Anthropic's IPO prospectus has given us the first hard look at the economics of selling AI. Set against how enterprises are actually buying it, four things are coming into focus:
AI is rapidly deflating at the unit level. The cost of a given level of performance has been falling 13-fold a year, with no end in sight.
The AI "industry" is still highly concentrated among the frontier labs, the hyperscalers and the chip makers. It has yet to spread into the rest of the economy.
Adoption across the Fortune 493 is still muted. These are, for the most part, industrial-era businesses still running experiments.
The canaries in the coal mine are starting to sing. Wall Street and tech services, the two industries adopting AI fastest, are showing early signals in headcount and pricing.
Let's take them one at a time, because together they paint a picture.
The Deflation: Epoch AI outlines that since 2023, the cost of a given level of AI performance has fallen 47% per quarter. That’s six times the pace of Moore Law, and the fastest decline ever recorded for a transformational technology. In January 2025, a graduate-level science answer cost 30 cents. Eighteen months later it cost four-hundredths of a cent. The equivalent is a $50,000 Mercedes marked down to $69.
The Concentration. A surprise in Anthropic's S-1 is how few places the money comes from. Two unnamed customers (rumored to be Meta and Alibaba) accounted for 24% of 2025 revenue, and another 47% flows through the Amazon and Google marketplaces. The money going out is just as concentrated: $518 billion in compute commitments, including $161.2 billion to Broadcom, $111.1 billion to Google and $110 billion to Amazon. Roughly 80% of that is binding and non-cancelable. Revenue can deflate, but those obligations cannot.
Muted Absorption: The Fortune 493 efforts with AI are not showing up in revenue or margin metrics. In short, the AI market is currently great for its sellers but not its buyers. While corporate leaders aggressively recruit talent to orchestrate agents across their operations, productivity gains aren’t widely apparent. There’s a collective assumption this will change, but for now that’s belief without much evidence.
The Canaries: That said, two Fortune 493 sectors are starting to move: Wall Street and tech services. Both sit at the vanguard of technology adoption, and AI is beginning to reshape their labor and pricing models.
Labor is shrinking: Over the past 15 months, the US financial and tech sectors have shed a combined 246,000 jobs, while all other industries added 812,000. In both sectors, revenue per head is now the metric CEOs watch most closely, and leading players are now posting gains of 10% or more.
Tech services pricing power is starting to erode: Last week, Accenture cut its revenue growth guidance to 3–4%, said pricing fell across much of its business last quarter and that it plans to hire fewer people next year than it did this year. (Translation: “We’ll sell more units, but at lower cost with fewer people.”) In India, the Nifty IT index is down 27% year to date, and the September quarter is shaping up as the sector's weakest in years. Kotak expects TCS to grow just 0.5% and Infosys 1.1% in what is normally their strongest quarter, and Jefferies sees Infosys cutting its full-year growth guidance to as low as 0.5%. Kotak's explanation: "AI deflation."
So what to make of all this? First, if you feel behind on AI, take a deep breath. The transition across the broad economy will take years. We obsess over how fast the machine is improving and pay too little attention to how slowly industry is absorbing it. Second, AI deflation is good news for everyone outside tech. Jevons paradox is alive and well: as the cost of technology work falls, companies will take on projects they could never have justified before, and absorption will speed up. The sellers have had their run; the buyers' turn is next, and the price of admission is falling.
The week's most important AI news, curated for you:
