FT Article by Simon Edelsten – ‘Why the lessons of history are making me sceptical of AI stocks’
Why the lessons of history are making me sceptical of AI stocks
Simon Edelsten
This article was originally published in the FT Adviser
There’s an old saying in investment: when someone strikes gold, buy the local hardware shop and sell shovels to prospectors.
Today’s gold rush in equities is focused on artificial intelligence. Oracle and Broadcom, both mentioned in previous columns, have recently enjoyed results that suggest the boom exceeds even fevered expectations.
We took some profits in Oracle on its share price rising a third in just one day. We like this shovel shop, but we’re not betting the ranch on it. Here’s my thinking.
A brief history of AI
Computing power and coding periodically make hops forward. In the past decade Nvidia chips have helped enable leaps in gaming and crypto and now AI.
To understand AI’s arc of development, it’s worth asking why the chicken didn’t cross the road.
In the late 2010s coders working on translation and other language software faced a problem called the ‘issue of reference’. In the sentences below software would struggle to work out whom the “it” refers to.
‘The chicken didn’t cross the road because it was too tired.
‘The chicken didn’t cross the road because it was too wide.’
“It” changes reference. By looking at text patterns, the programmers coded ‘transformers’ (the T in GPT: generative pre-trained transformer) to handle this challenge. But their process entailed looking at these sentences letter by letter, forwards and backwards. That isn’t how our brains do this. So is AI that intelligent after all?
It reminds me of my first Sinclair calculator in 1975. It could do impressively complex multiplication — well, it impressed me back then — and at incredible speed. But it was calculation, not intelligence. Throw a non sequitur or absurd question into the mix and your AI machine will struggle.
Training phase
Using lots of fast processors (Nvidia’s are the best), companies are training chips to get better at finding patterns in letters and code. This requires two things: lots of data centres (Oracle’s territory) and lots of coders. Both have become rather pricey.
The big surprise in the Oracle results was the company’s prediction that cloud revenues would rise from $18bn to $144bn in the next four years. There seems to be one large order at the centre of this boom, from OpenAI.
All very exciting, but OpenAI is itself quite a gamble. Being a private company, it has opaque finances. We do know that its staff and data centre costs are rocketing. Word is that its estimated cash burn next year could be $17bn rather than the $8bn previously thought, and $35bn rather than $20bn the year after.
Oracle has got its strategy right and built hardware stores on the new California gold fields. But will its biggest customer have the funds to pay for the order? In the last week Nvidia has invested $100bn into OpenAI, which will help considerably, but funding them to buy capacity in warehouses full of Nvidia chips seems a bit circular.
Application phase
Arguably, this depends on how much income it can generate. And there’s the rub. We’re already seeing useful applications of AI technology. We have better predictive text. We have GPT writing book reviews, summarising legal documents and optimising search results. However, OpenAI’s latest release, ChatGPT 4.5, cost a lot to develop and left reviewers underwhelmed. It looks like the folks at OpenAI called it 4.5 rather than 5 because they knew it was only half a step forward.
More exciting may be advances in applications based on code, like building websites and generating AI images. Microsoft has claimed it can give coders an ‘auto complete’ that reduces coding time by 20 per cent.
But does it translate to enhanced revenues? So far Office co-bots have failed to drive significant price rises. Customers expect improved software functionality with every update anyway. Only a big step-change in function will make price hikes palatable.
The broader application of advanced computing to various industrial processes — from factory automation to facial recognition, to diagnostics, to autonomous driving — looks more promising. This will doubtless involve chips designed for specific applications (which is where Broadcom comes in) and coding to run those chips (which AI will facilitate).
However, Baidu, the main Chinese search engine, has just said it will do AI with a chip designed in-house (Kunlun P800), so maybe China-designed chips are becoming competitive on price and effectiveness.
These applications will need lots of data centres to run, but demand for capacity might fall beyond the training phase once vast volumes of letters and code have been scanned to build models.
Oracle’s share price is arguably based on data centre demand carrying on growing into the application phase. It may. But it may not. And there’s competition out there.
Some applications will suit the Microsoft Azure cloud, some Amazon’s AWS and some new centres from Oracle and others. Working out how much cake each gets or what return on capital is likely after the initial surge is tricky, but it’s not the glory-or-bust bet of developing GPT.
So I’m still focused more on provisioning than prospecting. For me that means Oracle, IBM, Broadcom (which I’ve owned in preference to Nvidia), Cadence (which helps people design new chips) and Fortinet (as the data needs good cyber security). All these stocks are expensive, so, though I like them, I’m not getting carried away.
This research material is for distribution only to Professional and eligible counterparties as defined by the FCA and also to persons of a kind to whom it may lawfully be promoted by an authorised person by virtue of Section 238(5) of the Financial Services and Markets Act 2000, the Financial Services and Markets Act 2000 (Exemption) Order 2001 and COBS 4.12.4R. and should not be relied upon by any other persons. Therefore circulation of this research material must be restricted accordingly.
This material is provided for information purposes only and does not constitute a solicitation in any jurisdiction in which such solicitation is unlawful or to any person to whom it is unlawful. Moreover, it neither constitutes an offer to enter into an investment agreement with the recipient of this document nor an invitation to respond to it by making an offer to enter into an investment agreement. This research material may contain forward-looking information that is not purely historical in nature. Such information may include, among other things, projections, forecasts, estimates of yields or returns, and proposed or expected portfolio composition.
Moreover, certain historical performance information other investment vehicles or composite accounts included in this material and such performance information is presented by way of example only. No representation is made that the performance presented will be achieved, or that every assumption made in achieving, calculating or presenting either the forward-looking information or the historical performance information herein has been considered or stated in preparing this material. Any changes to assumptions that may have been made in preparing this material could have a material impact on the investment returns that are presented herein by way of example.
This material is not intended to be relied upon as a forecast, research or investment advice, and is not a recommendation, offer or solicitation to buy or sell any securities or to adopt any investment strategy. The opinions expressed are as of October 2024 and may change as subsequent conditions vary. The information and opinions contained in this research material are derived from non-proprietary sources deemed by the Firm to be reliable, are not necessarily all inclusive and are not guaranteed as to accuracy. There is no guarantee that any forecasts made will come to pass. Reliance upon information in this material is at the sole discretion of the reader. Past performance is not indicative of future results. The value of your investment may go down as well as up, and you may not receive upon redemption the full amount of your original investment.
The Investment Manager is authorised and regulated by the Financial Conduct Authority and registered in England and Wales (Company Number: OC304213). Its registered office address is 6 Stratton Street, Mayfair, London W1J 8LD United Kingdom.