Nvidia and AMD: Weapons of AI Chip War

The 20^th-century geopolitics were centered around oil; the 21^st century’s geopolitical dominance is determined by technological supremacy. Like in the previous century, the U.S. is defending its turf against another superpower, this time, not the Soviet Union but China; “the new oil” is the chips necessary for the development of AI technology.

As the technological rivalry between the two empires continues to intensify, the main focus is shifting more and more to Artificial Intelligence (AI). It’s much more than ChatGPT’s ability to write great texts and Midjourney’s capability to produce superb images, although these skills can produce a lot of political, economic, and social damage if they fall into the wrong hands (think about disinformation, deep fakes, espionage, etc.).

Scientists and tech business leaders agree that Artificial Intelligence will bring a major technological shift comparable to the invention of the Internet; the technology has the potential to upsurge economic productivity and competitiveness, as well as impact, or even predetermine military dominance. Earlier this year, the Pentagon, backed by tech leaders, sounded the alarm on China’s development of AI-powered warfare practices, aimed at amassing advanced capabilities to counter the U.S. in a potential conflict.

Cold War 2.0

The “AI arms race” officially started back in 2017, when China announced its plan to become the global leader by 2030. It was crystal clear back then, as it is today, that modern global leadership ambitions aren’t carried out by mass-producing cheap shoes for Western shoppers, but by winning on the technological advancement front; AI technology is at the forefront of those efforts. Since no AI advancement is possible without the most innovative chip technology, the U.S. efforts at countering China’s AI progress are concentrated around chips. China is investing heavily in the development of the domestic semiconductor industry, but it is still some years behind the U.S. in the AI-chip sphere. Besides, the progress of Chinese chipmaking ventures depends on their access to Western advanced chipmaking equipment.   

The Biden administration has continued and intensified the previous U.S. government’s policies on China. Last year, the administration imposed export controls, intended to prevent the sale of the most advanced chips, which are necessary to develop and train AI models and the chipmaking equipment needed to produce them, to entities operating in China. Back then, the sweeping restrictions were thought to deal a heavy blow to companies in the industry, especially to providers of chips and chipmaking equipment with high exposure to China, such as Applied Materials (AMAT), Lam Research (LRCX), KLA Corp. (KLAC), and others. However, although earnings have suffered, the industry adapted; this year’s rally in anything AI-related has swiped left on all of last year’s worries.

Source: Google Finance

Now, it’s time for the next stage in the U.S.-China AI divorce saga. According to the reports, the U.S. government is planning additional restrictions on AI chips and equipment in China to further ensure that the Asian giant stays behind the U.S. in the AI arms race.

The AI Arms Race Heats Up

After last year’s restrictions were put in place, the U.S. quickly found out that Chinese firms found ways to evade them. While smuggling or buying via third parties and shell companies are hard to clamp down on, the U.S. government is taking steps to close the main loopholes through which China has been gaining access to AI chips, post-ban.

The chipmakers are partly to blame themselves for the next wave of restrictions. In their attempt to defend their Chinese market share, they made all efforts possible to bypass the administration’s demands.

The main culprit of this is Nvidia (NVDA). After the previous restrictions were announced, the company designed a slightly lower-performing version of its AI GPU (graphics processing unit) A100, which fell within the limits set by the government for the chips shipped to China. The “Chinese chip,” called A800, operates at 70% of the speed of A100 GPUs that power AI computations in data centers in the U.S., but is still superior to anything produced in China; besides, it can be used in automated targeting systems for drones or in the surveillance systems. Under the new export curbs, A800 chips will need a government license for exports to China, i.e., effectively banned.    

Putting Ever Higher Fences Around AI

In addition to using previously legal “curtailed” Nvidia chips, Chinese companies have used other loopholes to gain access to AI-enabling technology. While the restrictions focused on physical exports of the chips, Chinese tech firms have been renting access to controlled chips via the cloud. Top cloud providers Amazon (AMZN), Microsoft (MSFT), and Alphabet (GOOGL) have been long offering remote access to the computing power of AI chips through leases; China has been making an accelerating use of these offers since the restrictions were introduced last year. In March this year, Nvidia said that it also plans to rent out supercomputing power to Chinese firms, willing to leverage their demand for its high-end technology.

Apparently, the Biden administration does not view this “in your face” approach lightly. According to the reports, the government is planning to restrict the leasing of cloud services to Chinese companies. If that’s not enough, the government, reportedly, is crafting a ban on the U.S. investment in advanced semiconductors, AI, and quantum computing firms in China, intending to prevent the U.S. money and expertise from supporting the development of advanced technologies “made in China.”

Nvidia and AMD at Crosshairs

The new restrictions will certainly hurt the revenues of  U.S. chipmakers such as Nvidia and Advanced Micro Devices (AMD), as well as companies trying to challenge Nvidia’s and AMD’s dominance in artificial intelligence chips, like Intel. They will also apply to firms that sell products containing advanced chips, such as servers for AI data centers. Dell Technologies (DELL), Hewlett Packard (HPE), and Super Micro Computer (SMCI) fall into the category. The cloud providers will also be impacted by the ban on their “rent-an-AI” service, albeit just marginally.

Nvidia controls about 80% of the market for the accelerator chips in the AI data centers run by Amazon, Alphabet, and Microsoft, and derives about 22% of total revenues from China, with the revenues from data center chips sold to the Asian superpower ranging between 7% and 10% of the whole revenue pie.

Source: Google Finance

AMD, the second-largest maker of GPUs, is aspiring for a larger market share; Intel (INTC), who’s been late in the AI game, has just started making advancements in the AI chip sphere. While Intel, an AI newcomer, has nothing yet to lose from the China AI chip ban, AMD will be hurt. Although it has significantly reduced its dependence on Chinese sales in the past decade, they still make up about a quarter of its total revenue (however, much of the income is derived from AI-unrelated products).

Of course, the new restrictions would present a serious obstacle to both Nvidia and AMD in terms of revenue generation in the short term, until they find a way to adapt. Citigroup (C) said that AI demand will exceed supply in the near term; the two chipmakers remain at the forefront of a surge in AI development. The Bank of America (BAC) forecasts that the Chinese market will ultimately represent less than 10% of the total addressable market for AI-related accelerator chips, expected to reach more than $100 billion; Evercore ISI called the additional restrictions “a little speed bump on the AI highway.”

The Global AI Chip Blockade

The U.S. is not alone in this battle, as all its allies have by now come under the anti-China banner – again, reminiscent of the Cold War era. The domiciles of global producers of advanced chipmaking equipment, Japan and the Netherlands, have also restricted exports of gear to China. The Dutch tech mega-cap ASML Holding NV (ASML) is facing a ban on exports of its chipmaking machines to China; the restrictions are expected to provide a blueprint for other European governments to follow suit.

ASML is the world’s single maker of high-end semiconductor lithography equipment, therefore a ban on its equipment exports is one of the most effective chokepoints for Chinese AI chip advancement. The company is a vast ecosystem of knowledge and effectiveness; there is no chance whatsoever for China to be able to replicate its products within a decade at the least. ASML is so overwhelmed by the global demand for its equipment that it doesn’t view the restrictions on exports to China as any kind of danger to its bottom lines.

Japan also joined the U.S. ranks earlier this year with restrictions on its domestic firms’ exports to China, related to 23 types of cutting-edge chipmaking technologies and equipment, which now need government licenses. The restrictions apply to companies such as Tokyo Electron (TOELF), Screen Holdings Co. (DINRF), Nikon Corp. (NINOF), and Lasertec Corp. (LSRCF), which are vital parts of the global AI chip supply chain. These companies derive 20% to 25% of their revenues from China sales, which means they will be seriously impacted by the ban, at least in the short term.

However, on a larger scale, this pain is worth the long-term gain. With Japan and Netherlands joining the restrictions club, Chinese efforts at making swift progress in AI will be severely curtailed, assuring that this round of the match is won by the U.S. and its Western allies.