AI Capex Concerns are Legitimate, but Also Unrealistic

Concerns about the payback from AI capital investment by hyperscale cloud computing giants including Alphabet, Microsoft and Amazon already have been an issue for equity investors. The day before the Alphabet earnings call (July 22, 2024), the stock price was $183.60. The day after the call the price was $174.37. 

Alphabet lost about $113.28 billion in equity value the day after its July 23, 2024 earnings call, and the total change in equity value for the following week was approximately $145.64 billion.

Similar damage could occur to other hyperscalers in the “cloud computing as a service” space, if investors do not see material increases in revenue and also hear forecasts of continued high capex. 

source: Reuters, LSEG 

Some of us not in the financial analyst business might find such expectations unreasonable. 

In part, that is because expectations for providers of software services generally anticipate high profit margins and relatively quick payback from capex, compared to providers of other services with a more utility-like character.

Even within the cloud computing business, capex might be expected to breakeven in two to four years, but not produce a payback for three to five years. In other capital-intensive industries, breakeven periods routinely range from five to 15 years, with payback taking seven to 20 years. 

Industry	Expected Breakeven Period	Expected Payback Period
Software	1-3 years	2-4 years
Cloud Computing	2-4 years	3-5 years
Communications Networks	5-7 years	7-10 years
Airlines	7-10 years	10-15 years
Real Estate	5-10 years	10-20 years
Utility Industries	10-15 years	15-20 years

Of course, financial analysts get paid to predict quarterly to annual results. Enterprise CEOs are judged on annual performance. But analysts and researchers often work with longer time frames. 

So firms will be punished for what is seen as “excessive” AI capex. What might not be immediately clear is the strategic impact half a decade to 20 years out. And that is the balance the cloud computing hyperscalers must now strike: investing in a prudent manner now while avoiding the risks of underinvesting. 

If AI winds up becoming a general-purpose technology, investing and adoption laggards might suffer to some degree. The problem is that nobody now knows what levels of investment are “too little” and which might be “too much.” 

Cloud computing provider revenues from customers are going to be the real test. But expectations about the degree of financial return, and the magnitude of return, have been unrealistic from the start.

Like it or not, many important capex investments take quite some time to show payback. So expectations of near-term financial gain seem quite unreasonable.