AI Cost Collapse

We are living through one of the most extraordinary economic phenomena in the history of technology — and most people are watching it without the framework to understand what it means.

The cost of AI intelligence per token has dropped more than 99% in under four years. That is not a typo. What cost $20 per million tokens at the launch of GPT-3 in 2020 now costs fractions of a penny. GPT-4o mini sits at $0.15 per million input tokens — 200 times cheaper than original GPT-4, which launched at $30 per million just two years earlier. The newest frontier models continue the freefall: GPT-5 nano is available today at $0.05 per million tokens. Claude Haiku 4.5 processes a million tokens for under $1. Google's Gemini 2.5 Flash runs at $0.075 per million.

Wright's Law: The Engine Behind the Collapse

In 1936, Theodore Wright discovered something counterintuitive while studying aircraft manufacturing costs. Every time the cumulative number of units produced doubled, costs fell by a predictable, constant percentage — regardless of time elapsed. This principle — Wright's Law — has proven itself across solar panels, lithium batteries, semiconductor chips, and now, unmistakably, AI inference.

The Solar Analogy: Faster Than Anyone Expected

From 2010 to 2024, the cost of solar electricity fell roughly 90% — a dramatic decline that analysts consistently underestimated. Now compare that to AI token pricing. Solar took 14 years to fall 90%. AI tokens achieved a comparable collapse in under 4 years — and the steepness of the AI decline makes solar look gradual by comparison.

Stanford's 2025 AI Index confirmed the trajectory quantitatively: inference costs for GPT-3.5-level performance dropped over 280-fold between 2022 and 2024 alone.

Jevons' Paradox: Why Cheaper AI Means Exponentially More AI

In 1865, William Stanley Jevons demonstrated that more efficient engines didn't reduce coal consumption — they tripled it by 1900. This is Jevons' Paradox: efficiency improvements in the use of a resource do not reduce total consumption. They increase it dramatically by unlocking demand that was previously economically impossible.

Per-token prices fell more than 1,000-fold in three years. Enterprise AI spending simultaneously surged 320% in 2025 alone. Microsoft CEO Satya Nadella acknowledged this directly: "Jevons Paradox strikes again. As AI gets more efficient and accessible, we will see its use skyrocket, turning it into a commodity we just can't get enough of."

AI Agents: Where Jevons' Paradox Becomes an Avalanche

An AI agent is not a chatbot. It is an autonomous software system that can independently plan multi-step tasks, make decisions, call external tools, and execute complex workflows without human intervention at each step. When you deploy an AI agent to research a client file, draft correspondence, cross-reference case law, and flag inconsistencies — you consume hundreds of thousands of tokens, automatically, every single time it runs.

Gartner projects that 40% of enterprise applications will feature task-specific AI agents by end of 2026 — up from less than 5% in 2025.

What This Means for the Legal Profession

The practice that deploys agentic AI for client intake, financial analysis, document drafting, compliance monitoring, and research doesn't compete with larger firms on volume by working more hours. It competes on architecture. The token cost to run a comprehensive agent stack across an active family law docket of 50 cases is, at current pricing, measurable in dollars per month. Not hundreds. Not thousands. Dollars.

The price of intelligence is collapsing. The demand for intelligence is about to detonate. The window to build on that foundation — before it becomes table stakes — is open right now.

Build accordingly.