Indices

Three measurements that didn’t exist. So I built them.

Is the price of compute actually falling, or is it just frontier hype?

$CPI—·

The price of a million tokens, weighted across four workload tiers: commodity, frontier, reasoning, and long-context. Volume-weighted across 2,000+ models. January 2025 = 100.

Methodology & data →

Is the AI market actually growing, or is the spend just chasing flat usage?

$AIU—·

A composite of three inputs: token throughput (60%), inferred spend (30%), and energy use (10%), read as a single number. February 2025 = 100.

Methodology & data →

When AI is cheaper than a person, how long does it take to actually replace one?

$LDI—·

The substitution rate between human and AI cost across five federal workloads. Human cost from BLS wage data; AI cost from the $CPI basket; substitution inferred from federal procurement records.

Methodology & data →

Tools

AI Services Price Reasonableness Worksheet → A fill-in procurement determination sheet for federal AI buyers, built on the $CPI basket. The benchmark rates predate the current data — treat them as a starting point, not a current quote. LLM Cost Calculator → Price a workload across models using the live Compute CPI rates. LDI Workload Calculator → Compute the human-vs-AI cost differential and substitution rate for a workload of your own. SNAP Eligibility — LDI Deep Dive → A worked example of the Labor Displacement Index applied end-to-end to one federal workload. Market Intelligence → Statistics that reprice market assumptions — quality-adjusted price, cognitive arbitrage, and tier analysis across the model market.