What Happens as AI Scales? Follow the Bottleneck
Most AI forecasts assume tech companies move in a straight line—either triggering a smooth boom or wiping out office jobs overnight. To test what happens when the real world pushes back, we placed nine competing institutions in late 2026—employers, electric utilities, banks, insurers, startups, workers, universities, the Fed, and governments—and let each act in its own self-interest every 6 months through late 2030.
The economy only moves as fast as its tightest constraint, and solving one bottleneck directly triggers the next: Electricity blocks AI first (2026–27) → rush building creates a +3.2 GW power glut and cheap cloud prices (2028–29) → cheap computing funds 94%-accurate multi-AI checks that replace exhausted human sign-offs (Early 2030) → firms cut middle managers while 24/7 AI checks burn up all the spare power, plunging the grid back into shortage (Late 2030).
Meet the 9 competing groups acting every 6 months ▾
1. The Chain Reaction in Four Acts (2026 → 2030)
2-minute overview of how each phase broke the nextPower & Errors Protect Office Jobs — Except Juniors
AI cannot immediately replace most office staff for two reasons: datacenters cannot get enough power from backlogged grids, and AI still botches roughly 1 in 3 multi-step tasks (68% accuracy). Because insurers require human review, big firms keep experienced staff on payroll—and protect margins by freezing graduate hiring while rushing to order power plants.
Rush Builds Flip Power Into a +3.2 GW Glut
All the solar, battery, and gas projects ordered in 2026–27 switch on together just as leaner open-source AI starts running on ordinary laptops. The power shortage flips into a +3.2 GW surplus (~3 nuclear plants). Half-empty datacenters default, cloud prices drop 30%, and burned lenders halt new power orders.
Dirt-Cheap Computing Ends "Human Sign-Offs"
Three years of frozen junior hiring leave senior staff buried in review queues—so they secretly use personal bots to rubber-stamp their own signatures. Meanwhile, cheap surplus power lets labs cross-check work across multiple AIs (94% accuracy). When audits expose fake human reviews in early 2030, insurers flip coverage to machine checks.
Manager Cuts Deepen & 24/7 AI Eats the Power Glut
With insurers backing automated checks, employers cut middle-management layers and startups plus overseas hubs take 29.4% of white-collar services—pushing grad hiring down 49% and workers' income share below half (49.3%). Running nonstop AI checks burns up all spare power, plunging the grid back into a –2.7 GW shortage.
1 GW ≈ 1 nuclear plant. Rush 2026 builds caused a 2029 glut (+3.2 GW), halting new builds before 24/7 AI checks ate the slack.
Once multi-AI cross-checking passed 94% accuracy in early 2030, insurers dropped human sign-offs and challengers gained share.
| Measure | Late 2026 | Early 2029 | Late 2030 | Plain-English Reason |
|---|---|---|---|---|
| AI Electricity Balance | –1.4 GW (short) | +3.2 GW (glut) | –2.7 GW (short) | 2026 rush created a 2029 glut; nonstop AI checks ate it |
| AI Multi-Step Accuracy | 68.0% | 88.0% | 95.8% | Cheap surplus compute funded multi-AI cross-checking |
| Routine Office Work Done by AI | 10.0% | 41.5% | 58.2% | Accelerated once insurers dropped human sign-offs in 2030 |
| Startups & Overseas Share | 4.0% | 19.6% | 29.4% | Lean challengers undercut slower traditional companies |
| Entry-Level Hiring Pace | 100% | 65.5% | 51.0% | Cut nearly in half as junior & audit tracks automated |
| Workers' Share of US Income | 56.5% | 52.4% | 49.3% | Middle-manager cuts pushed total wage share below 50% |