Sunil Venkataram
Back to garden
•
🌰 seedling
TAM of Intelligence is Infinite
type/noteaiinvestingeconomics

TAM of Intelligence is Infinite

The framing

A traditional TAM calculation asks: how many buyers exist, how much would each pay, what’s the top-down ceiling? This framing works for categories with fixed scope — CRM seats, database licenses, cars per household.

Intelligence doesn’t fit. Cognitive work is an input to:

  • Every knowledge worker’s output
  • Every decision under uncertainty
  • Every interface between a human and a system
  • Every analysis that precedes a transaction
  • Every creative artifact
  • Every line of code
  • Every diagnosis, legal opinion, strategy memo, design choice

The naive upper bound is “the fraction of GDP currently spent on cognitive labor,” which is already in the tens of trillions globally. But that’s still bounded thinking. The unbounded version is that cheaper, better, faster cognition creates new economic activity that didn’t exist when cognition was expensive. See: Jevons Paradox applied to knowledge work.

What the Anthropic ramp suggests

From All-In E230 - Anthropic Mythos OpenClaw Where AI Value Accrues:

  • Anthropic’s revenue trajectory: $1B → $4B → $9B → $30B across five quarters
  • Tripled in the last four months alone
  • Added “a combined Databricks-plus-Palantir of annualized revenue in a single month”
  • Still supply-constrained — only 2,500 employees and ~1.5 gigawatts of compute

This isn’t a company capturing a fixed market. The limiting factor is how fast they can install compute, with demand stacked up behind the meter. That’s what an infinite-TAM market looks like from the supply side.

Why this matters for positioning

If the TAM framing is right, three things follow:

  1. Traditional market sizing exercises misprice the opportunity. Analysts estimating “the AI market” at $500B or $1T by 2030 are applying SaaS-era math to a category that doesn’t behave like SaaS.

  2. Supply-side bets matter more than demand-side bets. The winners are the companies that can build compute, secure power, hire researchers, and ship capacity — not the companies that pick the right vertical. See also: AI is power and compute constrained.

  3. Value creation will continue to outrun value capture. Most of the value from cheaper intelligence accrues to downstream users (cheaper professional services, faster drug discovery, more productive knowledge workers) — not to the companies selling the intelligence. Jerry Neumann’s AI Will Not Make You Rich thesis remains compatible with this framing: the TAM is infinite and most of the value leaks to consumers.

The honest caveats

  • “Infinite” is a framing, not a number. In practice the ceiling is whatever fraction of global economic activity can be price-discovered against cognitive work. That’s vast, not literally infinite.
  • The ramp is supply-constrained, not demand-confirmed. The $30B ARR could reflect genuine end-user demand OR it could reflect enterprises over-subscribing in case they need it later. The 2027-2028 churn numbers will tell.
  • Jevons Paradox only applies if the technology is a genuine substitute for existing cognitive work. For tasks where human judgment is legally or socially required, the substitution rate is capped regardless of model quality.
  • The Jevons framing has a serious counter-argument worth holding. See Jevons Paradox vs Cognitive Displacement - The Unresolved Tension — Sahaj Garg and others argue that the usual reabsorption pattern fails in this transition because all three labor categories (cognitive, emotional, physical) are being automated on overlapping timelines, removing the “next category” that historically absorbed displaced workers. The Jevons-positive view (this note) and the displacement-positive view should be held simultaneously as live hypotheses until the hiring data of 2027-2028 resolves them.
Backlinks (6)
Connected Notes