AI Craze Boosting Demand for Chips, But Geopolitical Risks Linger

The market for semiconductors is expected to grow in the coming years, but US-China friction over AI remains a headline risk for chip producers.

While the demand for AI-interfacing chips is expected to continue into 2024, geopolitical risks emanating from the US-China rivalry could dampen the outlook. But domestic demand, rapid innovation, and a robust startup ecosystem will likely offset the impact of US-issued export restrictions.

Washington has implemented a number of policies to curb sales of AI-interfacing chips (A100s) and equipment (ASML’s EUV machines) to China as tension grow. Officials have expressed concern Beijing could be using the chips to advance breakthroughs in artificial intelligence with military applications.

China’s military strategy of Systems Confrontation and asymmetric warfare - or so-called “Assassin’s Mace” to bring in the Warring States era - strongly rely on data. Specifically, the ability to compute a lot of it and attack multiple systems in the most optimized method.

You can read more about that here.

Generative AI Boosting Demand for Chips

It’s no secret that Nvidia has emerged as the clear leader in artificial intelligence (AI) chips. The type they manufacture are the best for running deep learning models that generative AI like ChatGPT requires.

Nvidia’s biggest seller were their GPUs, and a report by UBS reported that “We expect the GPU and chip segment to be the best near-term beneficiary of strong AI spending”.

This complements forecasts that the revenue for the AI industry is expected to be $420 billion by 2027, up from $300 billion previously, a whopping 40% change. 

Approximately 90% of the demand for these AI-interfacing chips last year was related to processors user for training models. This is as opposed to what is called inference, the process of generating answers or results from those AI models. The UBS analysts expect inference to grow to represent 20% of AI chip demand by 2025.

Another bullish tailwind for companies like Nvidia comes from the fact that many of these special advanced chips have been experiencing a manufacturing shortage and that resulted in a delay of companies rolling out generative AI models. A report from Deloitte found that they expect the “imbalance to continue well into 2024”. 

The same report also predicts that total AI chip sales in 2024 will be 11% of the predicted global chip market of US$576 billion.

According to Barrons, “For the GPU and chip segment specifically, the UBS analysts estimate a 60% compound annual growth rate between 2022 and 2027, driving annual revenue from $15.8 billion to $165 billion over the same period”.

But no picture is perfect.

Geopolitical Risks Clouding Outlook

The most recent round of restrictions came in October of 2023, and are building on the previous export controls from October 2022. They include expanding licensing requirements for exporting advanced chips to over 40 additional countries that are considered risks for diversion to China.

It also includes closing existing loopholes which allow Chinese companies to access some advanced chips necessary for building out their AI program.

The recent measures also include restrictions on exporting certain semiconductor manufacturing equipment to China. This includes deep ultraviolet (DUV) lithography systems. The DUV systems, though less advanced than extreme ultraviolet (EUV) equipment, are capable of producing nearly as advanced chips.

As for ASML, which boasts a near-monopoly on mass-producing EUV (ultra-violet lithography) machines, China was the company’s biggest market in the third quarter of 2023, and responsible for 46% of the company's sales. Its machines can cost between 100-350 million euros per unit.

Following an order by the Dutch government, ASML will curb shipments of two of its machines to China. Specifically, the government has partially revoked the export License of the NXT:2050i and NXT:2100i lithography systems.

Nvidia’s strategic position in the market for generative AI means it was also naturally the target for US export restrictions. China accounts for about 20% of Nvidia’s revenue, but the company is supposedly working on a new lineup of AI chips customized for China. Specifically, the H20 and L20.

US-based semiconductor firms generate 30-40 percent of their sales in the Chinese market, compared to just 25% domestically. With that market share in mind, it is no wonder that geopolitical risks are casting a shadow over key players in the semiconductor industry.

And why companies like Nvidia are walking the commercial-political tightrope to avoid losing a key market. In the short term, the strategic restrictions are not expected to have an impact on the firm’s finances.

However, over the long run there is a risk of declining sales to China if tension between Beijing and Washington continues to swell. The geopolitical context suggests a meaningful detente is far beyond the horizon.