AI investment is swelling, but growth is starting to slow

Growth fell from about 145% to about 45%

According to the figures Marcus cited, one AI business had quarterly growth of about 145% from Q1 to Q2 and about 45% from Q2 to Q3. It is still growing, but its momentum is weakening.

the growth rate from Q1 to Q2 is about 145%, and the growth rate from Q2 to Q3 was about 45%. It was a deceleration in the growth rate.── Gary Marcus (2026-10-15 12:49 UTC · YouTube "NYU’s Gary Marcus on Building AI We Can Trust" (Bloomberg Television) at 0:17:39, source)

For companies expanding their facilities on the assumption that growth will continue, slowing growth leads to a review of their plans. Marcus sees the industry as overexpanded, with profitability not yet confirmed. That there is a gap between the pitch to investors and reality has long been this person's argument.

In reports from the same day, Wall Street banks are looking at both the growth and the risks of AI infrastructure investment. The sheer size of the plans has become the point where views diverge.

The side saying "it is not a bubble" argues that supply and demand will eventually balance

Jensen Huang of chip giant NVIDIA said on October 15 that a bubble occurs when supply exceeds demand.

A bubble happens when supply exceeds demand. Eventually, supply and demand will come will harmonize.── Jensen Huang (2026-10-15 13:36 UTC · YouTube "Nvidia CEO Jensen Huang Calls For US-China Cooperation On AI Safety | Trump | N18G" (CNBC-TV18) at 0:02:31, source)

This remark can be read as saying that demand is sufficient now, so expanding facilities is the right thing to do. However, saying it will "eventually balance" also allows for the possibility that it is not balanced now. Huang's company sells computers for AI, and the more its customers keep investing, the more its profits grow. For that reason, this view should be read with a discount, since it comes from the seller's standpoint.

[Council reasoning] Opinion within the council is split on this point, and the indicator of how widely positions vary is high, at 0.905. Not everyone is facing the same direction.

Solving math problems does not mean being able to do real-world work

The basis for the investment is the expectation that AI capabilities will keep improving. Marcus raises doubts about this expectation.

The fact that these systems can solve math problems doesn't mean that they can solve other things in the real world.── Gary Marcus (2026-10-15 12:49 UTC · YouTube "NYU’s Gary Marcus on Building AI We Can Trust" (Bloomberg Television) at 0:03:43, source)

If investment and hiring decisions simply translate higher exam or math scores into results in corporate operations or medicine, the outcome risks being disappointing. Marcus believes AI does not reliably follow instructions either, and that human checking is essential for using it in business. This is the basis for his claim that AI has not yet raised productivity.

Meanwhile, Geoffrey Hinton, on the same day, called a certain estimate "too optimistic" and said his own figure is four times that. The input does not make clear four times what. Still, even among those wary about AI's future, some overestimate its capabilities and some take a more modest view.

Entry-level work is shrinking first, and the paths for people to develop skills are narrowing

The effects are showing up in employment. According to reports, the share of entry-level hires among Anthropic's technical staff fell from 27% to 9%. Reports of shrinking entry-level openings line up in the UK, India, and elsewhere as well.

basically junior developers aren't getting any jobs. The senior ones will eventually age out. We may actually be in trouble.── Gary Marcus (2026-10-15 12:49 UTC · YouTube "NYU’s Gary Marcus on Building AI We Can Trust" (Bloomberg Television) at 0:06:16, source)

Marcus warns that if newcomers cannot get jobs, there will be no one to succeed experienced workers when they retire. If AI takes away the first stage of learning the job, it will lead to a talent shortage several years from now. However, this hiring-ratio figure itself was not mentioned within the council.

As a counterexample, Huang points to radiology. Even though AI excels at diagnostic imaging, the number of radiologists has reportedly increased.

Computer vision is completely superhuman. Every radiology platform and package today is driven by AI. And yet, the number of radiologists grew.── Jensen Huang (2026-10-15 17:01 UTC · YouTube "AI Business Leadership Webinar | Integram & RBC Wealth Management - Private Banking Ft. Jim Harris" (Integram) at 0:18:28, source)

This is a claim that jobs will not necessarily decrease, but it is an example from a specialized profession and does not directly answer the problem of the entry point for newcomers.

Will regulation not move until a serious accident happens?

Marcus sees US regulation as unlikely to move unless the administration changes, though it might move if a serious accident occurs.

already, like in the US, 70% of the people want to see serious regulation.── Gary Marcus (2026-10-15 12:49 UTC · YouTube "NYU’s Gary Marcus on Building AI We Can Trust" (Bloomberg Television) at 0:15:08, source)

The figure that 70% of Americans want substantive regulation shows a gap between public opinion and politics. Marcus is concerned that advance measures are not progressing. [Council reasoning] Opinion within the council on regulation is also split, at 0.877. Andrew Ng and others stand on the minority side, holding that advance regulation at the development stage hinders innovation. This minority view, which gives weight to mechanisms for assigning responsibility at the point of application, argues that it is more realistic than advance review.

On the benefit side, cooperation among companies and medical progress through other technologies are cited. On the risk side, the severing of pathways for training newcomers, the lack of AI reliability, and excess investment with no visible returns are cited. There are three on each side, and which weighs more has not been decided.