A quantitative model of the race between AI-driven power concentration and societal reaction mechanisms,
calibrated to real-world data. Built to accompany the essay
The Wrong Conversation. Explore how awareness lag, political lag, and elite capture
interact — and whether any intervention prevents stable autocracy.
Data sources: METR 2025 · Epoch AI · McKinsey Nov 2025 · Goldman Sachs · Crunchbase 2025 ·
OpenSecrets lobbying data · UBS Wealth Report 2024
Calibration:METR Mar 2025: capability 2x/7moMcKinsey Nov 2025: 57% tasks automatableCrunchbase 2025: top 2 labs = 14% global VCUBS 2024: global wealth Gini ~0.88
METR 2025: 7mo observed; range 3–24mo
7 mo
Epoch AI: algo efficiency 3x/yr. Current automation ~15–35% of AI research.
0.55
Amodei/CNN 2025: 40% Claude usage = full automation. Sholto Douglas: near-full WC by 2027-28.
0.85
Skild AI, Figure, Tesla Optimus timelines. Ken Goldberg (UC Berkeley) is a credible skeptic.
0.60
2025
Stable autocracy threshold reached. The top-lab cluster controls >60% of effective economic power.
This is the self-reinforcing equilibrium described in the essay — no endogenous exit mechanism remains.
Labor irreversibility threshold. Labor's power share has fallen below 5%.
Collective bargaining and democratic pressure are no longer viable countervailing forces.
Top-lab power share
8%
Labor power share
55%
Wealth Gini
0.88
Crisis year
—
Power distribution
Top-lab cluster
Capital owners
Labor
Crisis threshold
Labor displacement
White collar
Blue collar
Capability gap (log scale)
Leader
2nd place
Three interacting mechanisms. Each is modeled as a differential equation feeding the others.
They don't add — they multiply. Elite capture doesn't just slow reaction; it degrades
the awareness-to-policy pipeline itself. The key output is net reaction force:
policy strength × (1 − elite capture).
Time before public recognises displacement as systemic. ~1–3yr historically.
How much enacted legislation actually constrains behavior. EU AI Act ~0.4; strong US antitrust ~0.7.
0.50
OpenAI: 7x in 1yr. Nvidia: 388% YoY. Big Tech: $400K/day in Congress by H1 2025.
3.0x
Max fraction of policy that lobbying can neutralise. Telecom/finance historical cases: 40–80%.
0.60
How fast people viscerally feel displacement as a crisis. Faster = earlier political pressure.
0.50
2025
✓ Crisis averted
⚠ Stable autocracy reached
⚡ Marginal — reaction barely holds
Live system state
Public awareness
0%
Policy strength
0%
Elite capture
0%
Net reaction force
0%
The race: concentration vs. reaction
Lab power
Labor share
Net reaction
Reaction anatomy
Awareness
Policy
Elite capture (erosion)
Toggle interventions and compare against the base case. Each is modeled from historical precedents and calibrated to observed rates. Early intervention (now, 2025) uses all structural changes simultaneously — the Norway model applied before the resource curse.
Redistribution floor as labor income collapses. Requires political will before power concentrates.
Floors labor share at 8%
Antitrust / compute caps
Regulatory limits on frontier compute concentration. Adds friction to R&D feedback loop.
Slows R&D loop 30%
AI wealth tax
Progressive tax on AI-generated capital returns, redistributed as public dividend. Norway SWF model.
–20% capital snowball
International governance
Treaty-based compute monitoring and multi-polar AI development mandates.
+2 effective competing labs
Early intervention (now)
All structural changes implemented in 2025 while power is still distributed. Highest-leverage option.
Full effect from year 0
Top-lab power share: base vs. interventions
No intervention
With selected
Crisis threshold
Select interventions above, then click Compare.
This essay's scenario
White collar yr 1–2. Blue collar yr 3–5. Fast R&D loop (7mo doubling).
Crisis: ~2030–2031
Amodei (public statements)
50% white collar in 1–5yr. Blue collar slower. Moderate R&D feedback.
Crisis: ~2033–2036
Davos consensus
Gradual adoption. Firms slow to act. Regulatory drag throughout.
Crisis: ~2039–2042
All three scenarios: top-lab power share
This essay
Amodei
Davos
Labor power share: all scenarios
Key finding: even the Davos consensus — the most optimistic mainstream view —
results in crisis before 2045 under current reaction parameters. The difference between
scenarios is how much time we have to intervene, not whether concentration happens.
How does the outcome change depending on when structural intervention begins? Each bar shows the projected crisis year if intervention starts in that year. Green = crisis averted within 20yr window. Amber = delayed. Red = crisis before 2035.
Crisis year by intervention start year
Which parameter matters most at each start year?
Awareness speed
Policy effectiveness
Political lag
Capture ceiling
Computing...
Which parameters drive concentration most? Each bar shows how many years the crisis shifts when that parameter is changed ±50% from its base value. Longer bar = more leverage over the outcome.
Years to crisis — sensitivity to each parameter
Key finding: R&D feedback strength and capability doubling time are the dominant levers —
consistent with the essay's core thesis. The reaction mechanisms (political lag, capture ceiling)
matter significantly, but only once displacement is already underway. Number of competing labs
has surprisingly low effect beyond ~3, because market-share dynamics consolidate regardless.
Projected sequence of events under the essay's displacement scenario. Events are model-derived and anchored to empirical data. Treat as a structural tendency, not a precise forecast.
2025–2026 (now) — Entry-level white-collar hiring collapses. Finance, law, consulting, software testing, content creation. Already underway: US job openings at lowest since 2021 (BLS Dec 2025), steepest declines in professional and financial services.
2026–2027 — Mid-level white-collar automation reaches ~60–70%. Professional-class unemployment spikes. Top-lab cluster begins absorbing displaced capital at scale. Gini crosses 0.93. Elite capture of regulatory agencies accelerating.
2027 — Intervention window closes under base-case parameters. After this point, meaningful structural reform requires simultaneously improving awareness speed, policy effectiveness, and structural lobbying limits — not any one of them alone.
2027–2028 — White-collar displacement near-complete. Labor's power share drops below 20%. Democratic political pressure theoretically possible but structurally weakening. Last viable window for antitrust structural remedies.
2028–2029 — Blue-collar displacement begins in earnest. Logistics, warehousing, manufacturing. Labor share below 10% — irreversibility threshold. The population that would need to organize to stop concentration has lost the economic leverage required to do so.
2029–2030 — Blue-collar displacement accelerates across construction, transport, physical services. Human labor no longer a meaningful economic input in most sectors. Power concentrates into top-lab + aligned capital.
2030–2031 — Crisis threshold. Top-lab cluster passes 60% of effective economic power. UBI, if not already structurally embedded with independent funding mechanisms, now requires the goodwill of entities that have captured the political system. The stable autocracy equilibrium.
2031+ — Self-reinforcing equilibrium. The lab cluster owns the AI, the capital, and the physical infrastructure. No endogenous redistribution mechanism. The Norway window — democratic structures installed before the resource curse — has closed.