What might A.I. mean for investors?

The recent increase in stock prices in the U.S. day after day has been astounding.  For those of you that read my gibberish from time to time, I have laid the blame mostly at the feet of retail investors bitten by FOMO, accelerated by hedge funds that recognise and exploit the insatiable appetite of retail by piling into the same stocks (i.e. a momentum strategy).  The ongoing rally is clearly centred around the buzzword of the last couple of years – A.I., or “artificial intelligence”.  Companies involved in artificial intelligence, particularly those able to present themselves as such, have experienced significant increases in their stock prices.   In this respect, there are five things I would like to point out to my readers regarding A.I.:

• Tech dominance (A.I.-driven) of S&P 500: The Mag  7 stocks – Alphabet, Nvidia, Microsoft, Apple, Amazon, Meta Platforms and Tesla – have combined market value of nearly $21 trillion, which is 37.3% of the S&P 500 index.  All are involved to a certain extent to A.I.  As the most widely held stocks, they benefit the most from the broadening interest in anything and everything A.I.  Their concentration in the index and influence on overall returns (next bullet) has increased dramatically over several years. These same seven stocks collectively accounted for 21.5%, 28.6% and 35.3% of the S&P 500 index at the end of 2022, 2023 and 2024, respectively.

• Oversized influence of Mag 7 stocks on index returns:  Accounting for over one-third of the S&P 500 index, it goes without saying that the influence of the Mag 7 stocks as far as index returns is substantial.  The concentration and dominance of a handful of tech giants in the indices mean that fund managers which track the index (e.g. S&P 500 ETFs) or that measure their investment performance vis-à-vis the S&P 500 index have to adjust their portfolios accordingly, skewing their holdings towards the same popular stocks.  This is a reinforcing loop which simply adds further fuel to the proverbial fire.  The massive effect on index returns is visible when looking at the difference in returns between the actual S&P 500, which is market weighted (and hence dominated by the Mag 7), and a hypothetical S&P 500 that has the same stocks but is weighted on an equal basis.  To explain, the seven Mag 7 stocks – which currently account for 37.3% of the actual S&P 500 index – would only collectively account for 1.4% (0.2% each) of an equal-weighted hypothetical S&P 500 index.  Equal-weighted indices provide investors with the opportunity to “bet” on the other 493 stocks in the S&P 500 index.  However, the attractiveness of such a bet is suspect.  Year-to-date, the S&P 500 total return index has gained 14.5%, and the equal-weighted S&P 500 total return index has only gained 7.3%. 

  
  

• Substantial investment needed for A.I. rollout: The expected capital expenditures by companies to adopt and implement A.I. in the workplace are not entirely clear, but the investment for those companies on the leading edge of developing and promoting the roll-out of artificial intelligence and specifically large language models (“LLM”) will be significant.  Again, this can be seen most clearly with the Mag 7 companies.  It will take enormous investments to develop data centres, servers, software and other IT kit that moves A.I. forward. For example, Microsoft, Amazon and Meta have all said they will spend about $100 billion in capex in 2026 to fund their A.I. roll-out, and Alphabet (GOOG) is not far behind, pledging $85 billion in capex for 2026.  

• “Circular” A.I. deals in the epicentre:  Over the last few months, a number of A.I. related companies have struck deals with other companies in the A.I. value chain that have involved substantial amounts of investment.  Pundits have rightly pointed out that some of these arrangements are circular, meaning that an investment by one company in another usually has a quid pro quo in terms of the company receiving the investment then agreeing to buy product from the company that is investing.   In some respect although commercial, it has similarities to supply chain financing. For example, on September 22, NVIDIA announced that it would invest $100 billion progressively in OpenAI to develop 10 gigawatts of A.I. data centres that would use NVDA’s chips / systems (NVDA press release here).   This was followed on October 6 by a similar agreement between AMD and OpenAI in which AMD will sell up to 160 million shares of stock to OpenAI in exchange for OpenAI purchasing cutting edge A.I. chips from AMD (AMD press release here).  There are naturally concerns that these sorts of agreements are circular and so intertwined that the fortunes of these companies will rise – or potentially fall – together.  Nonetheless, each such agreement seems to boost the value of any company that is commercially involved with the two major players at the moment, NVIDIA and OpenAI.  The diagram below from #Bloomberg illustrates some of these intricate relationships in the epicentre of the A.I. value chain.

  
  

  

  
  

Recall that OpenAI is a private company and still a “not-for-profit” although recent investment rounds have valued the company at around $500 billion, which – if it were a public company in the S&P 500 index – would make it the 18^th most valuable company in the index.  Aside from employee / insider ownership, the largest single investor in OpenAI is Microsoft, with a stake estimated at 28% to 30% pro forma for recent transactions involving OpenAI.  The graphic below from the #FT illustrates the ownership of OpenAI.

• Collateral beneficiaries of A.I.: The last thing worth mentioning is that there are visible collateral effects of A.I. on companies that are just outside the nucleus of the direct A.I. participants.  The stocks of companies on the periphery needed to develop A.I. stand to benefit from the rollout.  The industry that most comes to mind as far as being a collateral beneficiary is the power industry, since the power requirements of A.I. data centres are expected to far exceed existing power capacity.  In a report published last year (here), Goldman Sachs estimated that the U.S. will need to add 47 gigawatts of new power capacity by 2030 to fund the A.I. evolution, which is roughly a threefold increase in capacity compared to 2023.   Most of this incremental capacity will come from regulated utilities, normally a “boring” low growth, low Beta sector because of its predictable cash flows.  The unprecedented growth in power demand over such a short period is reflected in the performance of the utility sector, which has outperformed the S&P 500 index since the beginning of 2024.  As far as the adaptation of A.I. by companies not directly or indirectly involved in its evolution (e.g. consumer products, healthcare, energy, etc), the uptake and benefits remain to be seen although the productivity improvements are expected to be enormous. 

The fuss about AI

There is little doubt that A.I. will dramatically change the world, and probably everyday life for many people.   Hopefully, changes overall – following what are likely to be “bumps in the road” during the transition period – will be for the better.  The questions are: what sort of productivity improvements resulting from the adaptation of A.I. can be realistically expected and how quickly they will occur, and how much investment will be required for these A.I.-related productivity enhancements to be realised? 

The case supporting higher valuations of companies that are involved in, or are able to spin a narrative about, A.I. is based on expectations of materially faster revenue growth, dramatic margin improvements attributable to productivity improvements, or some combination of both.  Setting aside the great unknown of how much and how quickly A.I. will improve corporate earnings, the questions that sceptical investors have include:  

• Is the substantial investment in AI “right sized” appropriately or is it “too much too soon” for tangible returns to keep up with the substantial investment required,

• More generally, what will be the trajectory of the productivity increases in non-tech companies (i.e. those on the periphery) which should eventually benefit from the adaptation of A.I. to reduce costs and lead to more efficiencies?

  
  

• Will A.I. generate sufficient productivity gains over the next few quarters / years to offset what appear to be some concerning headwinds facing the global – and specifically the U.S. – economy, including: growing federal debt / deficits; political gridlock in the U.S. and other countries as the electorate becomes more polarised; persistent inflation which remains above target; a slowing jobs market; and the fiscal effects of the Trump tariffs as the U.S. becomes ever-more isolationist?

There is no doubt that these three areas are intertwined.  And it spite of these cautionary concerns it is clear that many equity investors – particularly in the U.S. – are all-in on A.I.  Parallels to the dot-com boom and bust are often raised, and it’s hard to deny that there are similarities to this last memorable boom-and-bust cycle around 25 years ago.  Indeed, many investors are convinced that they have seen this movie before.  The counter argument from A.I. advocates is based on the premise that productivity improvements resulting from the adaptation of A.I. will be fundamentally more powerful than the productivity enhancements resulting from the rollout of the internet 25+ years ago. 

Personally, I do not feel particularly well qualified to comment on how the A.I. rollout will proceed, but I do remember well what happened in the early 2000’s to stocks which undoubtedly influences my view.  Excitement in the late 1990’s over the opportunity associated with the proliferation and availability of the internet led to investor FOMO, which caused nearly every stock even remotely tech-related to get bid to ridiculous valuation levels by early 2000.  The tech-heavy NASDAQ Composite index increased fourfold between 1995 and 2000, reaching a P/E at its peak of 200x.  Venture capital poured into into anything and everything internet related, and tech IPO’s were the hottest item in the investment community during this period.  However, it all came to a screeching halt starting in March 2000, when the bubble burst.  By September 2002, the NASDAQ Composite was down 77% from its March 2000 highs.  Money to finance internet-related ventures dried up completely, and several high profile tech and telecom companies simply ran out of cash and failed.  Showing how much over its skis the tech benchmark index had got, it would take 15 years for the NASDAQ Composite to reach its former high.   As many of you might recall, this episode around a productivity-enhancing new technology did not end well for those investors that were buying towards the end of the boom.

Whether or not you believe that there are parallels to the dot-com boom-and-bust cycle, I for one have little doubt that investors playing the A.I. theme will at some point get ahead of themselves.  This is my opinion from an investment perspective, not a commercial perspective.  I am increasingly concerned about valuations, along with the one-way travel higher of the price of any stock even remotely related to A.I..  Having said this, I completely  realise that there are compelling reasons for the adaptation of A.I. related to sharp expected improvements in productivity.  Hopefully, these benefits will eventually trickle down to everyone and improve all of mankind.  The march towards widespread adaptation of A.I. will be erratic, but it cannot and will not be stopped.  

I would not think of betting against A.I., but I do feel that the very material investment amounts needed, and the magnitude and speed of productivity improvements, are difficult to determine at this point, although stock investors appear to be betting on the most optimistic outlook.  I hope they will be proven right!