BENCHMARKS · GOVERNMENT POLICY
Policy Economics
Governments don't just regulate the labs — they move the unit economics. Pick a scenario — a federal procurement EO, export controls, residency mandates, energy rules — and watch break-even shift on the same hardware.
Policy scenario
Agencies directed to phase out a named provider's models — the Anthropic scenario. Targeted provider loses anchor demand; rivals absorb it but carry compliance overhead.
Model
Accelerator
Region
Break-even · before32%8×B300 · United States
Break-even · after33%+1 pts
Profit/mo · before$16,790at 50% util, off-peak price
Profit/mo · after$4,258at 38% util, after policy
DeepSeek V4 Pro on 8×B300 in United States — before vs after federal procurement eo
| Metric | Before | After policy | Δ |
|---|---|---|---|
| Node cost · $/hr | $40.00 | $40.00 | $0.00 |
| Effective price · $/1M output tok | $1.00 | $0.97 | -$0.03 |
| Throughput · output tok/s | 35,000 | 35,000 | — |
| Break-even utilization | 31.7% | 32.7% | +1.0 pts |
| Monthly profit · GPU rent only | $16,790 | $4,258 | -$12,532 |
Same node, same model, same token prices — only the policy levers move. Demand shift scales achievable utilization; compliance is taken off the top of revenue; GPU premium scales rent and (implicitly) build cost. Illustrative magnitudes, not forecasts.