Running in the background to the more visible, very real wars occurring worldwide, there is another, economic and industrial in nature. This is a trade war between two principals: the US and China. The potential consequences of this trade war are incredibly significant to global geopolitics, as how the interaction between the two parties is handled will determine a basis for any future cooperation between the two principals and their respective allies.
This trade war concerns the production of semiconductor chips, those small pieces of (mainly) electronic circuitry found in many modern-day devices, from smartphones to televisions, cars to computers. In IDTechEx’s recent report “AI Chips 2023-2033”, the market research company highlights the part that artificial intelligence (AI) plays in this trade war, where the race for AI supremacy has become a national concern.
The shortage, where demand exceeds supply
But before discussing the role of AI, it is more instructive to start at the beginning or thereabouts. This does not require going so far back in time – only five years, in fact, to 2018. From January of that year, under the presidency of Donald Trump, formerly and now President Joe Biden, the US has enforced several layers of restrictions and barriers to trade with China where semiconductors are concerned, each successive layer aiming to plug a hole that the previous layer(s) had left.
The reasons for these restrictions are multiple, starting with simple economics (where the US wishes to halt China’s growing market share in the semiconductor supply chain) before moving onto interwoven concerns such as risk exposure and using semiconductors for militaristic purposes.
Regarding geographic risk exposure, since 2020, there has been a global chip shortage, where demand for semiconductor chips has exceeded supply. This shortage exposed US design companies to the risk of relying upon South-East Asia manufacturing capabilities, as lead times elongated to upwards of 3 months by the beginning of the crisis.
Several complementing factors caused the global chip shortage. Principle among them is the Covid pandemic, which saw a rise in demand (due to more people working from home and so in need of personal computers) and a fall in supply (due to lockdowns across Asia resulting in plants being shut down).
Other factors include the rise of data mining (where GPUs are needed, thus once again increasing demand); a drought that hit Taiwan in 2021, resulting in problems with producing ultra-pure water to clean factories and wafers; fires at several fabrication facilities owned by Asahi Kaseri, Renesas and ASML; and difficulties procuring neon (used for lasers in chip manufacture) due to the Russia-Ukraine War, as Ukraine was responsible for providing more than 90% of the US semiconductor-grade neon.
These events combined would likely be enough for a country such as the US to consider investing in its own production capabilities (where the US Chips and Science Act of 2022 is discussed in more detail in the aforementioned IDTechEx report).
But China’s spending on semiconductor imports and the blurred line between commercial and military ventures in the country has made the US wary of outfitting a rival economy with the tools to surpass them in key technology areas. This is where AI comes into play.
AI as a motivator
In addition to the abovementioned restrictions, on 26th August 2022, the US government banned AMD and Nvidia from exporting chips that can be used to support AI workloads to China. According to an August SEC Filing made by Nvidia, this comes in the form of a license agreement, effective immediately, for any future export to China (including Hong Kong) and Russia of Nvidia’s A100 and forthcoming H100 integrated circuits.
Any systems that incorporate the A100 and H100 ICs are also covered by the new license requirement, and any future integrated circuits that are roughly as advanced as the A100. The filing asserts that the US government has indicated that the new license requirement will address the risk that China or Russia may use the covered products in a military capacity. Similarly, a spokesperson for AMD – speaking with Reuters – said that the company had received new license requirements that effectively stopped AMD’s MI250 AI chips exports to China.
This information alludes to the fact that the restrictions imposed upon Chinese companies by the US government are not simply a matter of trying to take some control of the supply chain from the APAC region but also a matter of national security.
The measures are an attempt by the US to arrest China’s AI surge by preventing the type of advanced technologies needed for China to realise AI supremacy (a not unfounded concern, as by 2018, China had filed 2.5X more patents in AI technologies than the US). Such is the interwoven aspect of commercial and military endeavours in China.
The Biden administration has effectively stopped trying to block military-affiliated exports while retaining commercial exports and the revenue generated. As such, to quote the Center for Strategic & International Studies, “High-end AI chips can no longer be sold to any entity operating in China, whether that is the Chinese military, a Chinese tech company, or even a US company operating a data centre in China.”
AI not only promises to be one of the biggest drivers of economic advances within the next quarter century, but for China, AI mastery represents the ability to perfect a governance model in keeping with existing architectures. The effectiveness of AI models comes largely down to the quality and breadth of the training data set provided. Given that China can collect significant volumes of citizen data, the country is poised to reap the benefits of widespread AI usage.
China has – until recently – been rather quiet in the face of these restrictions, although some displeasure was expressed when TSMC (a Taiwanese semiconductor fabricator that accounts for the vast majority of leading-edge node manufacture globally) announced plans to build new fabrication facilities in Arizona, US, last year. But in July 2023, China struck back, with restrictions placed on the export of Gallium and Germanium materials used in certain semiconductor chip manufacture.
Germanium is used in applications such as thermal imaging cameras, solar panels, and telecommunications, where germanium can be used in photodiodes to convert light signals to electrical ones. Gallium is often paired with arsenic to form gallium arsenide, a compound semiconductor that can operate at higher temperatures and frequencies than silicon.
According to the US Geological Survey, China currently produces around 98% of the world’s gallium and controls around 68% of global refined germanium production in various countries. The impact of these material export restrictions on the US and allied countries is not to be downplayed.
Both European and Asian delegates have warned against these ongoing restrictions by both principals, given that the onus on the shoring up of national interests currently involves concurrently punishing the other party. And the harder that the US pushes against China, the harder that China will lean into plugging money into their own domestic supply chain (and given that in 2021 China spent the equivalent of USD$432 billion on imported microprocessors, money that they may no longer be able to be spent on imports, China is certainly not short of funding).
As with most countries, China has not looked to develop an isolated, front-to-back domestic semiconductor supply chain to date, as they have had the option to work with superior foreign partners rather than domestic firms that do not meet the same standards; design companies could work with established fabs in Taiwan rather than inferior domestic fabs, and Chinese fabs could buy foreign semiconductor manufacturing equipment (SME) with proven quality.
Now that they are cut off from these possibilities, China must look to growing domestic capabilities.
In the short term, this will be very difficult for China. While the country has been stockpiling chips and SMEs in anticipation of these imposed controls (the 7 nm chip produced by SMIC was produced using existing deep ultraviolet (DUV) machines), these resources will eventually run out.
So China must look to develop domestic capabilities if their semiconductor industry isn’t to completely dry out, at least in terms of the more advanced node processes. It seems likely that China will have to rely on more mature node processes for the next few years to create new chips.
It is unlikely that China will have no foreign support, at least in terms of SME and component supply. Companies with a large stake in the Chinese market – such as Zeiss, a German company that supplies mirrors to ASML for their extreme ultraviolet (EUV) lithographic machines, and where China is their fastest growing market – may be unwilling to relinquish revenues generated from China.
To abide by the US export controls, companies may engineer out the US inputs or components in their products such that they can sell these without repercussions in China.
In the longer term, there is cause for cautious optimism for China. Forced to work together in a way that they have not done previously, China’s fabs, design companies, and SME firms may form an ecosystem that is not only stronger from the forced collaboration but also more thoroughly isolated from global supply chain disruptions than most other countries (and, in addition, free from US controls).
This comes with significant hurdles, but these may be removed or – at the least – lessened should other countries feel the adverse economic effects of being unable to bolster their raw material supply.
The story is far from being finished, but with a projected growth of US$257.6 billion by 2033 for AI chips alone, much will be gained and lost over the next 10 years.
IDTechEx forecasts that the global AI chips market will grow to US$257.6 billion by 2033. The report covers the global AI Chips market across eight industry verticals, with 10-year granular forecasts in seven categories (such as by geography, chip architecture, and application). In addition to the revenue forecasts for AI chips, costs at each stage of the supply chain (design, manufacture, assembly, test & packaging, and operation) are quantified for a leading-edge AI chip.
Rigorous calculations and a customizable template for customer use are provided, and analyses of comparative costs between leading and trailing edge node chips.
IDTechEx’s latest report, “AI Chips 2023-2033”, answers the major questions, challenges and opportunities faced by the AI chip value chain.