Most companies spend less on AI per worker than you pay for Netflix
1 Nov 2026 · News
A new essay says cheap intelligence will create a mass market, the way tea and cotton once did. The bills say the mass market hasn't shown up yet.
Venkatesh Rao argues that the real AI story is demand, and that ordinary people will want far more intelligence once it gets cheap. The spending data says that demand is still a luxury. Among U.S. businesses on Ramp that pay for AI, the median company spends $15.70 per employee per month, less than a $19.99 Netflix plan. The top 1% spends 609 times as much. We expect that picture to hold through March.
The case for abundance
Rao's essay, Cognitive Mercantilism, went up on his Contraptions newsletter on 25 October. It borrows from the historian Joyce Appleby. Before capitalism, mercantilists treated wealth as a fixed pile of land and bullion to hoard, and they saw ordinary people as labor, not customers. Rao says today's AI anxiety makes the same mistake with intelligence. It treats brains as scarce, distrusts the public and ignores demand.
His counterexample is the industrious revolution. Appleby describes ordinary Britons working more days a year, from about 250 to 300, partly by choice, so they could buy tea, sugar, tobacco and Indian cottons. Rao expects a version of that for thinking. When intelligence is cheap, paying attention pays off, so people think harder and want more. "There is no obvious upper bound on that demand," he writes. He even names a price. People may drop a streaming plan to afford a coding agent, since "both Claude Code and Netflix Standard cost around $20/month today."
His prices check out. Netflix's US plan page lists Standard at $19.99. Claude Pro, which includes Claude Code, costs $20 billed monthly on Anthropic's pricing page. History is on his side too. Food got cheap, and people found plenty to do besides farming.
What the bills show
Ramp, a corporate card company, tracks AI spending at more than 85,000 U.S. businesses in its AI Index. In September, 55.9% of them paid for any AI at all. Among the ones that paid, the median firm spent $15.70 per employee per month. Within the top 10%, the typical firm spent $813.43. Within the top 1%, it spent $9,567.53, which is 609 times the median. In August the gap was 581 times.
So most companies on Ramp spend less on AI per worker than one Netflix plan costs. About 44% spend nothing, and half of the rest spend under $15.70. Ramp says free tools and personal accounts don't show up in its data, so the real figures run higher. Free tools can't explain a 600-fold gap, though.
People paying out of their own pockets look much the same. In a16z's consumer AI report, 4.5% of eligible U.S. consumers in YipitData's panel paid for a personal ChatGPT, Gemini or Claude plan in August. That's up from 2.1% a year earlier, which is real growth from a small base. The median payer spends $25 a month, and a16z says that figure has barely moved.
Who buys the luxuries
One of Rao's mercantilist beliefs describes an economy where "the only real discretionary consumers were elites, who bought luxuries." That fits AI spending today. Apollo's chief economist Torsten Slok, citing Ramp data, says 80% of the top 1% of AI spenders are venture-backed, compared with 13% of the median spender. In Ramp's summer spending report, the median venture-backed firm spent $76.29 per employee a month. The median firm with neither venture nor private equity money spent $8.00, less than Netflix's $8.99 plan with ads.
At the far end sits Spawn, a ten-person game studio whose founder says it spent $3.1 million on tokens in a month. That works out to about $310,000 per person.
Part of Rao's story does show up, in usage. In Ramp's weekly token data, the price of OpenAI and Anthropic tokens fell 26.8% between the weeks ending 25 September and 23 October. Companies used 37.1% more tokens and spent only 8.9% more. Ramp says that sample isn't fully representative, but the direction is clear. Companies are buying a lot more intelligence without paying much more for it.
Then versus now
On 12 October we wrote that AI subscriptions are turning into meters, and that the top 1% of consumer payers made up 19.5% of consumer AI spend. On 24 October we reported that 1% of firms drive about 80% of company spending on OpenAI and Anthropic, and that a 25.4% token price cut came with 23% more spending. Since then, a similar price cut came with only 8.9% more spending. This time, cheaper tokens brought the labs much less new money. Rao's essay puts ordinary people's demand at the center of the AI story, and in the bills that is still the smallest part.
Our call
Rao has the direction right and the timing wrong. The industrious revolution he leans on took generations. So far the AI version is a small group of mostly venture-funded firms spending like aristocrats while the typical company spends about what a streaming plan costs. That makes the boom fragile. Slok's own warning is that a pullback in venture funding "could in turn weaken demand for compute."
In Ramp's March 2027 AI Index, the median firm paying for AI will still spend under $25 per employee per month, and the top 1% median will still be at least 300 times the median. If the median kept September's rise of 6.5% a month, it would reach about $21.50 in the February data that the March index reports. Ramp warns that its top 1% figure is volatile and gets revised, so our bar is half of today's gap. We're wrong if the median clears $25 or the gap drops below 300 times.
When the typical company spends more on AI per worker than a Netflix plan costs, Rao's mass market will have started to show up in the bills. That hasn't happened yet.