According to a recent report, there could be a potential global chip shortage as suppliers may find it difficult to keep up with the rapidly increasing demand for AI-related goods and services.
Up until 2027, AI workloads could increase by 25% to 35% yearly, predicts consulting firm Bain and Company. Still, there’s a good chance that a mere 20% increase in demand will upset the balance and cause another global chip shortage.
According to the authors of the Global Technology Report 2024, the proliferation of AI in the convergence of the major end markets could easily cross that threshold and result in weak spots throughout the supply chain.
Building larger data centers with a capacity of more than a gigawatt will also be necessary due to our insatiable thirst for AI. Most current data centers have a capacity of 50–200 megawatts.
The market for AI software and hardware is anticipated to grow between 40% and 55% annually over the next three years, when the demand for AI infrastructure and AI-enabled products is combined.
According to the report, if the current cost of large data centers is between $1 billion and $4 billion, in five years it could reach between $10 billion and $25 billion. This leads to an estimated total AI market value for 2027 of $780 billion to $990 billion (£584 billion to £741 billion).
The supply spider’s web, and the pressure it’s under
The supply chain for AI components needs to be able to expand at the same rate in order to meet this growing demand. However, the raw materials for the chips are actually at the center of the chain, which resembles a complicated spider’s web.
The infrastructure and fabrication lines needed to increase chip production flow in one direction, while data centers are necessary for AI products to operate in another. According to Bain, each has a lead time ranging from three and a half to more than five years, which presents a major obstacle to meeting demand.
The most vulnerable link, according to the report, is bleeding-edge fabs that produce the most sophisticated chips. Between 2023 and 2026, they will have to increase their output by 25% to 35% in order to keep up with the anticipated 31% and 15% growth in PC and smartphone sales, respectively.
To keep up, up to five more state-of-the-art fabrication facilities would have to be built, at a cost of between $40 billion and $75 billion.
There is also the supply chain involved in transforming chips into smartphones and PCs with on-device AI functionality, such as Apple Intelligence devices, which are becoming increasingly popular as consumers demand greater data security.
In fact, to make room for the on-device neural processing engines, the silicon surface area of the typical notebook core processor and smartphone processor has already increased by 5% and16%, respectively. According to Bain, by 2026, these products may raise the need for upstream components by at least 30%.
Another aspect of the web is packaging, and suppliers would have to triple their production capacity if GPU demand doubled by 2026. In addition, different power and cooling needs connect all process components to utility companies, which must also adjust to demand.
The previous worldwide shortage of chips
Chipmakers have prospered since the current generative AI boom began. Leading vendor of graphics processing units NVIDIA reported record earnings for the second quarter of 2024 of $30 billion (£24.7 billion). The company is valued over $3 trillion (£2.2 trillion) on the stock market. Similar success has been experienced by memory chip manufacturer SK Hynix and switch manufacturer Broadcom.
Few core companies, controlling significant portions of the supply chain, have realized these record profits. An American company called NVIDIA creates most of the GPUs used for AI model training. On the other hand, Taiwan’s TSMC makes them. Additionally, only TSMC and Samsung Electronics are able to produce the most advanced chips on a wide scale.
However, things haven’t always been easy in this sector. The COVID-19 pandemic in early 2020 caused a global chip shortage. For more than three years, there were supply problems among these relatively small number of businesses, which had an effect on consumer electronics and artificial intelligence.
The supply chain for semiconductors was already unstable before the pandemic struck because of a number of factors, such as trade disputes between the United States and China and Japan and Korea that affected the distribution and price of commodities. The scarcity of raw materials was also exacerbated by natural calamities, such as the three plant fires in Japan between 2019 and 2021 and the drought in Taiwan.
According to the Bain and Company report, “over the past ten years, extreme weather, natural disasters, geopolitical strife, a pandemic, and other major disruptions have made abundantly clear how supply shocks can severely limit the industry’s ability to meet demand.”
Chip scarcity might worsen if AI sovereignty is desired
A second global chip shortage could result from more than just a lack of manufacturing capacity.
The global semiconductor supply chain is still facing significant risks from geopolitical tensions, trade restrictions, and the decoupling of supply chains from China by multinational tech companies. According to the research, uncontrollable circumstances like shortages of materials, delays in factory construction, and others might also result in pinch points.
For instance, export restrictions pertaining to chips have been implemented by the United States on the sale of semiconductors to China, the Netherlands, and Japan. In 2023, the U.K. obstructed most license applications submitted by businesses looking to export semiconductor technology to China.
Additionally, the Chinese Ministry of Commerce declared that, “to safeguard national security and interests,” export restrictions on goods related to gallium and germanium would be put in place. China produces 98% of the world’s supply of gallium and 54% of the world’s supply of germanium, two rare metals that are crucial to the manufacturing of chips.
To lessen their dependency on other nations, governments throughout the world are also investing billions of dollars to increase their own capacity for semiconductor production. Data security, however, is also important because it allows authorities to better defend against cyberattacks and espionage by maintaining the supply chain inside their borders.
In order to bolster America’s economy, national security, and supply chains, as well as to provide much-needed semiconductor research investments and manufacturing incentives, the United States passed the CHIPS Act in 2022. Along with investing in the proof-of-concept for a shared national AI research infrastructure, the White House has also unveiled a blueprint for an AI Bill of Rights to aid in domestic regulation of AI.
The four biggest memory chip companies in the world, Intel, TSMC, Texas Instruments, and Samsung, have all declared their intentions to construct factories in the United States.
It was announced in August 2023 that the government of the United Kingdom will invest £100 million ($126 million) to support the development of AI hardware and prevent any potential shortages of computer chips. Only this month, Amazon Web Services declared that it would spend £8 billion over the following five years on data centers in the nation.
With the passage of the European Chips Act in July 2023, the European Union provided €43 billion ($46 billion) in subsidies to support the semiconductor industry in that country. By 2030, the bloc hopes to have achieved its ambitious target of producing 20% of the world’s semiconductors.
The leader of Bain’s Global Technology practice, Anne Hoecker, however, predicted that achieving data sovereignty will be “time-consuming and extremely expensive.”
“While less complex in some ways than building semiconductor fabs, these projects require more than just securing local subsidies,” the speaker stated in a press release. Big tech firms like hyperscalers might keep spending money on localized AI operations to maintain a sizable competitive edge.
According to the Bain report, since small language models handle a lot of the networking, computing, and storage tasks close to where AI data is stored, they could benefit from data sovereignty. These algorithms use RAG, or retrieval-augmented generation, and vector embeddings.
Suggestions for AI supply chain executives on how to deal with a shortage of chips
In order to avoid another worldwide chip shortage, the Bain report offers the following advice to businesses that use semiconductors:
- Gain a thorough understanding of and keep tabs on the whole AI supply chain, which includes peripherals like routers and network equipment, as well as PCs, smartphones, and data center components.
- To guarantee access to chips in the event of a disruption, sign long-term purchase agreements.
- To maximize compatibility with various suppliers and sourcing flexibility, design products using industry-standard semiconductors rather than application-specific chips.
- By sourcing components from several locations and diversifying your suppliers, you can fortify your supply chain against geopolitical risks like tariffs and regulations.
While executives may still be fatigued from the disruptions in semiconductor supply caused by the pandemic, there is no time to relax as the next major supply shock is approaching, the report’s authors wrote. But this time, there are obvious indicators, and the business community has time to get ready.
The way forward necessitates alertness, strategic planning, and prompt action to strengthen supply chains. Business executives can secure their success and resilience in an increasingly AI-enabled world by taking proactive steps.