In the week ending 24 January, a previously unknown Chinese start-up presented an artificial intelligence large language model apparently showing a performance comparable to that of the leading US models. The unexpected development could disrupt the paradigm of a quasi-monopolistic position for leading US chip manufacturer. This concern led to a major fall in valuations of some prominent tech companies. Markets have since recovered most of their poise. Here we give our view on these developments.
On Monday 27 January, the stock price of the leading US producer of powerful artificial intelligence (AI) processors fell by nearly 17%. The loss was partly reversed the following day when the same stock rallied by 9%, leading other US technology stocks higher. The NASDAQ Composite index rose by 2% on 28 January after the previous day’s sharp 3.1% decline.
The 27 January sell-off appears to have been something of a perfect storm created by factors including:
- Valuations priced for perfection
- Crowded positioning in companies exposed to datacentres
- A complete surprise to the market triggering a ‘shoot first, think later’ reaction.
Artificial intelligence is still a secular growth theme
We remain positive on artificial intelligence (AI) as a secular growth theme. More efficient models should drive broader adoption of AI faster and there is still a lot of room for innovation requiring more computing power.
Companies investing in AI will continue to invest to develop and train models, while at the same time adopting techniques to make their models as efficient as possible. This could result in investment and implementation cycles which are likely to be unsynchronised between companies.
In our base case, there is potential for slower growth in capital expenditure. In the worst case of simultaneous near-term cuts in spending, there would be a rollout period where new technology was implemented followed by resumed growth. This would have implications for the pace of demand for semiconductors and digital infrastructure equipment.
Implications for selected industries
Major cloud service providers remain well-positioned to benefit from the democratisation of AI. Reduced inference costs (inference being the process chips use to derive logical conclusions from premises known or assumed) and the increasing use of AI applications are expected to drive more IT migrations to the cloud.
A scenario with slower growth in capital expenditure expectations should ease the investment burden and potentially improve returns on investment for cloud players.
Additionally, hyperscalers (the large-scale datacentres that offer massive computing resources, typically in the form of elastic cloud platforms) can monetise AI innovations by building applications above the digital infrastructure layer or using AI to enhance existing services.
Application and platform software providers will largely be unaffected or could slightly benefit, depending on their applications of AI. Lower inferencing costs will result in broader adoption of AI features. Software margins could benefit (slightly) in scenarios where they currently bear the cost burden of inference (which is falling) and do not pass along any costs to the end user. We expect AI features to proliferate within software and potentially create stronger engagement trends for those providers who execute well.
On the cusp of more demand?
We continue to pursue a diverse approach to investing in AI, so we plan no major changes to our holdings in the short term. Carefully monitoring of earnings results, trends in capital expenditure and further AI model innovations continues, so we can react appropriately to new information.
If this new Chinese model is indeed more efficient (and that remains to be confirmed), this breakthrough could speed up the adoption of AI, creating more power demand overall. In other words, economies would be subject to the Jevons paradox, the effect named after English economist William Jevons, who observed in 1865 that technological breakthroughs leading to the more efficient use of coal ultimately increased the overall consumption of coal.
Also read Artificial intelligence is everywhere.
