The Western world’s increasing adoption of ChatGPT and other artificial intelligence tools is sparking a surge in investments in generative artificial intelligence. These investments are in turn driving robust demand for high-performance graphics processing units and other semiconductor devices.
Technology research firm Gartner has predicted that semiconductors designed to execute AI workloads will present revenue opportunities amounting to US$53.4 billion this year for the global semiconductor sector.
Globally, AI semiconductor revenue is expected to experience consistent double-digit growth year-on-year, with a forecasted value of US$67.1 billion in 2024. By 2027, this figure is projected to soar to an estimated US$119.4 billion, more than a twofold increase over 2023 revenues.
Taiwan is positioned to reap significant and immediate benefits from this emerging trend. “AI demand will play a pivotal role in driving Taiwan’s economic growth,” notes Ma Tieying, an economist with the Development Bank of Singapore (DBS). She emphasizes that Taiwanese foundries hold a distinct advantage, as they currently command approximately 90% of the global production market for advanced logic chips with dimensions under 10 nanometers.
However, supply issues are expected in the coming year. Taiwan Semiconductor Manufacturing Co. (TSMC) in September said supply constraints on AI chips are likely to last for around 18 months due to limitations in advanced chip packaging services. Chairman Mark Liu said the problem lies in the shortage of advanced chip packaging (CoWoS) capacity. TSMC is the exclusive manufacturer of Nvidia’s H100 and A100 AI processors, which are vital for applications like ChatGPT.
Liu says the shortage should be temporary, easing by the end of 2024 as TSMC expands its advanced chip packaging capacity with a new facility in Taiwan. Stressing the need for a “paradigm shift” in semiconductor technology integration, he emphasized that tightly interconnected chips are needed to accommodate more transistors, leading to higher computing power.
TSMC envisions chips having over a trillion transistors within the next decade, made possible by advanced chip packaging techniques. Other chipmakers are also focusing on innovative packaging, with Intel aiming to quadruple its advanced chip packaging capacity by 2025 in its bid to regain global leadership in semiconductor manufacturing.
“TSMC has already secured a substantial share of the AI market,” says Ma, noting that it already manufactures graphic processing units (GPUs) for Nvidia and AMD. “Industry giants including Microsoft, Google, Amazon, and Meta are reportedly pursuing their custom AI chip development, which will also leverage TSMC’s advanced nodes for fabrication.”
Bum Ki Son, a regional economist with Barclays, says that since the second quarter of this year, “we have seen very high-end semiconductor demand picking up, including demand for AI-related GPUs. GPUs are important as they make the processing of data a lot faster. The use of AI means that the data we need to process in our normal day-to-day lives will increase exponentially. TSMC has an edge in these high-end productions. The rise in AI demand is a structural factor that will benefit Taiwan.”
However, Ma points out that a potential risk to the evolution of the AI chip industry is the possibility of escalating U.S.-China trade tensions. She notes that in 2022 the United States imposed a series of export controls to restrict the sale of advanced chips to China, particularly those used in AI applications. Some chip companies, however, managed to circumvent the restrictions by developing customized AI chips specifically for the Chinese market.
This development forced Washington to announce new export restrictions in October that would effectively block Nvidia from selling two of its modified advanced AI chips to China. Nvidia appears to be undaunted. Chinese media reported in early November that Nvidia had created yet another three AI chips custom-made for the mainland.
Nvidia has a more than 90% share of China’s US$7 billion AI chip market, and analysts predict that U.S. restrictions will allow domestic Chinese firms such as Huawei Technologies, which is also a major TSMC customer, to gain a foothold in that market.
Chinese internet giant Baidu reportedly placed a sizeable order for Huawei AI chips this year. According to one source, Baidu placed the order before the U.S. curbs were announced in preparation for a future when it could no longer purchase from Nvidia.
Seeking to address worries surrounding the latest U.S. bans on the export to China of certain advanced AI chips, Taiwan’s Minister of Economic Affairs, Wang Mei-hua, told journalists in late October that they were unlikely to have a big impact on Taiwan.
While acknowledging the need for control over exports to China, Wang pointed out that Taiwan primarily serves as an original equipment manufacturer (OEM), tailoring its shipments to meet customers’ needs. She added that the current high demand for AI chips and servers already surpasses supply, so the ban should have a limited effect on Taiwan’s economy.
In parallel, Ma notes, Chinese electronics firms like Huawei and SMIC have made technological advancements that could also attract attention from American regulators. All of this could potentially lead to increased regulatory scrutiny and tighter export controls by the U.S. government. “This could impact the global semiconductor sector’s recovery,” she says.