The following quote is a slice of stock market history. It was delivered in a letter to shareholders by the CEO of Sun Microsystems, a US technology firm founded in 1982, in the aftermath of the dot-com bubble.
31st March 2002
These words have gone on to serve as a stark illustration of the irrational exuberance that drove stocks to absurdly high valuations in that period. Sun Microsystem’s share price vaulted 965% in the three years to September 2000, fuelled by a 250% rise in profits. Earnings, however, proved unsustainable. Just two years later, the company was loss-making and the share price had cratered 96% from its $259 peak. The company subsequently cycled in and out of profitability before being sold to Oracle in 2010 at just $9.50 per share.
Investors seem to have short memories: today, valuations for a worryingly large proportion of the US stock market rely on similarly fanciful assumptions as those highlighted by Scott McNealy. As shown in the chart overleaf, the proportion of the largest US companies (as measured by their market capitalisation) trading at or above the 10x price-to-sales ratio that McNealy considered “ridiculous” has never been higher. This ought to send a shiver down investors’ spines: it took fifteen years for the tech-heavy Nasdaq index to recover its post-bubble losses.
Fig.1: Weight of Names in the S&P 500 Index Trading at a Price-to-Sales Ratio of 10x or More
Source: GQG Partners LLC. GQG provides this information for informational purposes only. GQG has gathered the information in good faith from sources it believes to be reliable, including its own resources and third parties. However, GQG does not represent or warrant that any information and any third-party information provided, is accurate, reliable or complete, and it should not be relied upon as such. GQG has not independently verified any information used or presented that is derived from third parties, which is subject to change.
The common riposte to Scott McNealy’s scathing takedown of his shareholders’ financial literacy – or the lack of it – is that, as CEO, he simply failed to deliver the expected growth that had underpinned Sun Microsystems’ exalted valuation. We hear this same argument touted widely and repeatedly today: an unstoppable AI revolution will power growth for years to come, and what appear high valuations today will eventually be revealed as bargains as the scale of the opportunity is captured in corporate profits. We take issue with this argument. As Howard Marks, the sharp-witted founder of Oaktree Capital Management, likes to emphasise, “there is no investment idea that is so good that it can’t be spoiled by too high an entry price”.
To illustrate our concerns, let us examine some of the candidates that, we believe, could prove to be the Sun Microsystems of 2025. To begin, consider Nvidia, the dominant producer of the GPU computer chips that have powered the AI goldrush thus far. It currently trades at a price-to-sales ratio of 23x (nearly two-and-a-half times Scott McNealy’s “ridiculous” threshold), giving it a market value of $4.4trillion – the highest of any company in the world.
Nvidia is expected to grow its earnings at a mind-boggling 40% per annum for the next five years. If we accept these forecasts and assume Nvidia is able to defend an industry-leading profit margin of 55% over this period, implied revenues five years down the line are $1.1trillion – not far off the total current revenues of Apple, Alphabet, Microsoft and Meta combined. Place a price-to-sales ratio of 2.4x on this – in line with the long-term average for large US companies – and you arrive at a market capitalisation of $2.7trillion, still some 40% below Nvidia’s current level.
Clearly, Nvidia’s unprecedented rate of earnings growth will have to persist for more than five years to justify current valuations. Let’s stetch credibility and assume it delivers 40% growth and a 55% profit margin for another five years1 . At this point, an average price-to-sales ratio would justify a market capitalisation of $14.5trillion, representing a healthy – though not stratospheric – 12.6% annual gain for today’s investors.
The problem is that Nvidia’s annual revenues at this point would surpass $6trillion. We estimate this would amount to roughly one quarter of all the revenues of America’s 500 largest companies, or one seventh of total US GDP in 2035. To put it another way, it would require Nvidia to sell more than $16billion worth of computer chips every day of the year. At current prices, this means selling nearly 250,000 of Nvidia’s highest spec GPUs every day. This daily target is more than ten times the total number that Microsoft’s Loughton data centre – the UK’s largest ‘supercomputer’ – will eventually house by the time of its completion in 2028. We question how – economically and logistically – this can possibly be achieved.
For a second example, let us now turn to OpenAI, the progenitor of ChatGPT. OpenAI’s latest funding round ascribed it a value of $500billion, representing a price-to-sales multiple of some 38x its estimated revenues (nearly four times Scott McNealy’s threshold).
OpenAI is deeply loss-making, with CEO Sam Altman recently declaring that becoming profitable is “not in [his] top ten concerns” – just as well, given OpenAI expects to burn through $115billion of cash by the end of 2029. The firm’s deeply negative cashflow has not prevented it from signing a plethora of deals with a far-reaching network of US tech companies that, in aggregate, commit OpenAI to spending around $1.4trillion in the next seven or so years. Investors should question where Sam Altman hopes to find the money.
True, ChatGPT has enjoyed the fastest growth in users of any consumer-facing application in history and expects to hit one billion users by the end of this year. If it can just make $200 off each of these users in every one of the next seven years, then that $1.4trillion investment might perhaps be affordable. Unfortunately, an estimated 95% of ChatGPT’s users are currently to be found on the free version: the firm has just 35million paying individuals, and 1million business subscribers. With so few subscribers, it needs to make $5,555 from each of them every year to pay for the planned investment. But if it can multiply its subscriber base by a factor of ten, then annual per user revenue of $555 is a much less daunting hurdle.
However, there is a difference between revenue and profit. Even if there are 360million people willing to subscribe to the service2 , growth may not, in fact, be the answer. It is thought that OpenAI currently makes a loss each and every time a paying subscriber runs a query on ChatGPT. Growing the paying user base may increase revenues, but only at the expense of ever larger operating losses.
As challenging as these examples may be, we believe that to consider them in isolation is to miss the bigger picture. To meet its growth targets and justify its current valuation, it seems highly likely that Nvidia needs OpenAI to succeed. If it does so, then perhaps the valuations of Microsoft ($3.6trillion market cap, 12x price-to-sales), Oracle ($592billion market cap, 10x price-to-sales) and AMD ($354billion market cap, 11x price-tosales) among others are supportable – these three are among those entangled in OpenAI’s ambitious network of promised investment. Even still, how profitable can these companies expect to be if they are the source of the astronomical revenues Nvidia needs to generate to justify its status as the world’s most valuable company?
Furthermore, success for OpenAI and Nvidia would surely spell doom for Alphabet ($3.9trillion market cap, 10x price-to sales), Meta ($1.6trillion market cap, 9x price-to-sales) and the many other AI firms hoping to emerge triumphant in a ‘winner takes all’ market.
It is possible that AI proves revolutionary in ways that we currently cannot imagine. However, even in this scenario, it stetches credulity to believe that the growth assumptions underpinning the ambitious valuations of all – or even most – of the participants can simultaneously be fulfilled. To repeat Howard Marks, “there is no investment idea that is so good that it can’t be spoiled by too high a price”.
We strive to keep abreast of developments and to maintain an open mind but, at present, we are not convinced that AI actually is a particularly good investment idea (we will explore this in more detail in our next investment note). However, we are convinced that today’s valuations are more than sufficient to ruin it.
Though the fear of missing out can be uncomfortable, we argue that it pales in comparison to the discomfort felt when unsustainable valuations and growth aspirations eventually collide with reality. As stewards of our clients’ wealth, we consider it paramount to never put you in a position which leads you to ask of us “what were you thinking?” We therefore continue to steer portfolios well clear of the areas of the market where valuations are most precipitous, where the growth forecasts are most demanding, and where the web of interconnected risks is most entangled.
