Reasons why the AI stock correction may be overdone

Unlock the Editor’s Digest for free

Roula Khalaf, Editor of the FT, selects her favourite stories in this weekly newsletter.

You know stocks are getting pricey when investors get jumpy about even the slightest change in earnings expectations. This can be seen in the sharp moves in AI-related stocks in recent weeks, especially the sector’s top picks such as Nvidia, ASML, Arm and US and Asian chipmakers. But these need not move in step with each other

Take for example, ASML. The Dutch chip equipment maker’s weaker-than-expected orders sparked extreme volatility in AI-related stocks around the world. The logic seemed to be: since new orders for machines made by the world’s largest advanced chip equipment maker fell short of market expectations, the outlook for artificial intelligence chip growth must also be deteriorating.

Indeed, conditions are ripe for a correction. The AI stock market boom has been pricing in too much too fast. The hype has meant disproportionate gains for even some stocks that are unlikely to become significant beneficiaries of AI-driven growth. Meanwhile, the sector remains heavily exposed to geopolitical risk. China is a key market for most chip-related companies including Nvidia and ASML. For the latter, it is its biggest market, accounting for nearly half its system sales in the first quarter

Even so, recent sharp swings in AI stocks highlight market hypersensitivity to relatively benign cyclical moves more than anything else. The reality is that the AI chip market can continue to expand rapidly despite apparent weakness in the broader semiconductor sector. That is because chips made for AI applications, despite their phenomenal sales growth in the past year, still account for a tiny fraction of the world’s supply of chips. For example, take high bandwidth memory chips, a critical component in all AI chips. Around six of these advanced memory chips are needed to power a single AI chip, including those from Nvidia. 

Yet by sales volume, they currently account for just about 1 per cent of the total memory chip market. Even at today’s growth rates, AI chips cannot offset the cyclical downturn in the rest of the vast market, which remains dominated by traditional memory, storage and processing chips used for smartphones, cars and electronic devices. 

Here, companies are currently working through an excessive level of inventory — a result of overproduction by chipmakers and hoarding by device makers during the pandemic shortage. This higher than normal inventory is keeping end market demand suppressed. Thus, there has been little urgency for chipmakers to increase production capacity, and new chipmaking gear orders, in recent years. Prices of memory chips have dropped more than 50 per cent since 2021, with spot price declines continuing into this month. Demand recovery remains especially slow among auto and smartphone makers, two of the largest consumers of chips.

Moreover, fluctuations in ASML’s orders by quarter has not usually been reason for concern. ASML sells just about 100 new lithography systems each quarter. Around three companies buy most of its advanced machines, making quarterly numbers prone to fluctuations. One quarter’s miss is not necessarily indicative of results for the rest of the year.

On the contrary, a prolonged period of lower than expected equipment orders from chipmakers would likely mean a higher probability of large orders in the coming quarters. Around the world, more than 70 new chip fabrication projects are being built. TSMC and Samsung are expected to begin mass production of next-generation 2-nanometre chips next year. For that, orders of new equipment — which only ASML can make — for the new fabrication lines would need to be placed this year.

The latest results from TSMC help confirm this. Despite cutting its expectations for chip market growth, excluding memory chips, this year, it has left its capital spending plans for this year unchanged at between $28bn and $32bn. It is currently building new plants in countries including the US, Japan and Germany, and plans to start production of ultra-advanced 1.6-nanometre chips are already in place for 2026.

It is worth remembering that the underlying drivers of the AI stock boom over the past two years has had little to do with the cyclical ups and downs of the traditional chip and equipment sectors, and more to do with expectations of AI’s business transformation potential. For those looking for a sign of peak AI hype, the focus should be on the mismatch between steep valuations and the slower than expected pace of broader enterprise adoption of AI, rather than normal cyclical forces.

june.yoon@ft.com