Skip to the content

Cloud giants face AI capex conundrum

05 December 2024

Growth capex stacks up favourably compared to the other ways companies put surplus cash to work.

By James Knoedler,

Evenlode Global Equity

There have been significant changes in the dynamics affecting many companies from a bottom-up perspective. An example is the current boom in capital expenditure (capex) in the large cloud service provider companies, which has driven us to think further about the valuation of these companies.

Companies have four primary uses of surplus operating cashflow, having paid for maintenance capex: growth capex, mergers and acquisitions (M&A), dividends and share repurchase (for this analysis we exclude ‘pile it up in the bank’).

Traditional free cashflow (FCF) valuation measures deduct growth capex but ignore the other three uses, which in our view is overly punitive to companies that focus on organic growth and overly favourable to those which grow inorganically and/or overdistribute earnings.

 

Compound edge

We exclusively look at companies with profound competitive advantages which enjoy returns on capital meaningfully above their cost of capital.

The greatest of the privileges this confers is not the excess cashflow it yields every year, but the ability to reinvest some or all this cash into new assets with predictable returns well above those available in capital markets. This is the critical element for compounded share price returns.

If we compare growth capex to the other uses, it stacks up favourably in several dimensions. Dividends are arguably just as fixed a commitment as maintenance capex, and investors who do not have immediate uses for the cash are burdened with reinvestment risk, usually at returns far below those available on incremental internal investment.

Share repurchases carry valuation risk. Incremental returns on repurchases are very price sensitive and our view is that share repurchases are on the whole best used in opportunistic, ‘big bang’ transactions to retire equity in turbulent markets, rather than as a continuous capital-return mechanism.

M&A carries both valuation and due diligence risk and tends to scale badly.

In conclusion, as long as there is scope for reinvestment, growth capex is our preferred use of cashflow. Almost all company managements will also say this when asked, and we think it makes sense to value companies accordingly, rather than punishing them for organic reinvestment.

Our approach therefore focuses on gross cashflow, unlevered and with share option expense deducted. This means that we capitalise the returns to come from current growth capex.

Given the attractive competitive advantages our companies enjoy, they usually bear less risk of returns on capex not materialising than the average company. This has served us well in the past, allowing us to ‘look through’ investment cycles at holdings like Broadridge and Amazon.

 

Capex crossroads

The size, pace and feverish commentary around artificial intelligence (AI) capex, however, has given us some pause. While there is huge demand for AI computing capacity, there are big unresolved debates ongoing about the shape and timing of this demand.

This is an important difference from past episodes of technology company investment and digestion. If we look at the analogous instance of Alphabet’s capex ramp in the late 2010s, it was founded on very predictable returns from a) growth into enterprise compute to catch up with Amazon and Microsoft and b) deployment of machine learning models into the search advertising market, one which Alphabet had dominated for years.

Currently, a large chunk of demand is for model training – the ‘build it’ phase – rather than inference, which is driven by end users interacting with models – the ‘and they will come’ phase.

The signals which we can see suggest, so far, limited uptake of generative AI, particularly in revenue-generating settings. Model training is getting exponentially more expensive in a multi-player arms race, but the payoff for the winner(s) is not yet clear. It could be that generative AI yields artificial general intelligence (AGI), in which case all bets are literally off; or that it fizzles out like the metaverse and blockchain; or, most likely, it lands somewhere in between.

We believe that these concerns over the short-term profitability of AI are becoming more widespread and likely contributed to the slowing price of the ‘Alternative Seven’ (Alphabet, Amazon, Apple, Broadcom, Eli Lilly, Microsoft and Nvidia), of which six have direct exposure to AI.

Within this group, Microsoft, Amazon and Google are holdings in the Evenlode Global Equity portfolio. Our view is that while their core business model in cloud remains very attractive, the spread of possible outcomes has broadened materially and the risk of future returns on capital being structurally lower cannot be dismissed out of hand. Given this situation we have been reassessing our position sizes and position risk limits for these companies.

James Knoedler is a portfolio manager on the Evenlode Global Equity fund. The views expressed above should not be taken as investment advice.

Editor's Picks

Loading...

Videos from BNY Mellon Investment Management

Loading...

Data provided by FE fundinfo. Care has been taken to ensure that the information is correct, but FE fundinfo neither warrants, represents nor guarantees the contents of information, nor does it accept any responsibility for errors, inaccuracies, omissions or any inconsistencies herein. Past performance does not predict future performance, it should not be the main or sole reason for making an investment decision. The value of investments and any income from them can fall as well as rise.