The hype around AI has continued to gather momentum over the last few years, and with the AI supply chain predominantly in emerging markets countries, the theme is a key one for investors in the asset class. Guido Giammattei, gives his views on how AI stocks could dictate performance in the near term.
While there is little doubt about the structural growth offered by the AI theme, the performance of stocks exposed to this theme has become quite extended. On the one hand, investors may continue to chase these stocks, and this continues to drive the markets higher. After all, demand remains strong, as does CapEx, and there are no signs of earnings downgrades. Valuations are very elevated but not ‘crazy’. However, eventually, markets correct. The dot com bubble hit a wall in March 2000, and ‘why’ and ‘how’ the correction started remains a mystery. It just happened.
The following four factors suggest caution, in our view:
The magnitude and speed of equity gains: since the start of 2023, Taiwanese equities have rallied 85%, led by a 150% rally in chipmaker TSMC1. In Korea, SK Hynix, the main supplier of memory for Nvidia GPUs, has gained 220%2. In the US, Nvidia has risen even further, gaining 800% over the same timeframe, while an HSBC AI index has gained 160% in the last 18 months3. Another measure of extreme price action, the MSCI EM Information Technology Index versus the MSCI Consumer Staples Index, was beyond the IT relative peak during the internet bubble of 1999/2000 until recently.
Valuations: most of the AI supply chain is located in Taiwan, where approximately 85% of market cap is in tech or tech-related industries4. Despite the recent correction, Taiwan has continued to trade at the largest premium to emerging markets (“EM”) over the last three decades, and valuations remain elevated in this area of the market in both absolute and relative terms.
Breadth: the 7.7% return in the MSCI EM Index to August 2024 was the strongest start to the year in five years but also the narrowest, with just three stocks accounting for 70% of the index returns (in North Asia AI and internet themes)7. In the US, where the AI theme originated, a mere 20% of stocks were outperforming (as at August 2024), which is the narrowest breadth in 20 years5. TSMC, the leading foundry business globally, reached 10% of the MSCI EM Index8 in the summer, the highest weight a single stock has had in the last 25 years.
AI monetisation remains unknown: last but definitely not least, AI monetisation is unknown. With the US hyperscalers’ capital expenditures poised to more than double from USD366 billion in the four years to 2021 to circa USD750 billion by the end of 20256, the question of monetisation against these investments rises to the forefront for many investors. This is because these CapEx are not generating any revenues at present. A lack of visibility around the expected margin profile of newer GenAI solutions, the split of investment between the training of models (potential future monetisation) and inference (supporting current monetisation), the accounting around the OpenAI relationship, and re-usability of training CapEx for future inference capacity leaves more questions than answers. This poses a risk to the continuation of the current earnings cycle for AI stocks, in our view.
Recently, we have seen a correction in tech stocks but this is very modest and doesn’t alter the view that the IT sector, and AI stocks in particular, remain vulnerable to further pressure, given stretched valuations, positioning and technicals.
Whether the recent pullback in these stocks is the start of a deeper, broader downward move or simply profit taking, we believe EM may be in a stronger position than developed markets (DM). The valuation gap between EM and DM tech is too large to ignore, and it should provide a buffer in case of further weakness or a tailwind in the event of ongoing strength.