Jan 23, 2026

Rajiv Jain on AI malinvestment

I foresee disappointment.  We talk to dozens of companies monthly and large consulting firms focused on the AI rollout.  There is almost no evidence of profit-margin improvement.  People use AI to summarize email and do ChatGPT-type searches.  Ninety percent of ChatGPT use is outside the U.S. 

An important MIT study came out over the summer that found that 95% of AI projects found no real use.  In our view, Microsoft Copilot has been a total failure…  GPU pricing is in free fall.  Based on our research, Nvidia’s Blackwell chips can be rented for $4.00 an hour.  Blackwell chips are already being rented at a discount versus just a few months ago.  If demand was so strong, why have Blackwell rentals been collapsing? 

There are 200 neocloud companies today.  Depreciation rates are out of line with reality.  H200 chips, for example, were selling at 50%-plus off within nine months of release.  A few large enterprises mentioned to us that real economic life when used for training is barely 18 to 24 months.

Most of the growth at the cloud hyperscalers is coming from start-ups, and who is funding these start-ups?  More than half of the $150 billion that start-ups invested in the cloud was funded by the large tech companies.  This is a giant, circular trade.  I agree with Henry [Ellenbogen] that a lot of smaller companies will see productivity gains from AI, but margins are falling at the cloud providers.  There is an incentive for semiconductor companies and other tech companies to talk up the benefits of AI, but there is little evidence of profit-margin improvement almost anywhere because of AI, except at companies selling shovels to the gold diggers.

~ Rajiv Jain, "The Roundtable: Part 1," Barron's, January 10, 2026

 

No comments: