The Capture Gap
AI is breaking the old link between capability, work and prosperity. Economic development needs a new theory of value retained.
A C$13-billion data center that creates 300 permanent jobs is no longer an edge case; it is the signature project of this investment cycle. This report explains why the jobs metric is failing at the negotiating table, and lays out what should replace it: the capture gap, a project ledger, and four families of enforceable claims.

Get the full report
Enter your details and the PDF downloads straight away.

C$43M
of capital per permanent job in Meta's C$13-billion Alberta data center, announced July 2026
US$0.5T
global data-center investment in 2024, with electricity demand set to more than double by 2030 (IEA)
×20
the gap in employment density between a hyperscale data center and a battery plant, on the same headline capital
For 70 years the profession could treat jobs as the measure of a deal, because capability, employment and ownership arrived bundled together. AI is pulling that bundle apart. The places that prosper will be the ones that learn quickly, bargain competently, and keep a durable claim on what their assets make possible.
Fifteen sections, from diagnosis to operating model.
The report opens at the negotiating table, traces the theory the profession inherited and what AI does to it, then builds the replacement piece by piece: the ledger, the claims, the institution. Six of the fifteen sections appear below.
The question that stopped working
Why the direct-jobs number can no longer carry a deal, and what the drift into “ecosystem effects” is really telling you.
What AI changes, and what it does not
Enough cognitive work is being commoditized to move the returns elsewhere, and no claim of superintelligence is required.
Scarcity after cognition
Power, context, jurisdiction, organizational capability, capital, and clusters: the assets that stay scarce when cognition gets cheap.
Measuring the capture gap
A four-column ledger (value retained, public cost, distribution, duration) built to include everything the jobs count excludes.
The architecture of claims
Fiscal, ownership, contractual and institutional claims, and the difference between a claim and a promise.
The institution this requires
Attraction, valuation, negotiation, diffusion, and stewardship, plus why term sheets beat tax holidays.
The framework, tested on the current cycle.
Three projects from 2025–26 run through the framework in full:
Sturgeon County, Alberta
Meta's C$13 billion data center
Meta's July 2026 announcement set out one gigawatt, 1,750 acres, and 300-plus permanent jobs. The announced terms are more interesting than the jobs count, and the report shows which ledger questions are still open.
Abilene, Texas
The Stargate campus
Six thousand construction workers at peak, and permanent-jobs estimates that vary five-fold across published accounts of the same site. The metric itself has stopped doing the work.
Windsor, Ontario
The NextStar battery plant
Twenty times the employment density of the Alberta campus, and an ownership stake that changed hands anyway. Even jobs-rich projects argue for a ledger that counts more than jobs.
What you'll walk away knowing.
The report ends with the redesigned development institution: five roles, a new operating rhythm, and performance measured in claims secured rather than jobs announced.
Written for the leaders of economic development organizations and the officials who negotiate investment projects.
- Why the jobs metric is failing structurally, and what to say in the room when it does
- A capture ledger you can put in front of a minister before the incentive conversation starts
- How to read a data-center proposal: what to charge for, what to trade, and when to decline
- Live cases from the current cycle (Alberta, Abilene, Windsor) run through the framework
The full report, from the capture ledger to the four families of claims.
Leave your name and email and the report is yours. We'll keep you posted on what Doyen Collective publishes next, and you can unsubscribe anytime.