Hi Traders,

In May, a San Francisco outfit called Higgsfield AI put out a 95-minute feature.

It was a team of 15, took 14 days, and cost $500,000 with $400,000 going to compute.

Three months later, they did it again.

The Cully Hill Boys premiered in New York on August 5th. 110 minutes long, 28 person crew, four weeks long, and a cost of $2 million, with roughly half of that going to compute.

The budget went up four times.

I know. Hans, that's a media story. Stay with me, because the ratio underneath it is the whole thing.

Seventy Dollars a Second, Then Twice That

Hell Grind ran 5,700 finished seconds on $400,000 of compute. That's about $70 of silicon and electricity for every second on screen.

Cully Hill Boys ran 6,600 finished seconds on roughly $1 million. Call it $150 a second.

The compute bill per second of finished film more than doubled in 90 days, and it happened while published rates for the video models themselves were sliding toward pennies.

A real production shoots five or 10 takes because crew hours and daylight are expensive. When the next take costs a nickel, you don't shoot 10. You generate 200 and let a model pick.

Published rates for the top models top out around 70 cents a second. At $150 a second of finished film, they burned north of 200 seconds of generation for every one that made the cut.

You don't have to take my word for the churn. Higgsfield open-sourced the entire production: roughly 1,000 assets, 100 generated locations, and a production log with 137 entries tracking what changed and what got thrown away.

That's Jevons paradox.

Jevons noticed in 1865 that better steam engines didn't cut Britain's coal use, they multiplied it, because cheap coal made coal worth burning everywhere.

What collapsed is the cost of the failed attempt. And failed attempts are where the compute burns.

Every Viewer Becomes a Bill

A rendered film gets generated once and watched a billion times. The compute is a fixed cost spread across the whole audience.

Interactive film flips that. Google's Genie 3 builds navigable worlds live at 720p and 24 frames a second, and Project Genie opened in January to US subscribers paying $250 a month, capped at 60 seconds a session.

Sixty seconds for $250 a month tells you what live generation still costs. Every viewer who wants their own cut becomes their own compute event.

Which is why I keep coming back to the grid. Goldman Sachs has US data center power demand running from 31 gigawatts last year to 41 this year to 66 in 2027, taking data centers from four percent of peak summer load to eight and a half in 24 months.

Cheap seconds, expensive electrons. That's the trade.

In Turbo Income we're selling premium against the names on the receiving end of that demand.

Members get the full list plus the spread structures we're using to collect while the market catches up. Tap this link to join.

Time to get strategically greedy.

Here for a good time AND a long time,

Hans