First post here. I'm Jason, a product manager writing from Pennsylvania coal country. I'm not an economist. I've been circling this idea for a while and used AI to organize it into something concrete enough to attack. I'm confident the problem is real, reasonably confident in the design, and genuinely unsure I've set the numbers right — which is why I'm posting it in this community. You built most of the parts I assembled, and I'd love for people to lean in and try to break it.
Five years ago, O'Keefe and colleagues published the Windfall Clause: firms commit in advance to sharing profits if AI ever makes them astronomically rich. It has mostly sat on the shelf since. My piece takes it off the shelf and attaches it to something that can start today. Here's the full proposal on Substack: https://jasonvalenti.substack.com/p/the-intelligence-dividend
The proposal in plain terms.
Layer 1, starts now. A tiny fee on the electricity of large AI datacenters: a few dollars per megawatt-hour, about half a cent per GPU-hour, only at facilities above 25MW of AI-chip load. Too small for any business model to notice. Big enough to build the fund, the public ledger, and the precedent. It attaches to the utility meter, so it catches open-weight and closed models alike and can't be dodged by shuffling APIs. Tenants churn; meters don't. There's a worked example in the piece using Stargate Abilene, where nearly every logo above the meter has changed in two years while the watts never moved.
Layer 2, signed now, pays later. The Windfall Clause with numbers attached: a graduated share of a firm's total profits above 0.25% of gross world product (about $275B a year, roughly triple the most profitable companies in history), scaling to 25% above 1% of GWP. Using total profits with civilization-scale thresholds deletes the "which profits count as AI profits" fight instead of litigating it forever.
Both feed an Alaska/Norway-style permanent fund paying a universal, equal dividend. The dividend does double duty as the anti-capture mechanism: everyone owns it, so everyone defends it.
Honest numbers, stated plainly in the piece: the first decade collects low billions and mails nobody anything. The check only exists in the world the lab founders keep predicting. If they're wrong, Layer 2 never fires and Layer 1 cost roughly nothing.
Where I'd most value attack:
Strongest surviving objections go into v1.1, with credit.
