For most of history, physical work was muscle. Then steam, combustion, and electric motors took it. Human muscle became a rounding error in the global energy budget. Cognitive work is on the same curve, compressed in time.

When the price of a scarce input collapses, supply floods and demand expands. Cheaper engines did not reduce coal use. They multiplied it. Cheaper cognition will not reduce thinking. It will multiply it. Most problems on Earth currently go unthought about because thinking is expensive.

A token is not a fixed unit of text. It is the output of a tokenizer. The same content can produce meaningfully different token counts depending on language, model version, and even quiet tokenizer updates. Non-English text routinely consumes more tokens. Output tokens are priced much higher than input tokens. Reasoning or “thinking” tokens are usually billed as output as well.

These mechanics create systematic opacity. Listed per-million-token prices are unreliable proxies for real cost. Cheaper models can end up more expensive overall because they consume more tokens to reach the same result. Subscription plans hide limits through silent model downgrades and non-comparable quotas. The practical consequence is that teams cannot budget accurately from published rate cards alone.

The dominant narrative says AI will cause mass unemployment. Three years of real data say otherwise.

Over 90% of U.S. firms that adopted AI reported zero change in headcount. Multiple surveys from the Atlanta Fed, Census Bureau, and other sources show the same pattern: almost no net job loss. What is happening instead is task reallocation inside existing roles. Recruiters do less screening and more judgment work. Lawyers review AI output rather than draft from scratch. Finance teams interpret rather than assemble data.

A small group of companies is currently generating enormous, non-recurring revenue by selling training data, expert judgment, simulated environments, evaluations, and physical capture to frontier AI labs. Some are closing tens of millions in purchase orders per month with teams of fewer than a dozen people. The cash is real. The contracts are not forever.

What is actually being sold falls into six categories: hours of labeling (already dying), high-end expert judgment, simulated worlds for agents to practice in, benchmarks and red-teaming (the referees), physical sensor data from real workers, and licensed rights to existing corpora. Almost everything arrives as a one-time purchase order against a deliverable. Nothing renews by default.

Fifteen years after Marc Andreessen declared that "software is eating the world," we are entering the next phase: intelligence itself is becoming abundant and nearly free.

Just as information became abundant through the internet, expertise and judgment — the scarce resource that was locked inside trained professionals — is now descending the cost curve at extraordinary speed. This descent is happening on two compounding curves at once: cheaper hardware and more efficient models. The result is that high-quality reasoning, once expensive and rationed, is rapidly becoming a commodity.

The dominant conversation in AI still revolves around which company has the smartest frontier model. However, the real economic transition underway is from maximizing raw intelligence to maximizing intelligence per dollar.

Enterprises are rapidly discovering that most workloads do not require the most expensive frontier models. As inference costs continue collapsing at an extraordinary rate, organizations are shifting from asking "Does AI work?" to "Does it work economically at scale?" This change favors cheaper, "good enough" models (including open-weight ones) for the vast majority of tasks, while reserving frontier intelligence only for high-value problems where additional capability creates outsized returns.

A new class of economic actor has arrived: autonomous agents that hold bank accounts, sign contracts, hire employees, and incur real financial obligations with no humans in the decision loop.

Traditional credit analysis collapses. The five classic pillars — management team, board oversight, audited statements, track record, and legal accountability — all assume human institutions. Remove the humans and the entire framework stops applying.

Frontier AI labs are not competing on the same playing field. Each has a unique identity that creates its own structural bottleneck.

OpenAI serves over 900 million monthly active users, mostly consumers. This produces a severe usage-mix problem: high volume of low-value prompts drives enormous inference costs with low revenue per token.