the world because China essentially proved that they can innovate with less. So even with the extreme restrictions that the US has put on China, China was still able to produce Deepseek V3 and Deepseek R1 that went head-to-head against OpenAI 01,

5:23 which was a huge breakthrough in the US. And to put things into perspective, the Llama 3.1 model that we covered earlier took nearly 3 months with 16,000 H100 GPUs that required 38 septillion operations. And compared to this, DeepSync not only used H800 GPUs, which trails behind the H100, but they only used 248 of them, and the training only required 3.8 septilian operations, which is 10 times less than the Llama 3.1. The implication of this is quite large on the infrastructure layer because if you can train an LLM more efficiently, the cost burden in the infrastructure layer is significantly less demanding. So from earlier we talked about how OpenAI is investing in Stargate that has capacity up to 2 million GPUs and XAI's huge investment in Memphis. If Deepc

6:08 continues to innovate state-of-the-art LLMs with less flops, it could totally undermine the US investments on the infrastructure layer that we saw on OpenAI and XAI. So the conundrum here is quite crazy since China is forced to innovate with less and they did innovate with less and the implication of that could be that maybe less is the norm which would completely weaken US's perceived strength because they control the supply layer. Regardless of the case the layer above supply is manufacturing meaning either way these advanced chips need to be manufactured to continue the innovation of LLM and competition in the manufacturing layer between the US and China is huge as well. Back in 2025, China put forward a policy called Made in China 2025, which was a movement to delever their dependencies on foreign countries, especially in the

6:53 semiconductor layer. Their goal was to have 70% of their core materials from Chinese domestic content. Essentially, China wants to produce their own chips instead of buying from companies like Nvidia. But the problem is this. Not only are fabs, which are factories that make chips, extremely expensive, it's also a logistical nightmare and requires extreme technical precision. In fact, there's really only one place in the world that produce most of these chips, and that's in Taiwan by a company called TSMC, that stands for Taiwan Semiconductor Manufacturing Company. So, China's effort to delever from companies like TSMC and Nvidia, would mean that they would at least have to produce chips internally at the same capacity as TSMC at the minimum. But most countries have given up trying due to yet another barrier to entry. But in this case, the

7:39 barrier to entry isn't just about money and capital, but it's also two different things: talent and equipment. While China has more STEM graduates than the US, in fact, nearly double the amount, which helps in the chip design, what they lack is equipment. More specifically, a $250 million equipment called EUV lithography. While most electronics that we use like microwaves, door sensors, and dishwashers can be made in fabs with lower grade equipment like the DUV, high performance GPUs like the H100 requires EUV to generate. The problem is EUV is also near impossible in replicating due to how accurate the equipment needs to be. We're talking precision down to a molecular level. And there's really only one place in the world that manufactures EUV and it's a Dutch company called ASML. And just like in the supply layer, US was banning

8:25 Nvidia from selling advanced chips to China. ASML is also banned from selling EUV equipment to China as well. In other words, in order for China to compete with the US in chip production, the barrier to entry here is not only the cost of fabs, which is around $20 billion for a factory. They also lack the equipment such as the EUV iththography to make the chips internally. So, as you can see in every layer, we see two sides of the same coin. one being the barrier to entry and the other being competition. Meaning in order for China to compete, they need to overcome every barrier to entry that the US is imposing on them. And the US has been making this barrier gradually difficult for China. While China in order to still compete but with less has been proving and almost flaunting at this point by releasing models like Deep Seek 3, Kim K2, Quan 3 coder as open

9:13 models as if to show China isn't too worried about the bans and restrictions that the US has been putting on every layer of the spectrum here. So the ultimate question is this, what does this all boil down to? At the end of the day, everything is contingent on the very last layer, the application layer, which is where most of us live and breathe. AI is only going to be as good as what the application come out of it. In other words, both the US and China are dumping billions of money in supporting companies like Intel, SMIC, Nvidia essentially at the top layer down because it's all about betting on what AI can do in the application layer like healthcare, military, and finance. So although we don't really experience the crazy competition that happens in the upper layers, it's really about us in justifying the competition at the top by innovating cool stuff like chacht manis

10:00 notebook LM replete lovable vzero bolt cursor client or any other domain specific solutions in military healthcare transportation science or anywhere else where AI could fit in. So ultimately, it's up to us in leveraging all the breakthroughs that we're seeing in each layers above to unlock more automation, more disruption, and more game changers that you create in your garage that could someday transform how we do Things.