While companies can't predict outbreaks, they can guard against the next big threat to the semiconductor supply chain.

The supply and demand for semiconductors is a delicate balance that can be influenced. In recent years, the industry and its customers have become aware of this. While chip shortages caused by the epidemic have eased, executives are preparing for a new crisis caused by AI.

As AI is used more, there will be less supply of GPUs for data centres. This is because the computational resources needed to train and run large language models appear to be limitless. However, there are supply chain constraints. The rise of AI will also lead to more people buying PCs and smartphones. This will affect the wider semiconductor supply chain.

The supply chain is very complex. If demand rises by more than 20%, there could be shortages. AI is growing fast in many markets. This could create problems in the supply chain.

It is difficult to balance supply and demand in the semiconductor industry because technology is changing quickly, it costs a lot to expand and it takes a long time to get things up and running. Chip suppliers and buyers must act fast to avoid the next crisis. Let's look at how things might change if there are big changes in demand or supply.

Data centre demand

There will be a big increase in late 2022 thanks to new AI technology, which is good for the semiconductor industry. Chipmakers' sales and valuations have risen. This includes leading GPU vendors such as Nvidia and companies supplying other chips to data centres, such as Broadcom (switches) and SK Hynix (high-bandwidth memory). Data centre spending on dedicated chips is growing. Cloud service providers are expected to spend 36% more on CAPEX in 2024. This is because they are investing in AI and accelerated computing. GPUs will continue to be in demand as large-scale language models (LLMs) are developed. Venture capitalists are also investing in AI startups.

If data centre demand for GPUs doubles by 2026, suppliers of key components will need to increase production by 30% or more. Demand is for advanced packaging and memory. Manufacturers of chip-on-wafer packaging components must triple production by 2026.

Building data centres and fabs, and ensuring access to advanced packaging and sufficient power are all part of the process of driving AI. These things often take a long time to get, which can result in unmet demand.

Importantly, many of these supply chain elements are shared with other parts of the technology ecosystem and are all subject to capital, geopolitical and timing risks. The absence of one chip can derail an entire system, as happened with the last shortage, when new cars were massively slowed down by the lack of a key chip.

PC and smartphone demand

Personal device makers are adding AI to their products. Our benchmarks show that laptops and smartphones have about 5% and 16% more silicon surface area to accommodate on-device AI.

As AI becomes more useful, people may buy more devices, causing a short-term increase in demand.

AI will affect the semiconductor supply chain for smartphones and PCs more than GPUs because there are more components in smartphones and PCs. The most vulnerable link in the supply chain for these devices will be the factories that make the most advanced chips. If AI is used more, there will be more PCs and smartphones. This means that leading-edge fabs will need to make more products by 2026. This would mean building four to five more leading-edge factories at a cost of $40 billion to $75 billion. This would justify the many factories already being built by major foundries.

Don't forget supply risks

Over the past decade, extreme weather, natural disasters, and other major events have shown that supply shocks can limit the industry's ability to meet demand. GPU supply has been affected over the past 18 months by disruptions in less visible parts of the supply chain, such as CoWoS advanced packaging.

Geopolitical tensions, trade restrictions and tech companies moving away from China are still a risk to the supply of semiconductors. Other problems like delays in building factories and shortages of materials could also cause problems. Without these uncertainties, we expect the greatest supply risk to come from greater demand for high-bandwidth memory components, advanced packaging, and substrate factories.

Insights from executives

Semiconductor buyers need to understand the components they are sourcing to navigate complex supply chains. Good leaders will focus on components used in AI data centres, such as switches, transceivers and power management ICs. They will watch PC and smartphone updates, as well as related items like Wi-Fi routers and networking equipment. This will affect the whole supply chain, so each area must be tracked.

Leading companies will learn from the chip crisis to keep their inventories at the right level. They will sign long-term agreements to ensure they can buy chips and manufacturing capacity when they need them. The 'just-in-time' inventory strategy will give way to a 'just-in-case' approach that is more costly but more resilient. More companies will use standard semiconductors in their products. They will also invest in supply chains to deal with political issues like tariffs. They will also monitor silicon packaging and substrate supply as they do front-end semiconductor manufacturing.

Semiconductor supply may still be disrupted by the outbreak, but there's no time to rest, as the next shock is coming. This time, the signs are clear and the industry can prepare. We must be vigilant, think ahead and act quickly to strengthen the supply chain. Business leaders can stay strong and successful in an AI-driven world by taking action.