We talked yesterday about Jensen Huang's presentation showing the production of software going vertical over the past few months — the "$9 trillion of productivity from $3 trillion of salaries."
And we ended by looking at the San Francisco Fed's report on total factor productivity, which decelerated in the first quarter, to 0.60%.
Let's talk about that gap.
Jensen focused on AI usage. For the past year, "token consumption" has been celebrated in Silicon Valley. Big companies (particularly big tech) took a clear strategic position: press the gas pedal on AI to the floor (i.e. burn tokens).
They turned usage into a scoreboard. Companies like Uber, Meta and Amazon ranked engineering teams by token consumption (the more, the more celebrated).
What's the problem? A high token count doesn't tell you a lot got done. It tells you a lot got used.
When you reward people for consumption, the meter just runs. But you don't necessarily solve problems, ship products, reduce costs, or create value.
And the tell, at this point, is in the recent behaviors of these same companies — they are cutting back on, or setting limits, on usage.
The Fed's Q1 number on total factor productivity reveals that story (it decelerated, not accelerated). It doesn't confirm a productivity boom, at least yet.
But it's early. We've seen the investment. We've seen hesitation, at the enterprise level. Now we've seen aggressive adoption and experimentation. We're seeing some waste. Now maybe some refinement, discipline and productivity gains.
With all of the above in mind, we get the May jobs report on Friday. Is the AI building phase increasing labor demand, or is AI visibly replacing labor?
Is AI proving "useful" as Jensen positioned it? Or is it just getting used a lot?
