Last October Jensen Huang stood on a stage in D.C. and said “over the next five quarters, there’s half a trillion dollars” to fulfill.

Today, Jensen was on a stage in San Jose, and said “right now, through 2027, at least $1 trillion” of high confidence demand.

And then he said they will be short on supply, and he’s certain it (demand) will be much higher than that.

Remember, on the earnings call last month, Jensen said “compute equals revenue.”

He said every dollar of compute capacity added (the huge industry capex spending that continues to grow) is being monetized the moment it comes online.

This is the self-reinforcing loop that is fueling what Jensen calls the biggest industrial buildout in human history: compute generates output, output generates revenue, revenue funds more compute.

And today, he drilled down into the unlock for it all — it’s power.

He says revenue equals tokens per watt.  Tokens/output is only constrained by watts/power.

That’s the big theme of this AI revolution:  Abundance is only constrained by scarcity.

With that, the Nvidia monopoly is well secured, because they’re churning out new advanced chips on about an annual schedule — which are proving to be multiples more efficient than the prior generation (more output per watt).

So, the Nvidia chip cycle alone is a global economic driver.  And those efficiency gains feed directly into Elon Musk’s economic-growth-tsunami theory.

And Elon’s “limitless” economic output theory then becomes plausible if you believe global energy capacity can be expanded.

Until then power is the limiting factor.

It’s scarce.

And with that, if AI is creating a world of abundance, then we should invest in scarcity.

We just launched a new site at AIInnovationPortfolio.com where you can see how we’re thinking about this “invest in scarcity” theme, and how we’re positioned.