BlueShip Research

In one line: Intelligence is getting cheap and permission is getting scarce, and I wrote fourteen dated, scoreable claims about it instead of describing the decade.

Abstract. On 30 July 2026 Ray Dalio described the coming decade for ninety minutes without attaching a probability or a date to any claim, on the stated ground that bubbles cannot be timed. The caution can be held while still committing to a number. As machine reasoning becomes abundant, the binding constraint migrates to what intelligence cannot manufacture for itself, and almost all of it is permissioned by somebody. The constraint is therefore a calendar rather than a capital budget, and the rent accrues to the holder of durable access.

1. Ninety minutes without a number

On the thirtieth of July, Ray Dalio sat down for an hour and a half and described the next decade. Debt service squeezing governments, an AI boom carrying the classic markings of a bubble, the share of business revenue going to workers falling, a world order going regional. Close to three million people have watched it.

He did not attach a single probability to anything, and he did not date a single claim. He says plainly that the future is deeply unknown and that even sophisticated investors struggle to time a bubble. Anyone who has tried it knows he is right. But a description is not a claim. You cannot be scored on one.

There was a single exception. His host reads him Jeremy Grantham’s view that the peak is very soon. Dalio answers in two words. He’s right. That is the only dated claim in ninety minutes, and he reaches it by agreeing with somebody else.

2. What I think the decade turns on

Here is the bet in one sentence. Intelligence is getting cheap and permission is getting scarce.

The cost of a unit of machine reasoning has collapsed and keeps collapsing, and the argument has moved on to who captures the value.

The binding constraint moves to what intelligence cannot manufacture for itself, and almost all of it is permissioned by somebody.

An agent can reason. It still needs an energized machine to run on, and a firm megawatt at a particular substation on a particular date is not something you can buy at any price. Eligible chips, customs clearance and an authorized settlement rail are the same story. Each is granted by a utility, a regulator, a commission or a charter, on a calendar that no amount of capital compresses.

Turbine order books stretched from 2028 to 2030 in the space of four months. The median wait from interconnection request to operation is around five years, against under two at the start of the century. Announced nuclear capacity for AI sounds enormous until you notice that almost none of it is under construction.

The rent accrues to whoever holds the durable access rather than whoever owns the best model. The question for the next five years is who is sitting on contracted power, licensed compute, strategic supply and chartered settlement when the abundance arrives.

3. The part that should bother you

A permissioned world is not a neutral one, and the hierarchy has a bottom. Countries that cannot obtain licensed compute will consume the capability without ever accumulating the capacity, a rental relationship dressed up as participation. Economies that import food will hike into supply shocks that the Federal Reserve is entitled to look through. Knowledge workers at entry level will lose the apprenticeship rung before the profession disappears.

Africa holds something like sixty per cent of the world’s best solar resource and attracts about two per cent of global renewable energy investment. That gap is a permission problem, the same one seen from underneath.

4. Writing it down

I have read a great many confident essays about this decade and watched almost none of their authors return to see how they did.

So I wrote fourteen of these claims down as binary questions, attached a probability to each, named in advance the published series that decides yes or no, and published them before any could resolve. Some will be wrong. You will be able to see which ones, on dates I chose before I knew the answer, against sources I do not control.

That is what the log is for, and it will still be there in 2030 whether it flatters me or not.

The fourteen forecasts, their resolution criteria and their running score are public in the Forecast Wing, marked against named series from third parties, win or lose.