China vs US: who’s winning the race for AI supremacy

The foremost global superpowers, the US and China, are competing for AI dominance, trying to complete the same puzzle using different pieces. And the US lead is not as clear-cut as many perceive.

“AI is expected to become a crucial component of economic and military power in the near future,” Stanford University’s Artificial Intelligence Index Report 2023 stated.

Shifting AI development will concentrate in certain places, reshaping the geopolitical landscape.

To limit China’s advances, the US put export controls on high-end chips, limiting its ability to obtain advanced computing chips, develop and maintain supercomputers, and manufacture advanced semiconductors for military and other purposes. China hit back with restrictions of its own on rare minerals.

How do China and the US compare in this race?

“After more than two decades of outsourcing, North America now finds itself in a worrisome predicament: it can design the most cutting-edge electronics but cannot manufacture them,” the study by the IPC, the global association for electronic manufacturing, stated.

Neither can China, which is the largest semiconductor importer.

A recipe for AI supremacy would likely include critical ingredients such as strong computational infrastructure, research community, and dedication.

The US is ahead in chip design

Semiconductors, providing computational power, are essential in training AI models and powering other systems.

Multiple stages are needed to build a final chip, and no single country is self-reliant on all of them. Offshoring of design, manufacturing, packaging, and final assembly spans the entire electronics ecosystem.

The US clearly dominates the semiconductor design market, having an 85% global share, with 5% left to Asian countries, according to the Department of Defence.

China, the largest importer and semiconductor market, depends on US chips. China represents 31.4% of worldwide final sales or $180 billion out of $574 billion in 2022, Citi Global Insights report revealed.

The recent US bans on exports of high-end components are a step back for China. However, this doesn’t mean that China is left without computational resources and can’t design its own chips.

Beijing, partly reacting to US tech bans, has implemented policies to catch up with global competitors.

Baidu, Tencent, and other Chinese tech companies announced that they have enough AI chips stockpiled to soften the impact of the US export ban while seeking Chinese replacements. Baidu has long been developing its own Kunlun chips for AI calculations.

China even boasted of developing a photonic chip that is “3,000 times faster than one of the most widely used commercial AI chips, Nvidia’s A100.”

In August, many experts were taken by surprise when Huawei released the Mate 60 Pro phone, running on a Chinese-made Kirin 9000s 7-nanometer processor. While not cutting edge, the 7nm process is still highly advanced, similar to Intel’s in the US.

Both China and the US dependent on fabs in Asia

The design itself is not a chip. The US chip companies, such as Nvidia or AMD, have long migrated to the fabless model, depending on third-party producers.

Only 12% of global chip manufacturing resides in the US. Asian companies lead across the entire semiconductor supply chain. Fabrication is dominated by the Taiwanese TSMC and South Korea's Samsung, offering the most advanced nodes. Foundries in other countries lag behind, such as China’s largest chipmaker, SMIC (Semiconductor Manufacturing International Corporation), and Intel in the US.

TSMC’s resources can be accessed by both China and the US.

13.1% of all TSMC’s revenues came from China during the three quarters of 2023, up from 10.5% during the same period last year. The US’s orders brought 63.1% of TSMC revenues during the last three quarters, down from 65.5% last year.

It’s harder for China to develop new capabilities, as SMIC is restricted from accessing new technologies from ASML, the dominant producer of advanced lithography systems for foundries.

However, SMIC has “amassed massed a large number of machines and potentially also a large stockpile of spare parts that it can draw from,” a report the Center for Strategic and International Studies (CSIS) revealed.

China, which set a target to reach 70% self-sufficiency in chips by 2025, appears to be working on manufacturing its own lithography machines, even if they will be dwarfed by ASML’s.

“If breakthroughs are to be made through its substitution strategy, Chinese players have a good track record in driving the costs down and could be quite competitive, at least in the home market,” the Citi Research report said.

China’s imports of integrated circuits are declining sharply from a peak of 637.9 billion units in November 2022 to 484.2 billion units in August 2023 (a trailing 12-month sum). Domestic production remained stable and above 300 million units.

The US semiconductor sector also depends on high profits generated from chip sales, which help maintain its market lead. The loss of the Chinese market could pose challenges to their business model.

“The US strategy could, in fact, strengthen Beijing’s determination to achieve tech self-reliance and could backfire on its own (US) IC companies,” the CITI report said.

China owns boards and soldering

Silicon wafer is still not a final product. Asia dominates substrate manufacturing, advanced packaging, PCB fabrication, and final hardware assembly capability.

Advanced IC-substrated fabrication is carried out by companies such as Unimicron, Ibiden, SEMCO, Nan Ya, and Shinko. Here, the US’s global share is 12%.

On November 20th, the Biden-Harris administration announced $3 billion in funding for the National Advanced Packing program “to drive US leadership in advanced packaging,” part of the CHIPS and Science Act.

Advanced packaging, part of semiconductor manufacturing, refers to the tight assembly of many chips with diverse functions on a substrate at extremely fine dimensions.

Chinese media reacted with a comparison: the US’s chip packaging production capacity accounts for only 3% of the world’s total, while China's packaging capacity is estimated at 38%. It takes years to develop new capabilities.

When everything needs to be stitched together in the advanced packaging step, 97% of this work is done in Asia. Taiwan is again the leader with nine companies, including ASE+SPIL and Powertech Technology.

“North America lags behind Asia in capacity, but technical capability is a source of concern as well,” IPC study reads.

China's dominance in commodity board, multilayered board, and high-density interconnect board markets is over 50% – it has significant capacities for IC packaging flex circuits.

China has the minerals

There are more ways China could retaliate, hindering US companies from producing advanced electronics.

China has already introduced export restrictions on gallium and germanium. However, there are dozens of other critical minerals that the US does not produce domestically, and China dominates the production of.

In 2022, the US was 100% net import reliant for 15 nonfuel mineral commodities. The list could be extended to 51 minerals with reliance larger than 50%. China is the leading source of 26 critical commodities, Cybernews already reported.

AI research: China has the lead

Semiconductors, as difficult to produce as they can be, are useless on their own. They are supposed to run the software.

And China’s strides in AI research are overwhelming. China leads in total AI journals, conferences, and repository publications, as revealed by Stanford University’s Artificial Intelligence Index Report 2023.

Chinese researchers published 40% of all global AI publications in 2021, exceeding the United Kingdom, Europe (15%), and the United States (10%) combined.

In 2021, China produced the greatest share of the world’s AI conference publications at 26.2%.

While the US is still ahead in AI conferences and repository citations, those leads are slowly eroding.

Nine of the top ten largest AI publishing institutions are from China, with the Massachusetts Institute of Technology taking tenth place.

US big tech produces more actual AI products

Most significant AI-based systems are now produced by industries rather than academia.

American companies and institutions developed most of the world’s large language and multimodal models, 54% in 2022.

In 2022, China developed only three “significant machine learning systems,” while the US produced 16.

The notable American models included OpenAI’s DALL-E 2 and GPT-3, or Google’s PaLM. The only Chinese large language and multimodal model released in 2022 was bilingual GLM-130B. In 2023, we’re hearing about new AI models every month.

The story is similar with private investments. US’s private investors brought $47.4 billion to the table in 2022, roughly 3.5 times more than China’s. The US leads globally in the total number of newly funded AI companies, having 3.4 times more than China.

The only areas where China's investments in AI were more extensive than the US’s were semiconductors, industrial automation and networks, and sales enablement.

Six American corporate giants dominate the AI market. However, Chinese companies such as Tencent, Alibaba, and Huawei are on their heels.

“Chinese companies are gaining traction, however. Only one company was in the top 10 in 2012; there were four in 2021. Tencent, Alibaba, and Huawei have forged ahead in the number of AI papers that they produce and the citations that these publications receive,” Stanford University’s Report noted.

Baidu even claimed its Ernie AI chatbot is better than OpenAI's ChatGPT, even if it was unable to answer what happened on June 4th, 1989, or evaluate China’s “zero Covid” policy.

The gap is also closing in AI software development. Chinese developers contributed 11.7% of high-impact changes to GitHub’s public AI projects in 2022, while their share was only 1.9% in 2011.

US developer contribution to AI projects was 20.3% in 2022, down from 40% in 2011, the statistics by the Organization for Economic Co-operation and Development (OECD) reveal.

Chinese support AI more than Americans

While the political will to develop AI seems strong in both countries, the same could not be said about the general population.

Most of the Chinese population, 78% of respondents, agreed with the statement that products and services using AI have more benefits than drawbacks, according to the 2022 IPSOS survey. That’s the highest proportion of surveyed countries.

Only 35% of sampled Americans agreed, having one of the lowest rates among surveyed countries.

Analyst: US has the edge, but it’s eroding

“The US has certainly better capacities because it already has the technology, the equipment, and the people,” said Irina Tsukerman, a geopolitical analyst, national security lawyer, and President of Scarab Rising advisory.

However, she believes that it’s only a matter of time before China becomes self-sufficient in high-end technologies. The main goal for the US is to stay ahead.

“The US has not been particularly great at restricting and sanctions in general, whether it comes to China or anybody else. I think there's a good chance that sooner or later, China will catch up unless something changes and the US and other countries become much more vigilant about their own technologies,” Tsukerman said.

In the wake of the chip wars, the problem for the US is the need to ramp up its own production to be less reliant on companies in Asia. However, the US lacks trained personnel and skill sets that have moved to India or China. Even work attitude raises concerns, as machinery must run non-stop for 24 hours.

While the US’s restrictions have slowed China’s advance, Tsukerman believes that the country is making breakthroughs faster than expected.

“The ban is not complete. It's a restriction of some but not all the chips and related matters to China,” said Tsukerman. “The idea is to slow down or prevent China from being able to utilize this knowledge to get ahead of the US in building military equipment and certain types of computer equipment that's heavily reliant on these technologies.”

Despite the restrictions, the chipmaking equipment and the advanced chips have been reaching China, sometimes even smuggled via third countries.

“Clearly, China has enough AI chips to be able to produce these vehicles en masse and to export them for sale outside the country and to remain dominant in that area,” Tsukerman said. “Anyway, the US and all its partners are more concerned about the military production.”

She believes that restrictions have caused some slowdown in China’s military technology development. However, it is impossible to evaluate the effect due to low transparency.

“Still, it has not fully caught up to the United States,” Tsukerman said. “We do know that it is desperate to reach superiority because it's working so hard to find a way to import either the chips or the technologies and produce them domestically. So clearly, it is very much an issue.”

While some chips, in theory, could be contained, the US’s competitive edge may deteriorate over time as China works on its own parts.

“The ban on the chips will not affect the fact that there are people who are able to actually create those technologies,” Tsukerman said.

China also has the resources to recruit scientists or even steal technology and knowledge from other countries. It already dominates the raw material markets.

“What it lacks is the advanced knowledge of the current development process. China is quite successful in circumventing intelligence and local company restrictions, infiltrating with their own scientists and stealing that knowledge base,” she said.

Despite all the hurdles, Tsukerman wouldn’t expect China to overtake the US in technology unless “the US is not vigilant about its own industries, which it has not been very well up until now.”

The bureaucratic communist structure makes it difficult for innovation and creativity to blossom.

“China has had cultural and economic issues preventing it from becoming the number one power. Otherwise, it would not have had to rely so much on exports and stealing,” Tsukerman concluded. “It’s good at replicating, building capacity, and luring foreign scientists. But when it comes to the domestic knowledge base, training, and creating entirely innovative industries, they are subpar.”