The Promo Price of Artificial Intelligence

Earlier this year, an IT consulting firm opened a renewal notice from Cursor, the AI coding tool its engineers had come to depend on. The firm had paid about $200,000 for 800 licenses. For the same usage, the renewal quote was roughly $1.5 million — a seven-and-a-half-fold increase for the identical product. The firm negotiated hard and got it down to $250,000. Sanofi, the pharmaceutical giant, and Druva, the security vendor, both reported renewal quotes around five times what they'd paid the year before, according to reporting by The Information. These weren't rate adjustments. They were the sound of a subsidy ending.

If you've been running a business for more than a decade, you've heard this sound before. You heard it in 2022, standing on a curb, staring at your phone, wondering when exactly an Uber to the airport started costing $70. You swear the ride used to be $12. What happened in between wasn't gas prices or driver pay. It was a business model completing its arc: subsidize the habit until the alternatives are gone, then hand the company to shareholders who are owed the difference.

The AI industry is running that same arc right now — at a hundred times the scale. The only questions left are how large the difference is, and how much of your operation will be built on the promotional price by the time you're asked to cover it.

I. The Precedent: How the $5 Ride Actually Worked

Start with what the cheap Uber ride actually was, because most people never itemized it. It was not a technology dividend. It was not efficiency. It was venture capital, transferred directly into your pocket, one ride at a time. Uber lost more than $30 billion in roughly the five years after its finances became public — and that's on top of years of private losses before that. Lyft ran the same strategy. Together they executed what one analyst memorably called an enormous, investor-fueled subsidy of America's ride-hailing habit.

The word "habit" is doing the important work in that sentence. The losses weren't waste — they were the product. The point of the $5 ride was never the $5 ride. It was what the $5 ride did to you: it rewired your defaults. You stopped checking bus schedules. You stopped keeping a car in the city. You built your commute, your nights out, your airport logistics around the assumption that a car appears in four minutes and costs less than parking. Cities rezoned around it. An entire generation entered adulthood never having memorized a taxi dispatch number. The subsidy wasn't buying rides. It was buying the deletion of your alternatives.

Then came the IPOs — Lyft in March 2019, Uber that May — and with them, a new owner: the public market, which has exactly one demand. In a 2022 memo to employees, Uber CEO Dara Khosrowshahi said the quiet part with unusual clarity: the era of growth at all costs was over, and the company had to make its unit economics work. Translation followed swiftly. Average Uber prices rose 92 percent between 2018 and 2021, according to Rakuten data; a U.S. Senate Banking Committee investigation put the increase at roughly 83 percent between 2018 and 2022. The companies' take — their share of each fare — climbed to between 43 and 49.5 percent, per a 2026 Consumer Reports investigation that also found the same ride, requested at the same moment, could be priced 160 percent apart for different riders, thanks to algorithmic pricing systems that replaced transparent per-mile rates a decade ago.

And it worked. In February 2024, Uber announced its first annual profit in fifteen years of existence — and, in the same breath, a $7 billion stock buyback. By 2025 its annual profits had reached nearly $8 billion. The machine did exactly what it was built to do: subsidize until the alternatives died, then harvest. Riders didn't vote for the second decade of Uber. They just couldn't leave.

II. The Current Subsidy, Itemized

Now hold that template up against AI, and check the fit line by line.

Is there a subsidy? Leaked audited financials — obtained by the writer Ed Zitron and verified by the Financial Times — show OpenAI generated $13.1 billion in revenue in 2025 while posting an operating loss of roughly $21 billion. Read that again: the operating loss was larger than the revenue. In 2024, the company lost about $5 billion on $3.7 billion in revenue. The trajectory of the ratio is improving — OpenAI spent $1.60 for every dollar it earned in 2025, down from $2.37 the year before — but the absolute burn is accelerating, not shrinking. ChatGPT reportedly serves more than 900 million weekly users, of whom only about 50 million pay. That is the free-ride ratio of a company still deep in market-share mode: roughly eighteen riders in the back seat for every one who's covering gas.

And that's just one company's ledger. The deeper subsidy sits a layer down, in infrastructure. Microsoft, Amazon, Alphabet, Meta, and Oracle have collectively committed somewhere between $660 and $690 billion in capital expenditure for 2026 alone — nearly double 2025 levels — the overwhelming majority aimed at AI compute. McKinsey projects the global buildout could require $6.7 trillion by 2030. Meanwhile, the combined revenues of the AI companies this infrastructure serves remain a small fraction of what's being spent on their behalf. That gap — hundreds of billions of dollars a year between what the machine costs and what customers currently pay — is the subsidy. It is the largest below-cost pricing operation ever run, and every prompt you send is a beneficiary of it.

Is it buying habit deletion, the way cheap rides did? Ask any operator honestly. Businesses from five-person shops to the Fortune 500 have spent the last three years wiring AI into quoting, scheduling, support, code, content, and analysis. Junior roles have gone unfilled because the tool covers the work. Workflows have been rebuilt around the assumption that intelligence is effectively free and always available. That's not adoption. That's the same rewiring of defaults the $5 ride performed — except this time the deleted alternative isn't the bus. It's headcount, institutional knowledge, and in some cases the muscle memory of doing the work at all.

III. The IPO Is the Trigger, and It's Loaded

Here is where the timeline stops being historical analogy and becomes current events. Uber's price escalation didn't begin when the technology matured. It began when the ownership changed — when patient venture capital handed the company to impatient public markets. That handoff, for AI, is happening right now.

OpenAI has confidentially filed its S-1 with the SEC, with Goldman Sachs and Morgan Stanley attached and a valuation discussed near $1 trillion. Anthropic filed its own draft S-1 in June 2026. The leaked OpenAI financials surfaced during what was supposed to be a controlled pre-IPO quiet period — which is exactly why they matter. A company can lose $21 billion a year as a private darling with sovereign wealth funds writing checks. A public company at a trillion-dollar valuation cannot. It must show a credible, quarterly-audited path from $1.60-in-per-dollar-out to something a pension fund can hold. Every mechanism that closes that gap — price, packaging, usage terms, take rate — will be under permanent, earnings-call pressure to move in one direction.

Khosrowshahi's 2022 memo — we have to make sure our unit economics work — is not a historical document. It is a template. Some version of it will be written inside every public AI company within eighteen months of listing, because the structure of public ownership requires it. The only genuinely open question is the one you'd actually want answered: what does extraction look like when it arrives? Because here, the skeptics have a real point — and it deserves a fair hearing.

IV. The Strongest Objection: Prices Are Falling, Not Rising

The obvious rebuttal to everything above is that AI, unlike Uber, is getting dramatically cheaper. Per-token prices for frontier-class models have collapsed since 2023 — in some cases by more than 90 percent. Competition among labs is ferocious in a way the Uber-Lyft duopoly never was; there are credible reports of an impending token price war. And open-weight models offer an escape hatch that ride-hailing never had — nobody could download a free Uber and run it in their garage. If this is a subsidy trap, why does the sticker price keep going down?

Concede the facts: all true. The per-unit price of intelligence is falling and will likely keep falling. But the conclusion doesn't follow, because it confuses the price of the unit with the size of the bill — and the bill is what businesses pay. This is precisely the mechanism the Cursor episode exposed. In June 2025, Cursor moved from flat-rate plans to usage-based billing pegged to underlying compute costs. The rollout was botched badly enough to require a public apology and refunds that July. But the structural revelation was bigger than the PR failure: users discovered that agentic workflows — AI that works autonomously, in long contexts, across many steps — consume tokens at a rate that made effective costs for some workflows twenty times higher than the old flat rate, even as per-token prices fell. One industry observer put the paradox in a sentence: per-token prices keep falling, but autonomous agents burn so many tokens that enterprise bills keep climbing.

This is the shape of extraction in AI, and it's more elegant than Uber's, because it doesn't require raising prices at all. It requires expanding consumption. Every capability leap — longer contexts, multi-step agents, always-on background work — multiplies token burn faster than token prices decline. The industry doesn't need to charge you more per unit. It needs you to need a thousand times more units, which is exactly where the entire product roadmap points. Add the renewal lever (the Cursor quotes, the Sanofi and Druva quintuplings), the tier lever (flat plans quietly restructured into metered credits, "unlimited" redefined by rate limits — a pattern that has already touched nearly every major AI tool, Claude included), and the lock-in lever — your prompts tuned, your agents built, your team trained, your data threaded through one vendor's stack — and you get Uber's second decade without a single headline fare hike. The unit gets cheaper forever. The invoice never does.

What the objection does correctly predict is a ceiling. Open-weight models and genuine lab competition mean AI extraction will likely be constrained in a way ride-share extraction wasn't — the escape hatches are real, for businesses disciplined enough to keep them open. That's not a reason to dismiss the thesis. It's the reason the thesis is actionable.

V. How Big a Problem Is This, Actually?

Size it honestly, because "big" is not an analysis. The exposure splits three ways.

For consumers, modest. The free tiers will survive in some form because 900 million users are the moat, and ads — the oldest subsidy-recovery mechanism on the internet — can carry much of that load. Your personal chatbot is not the target.

For enterprises, significant but survivable. Big companies have procurement teams, multi-vendor leverage, and the Sanofi option: threaten to walk, and mean it. They'll absorb 2x and negotiate away the rest. Their AI line item becomes another cloud bill — painful, managed, budgeted.

The acute exposure is the middle: the millions of small and mid-sized businesses that have rebuilt their operations on subsidized intelligence with no procurement team, no negotiating leverage, and no plan B. The contractor whose quoting runs through an AI tool. The five-person agency producing the output of fifteen. The SMB software vendor whose margins exist only because inference is priced below cost. These businesses did the rational thing — they adopted early and restructured deeply — and in doing so they concentrated a risk nobody itemized: their cost structure now contains a subsidy they don't control, on a countdown they can't see, held by companies whose future shareholders will be paid precisely by ending it. When the gap between $690 billion in annual buildout and the revenue currently covering it gets closed — and public markets exist to close it — the money comes from the customers who can't leave. It always does.

How big? Run the Uber math forward. Ride-share prices rose 83 to 92 percent in roughly four years once profitability became mandatory, in a market with a visible competitor. An SMB spending $2,000 a month on AI tooling today, with workflows too deep to unwind, should be underwriting $4,000 to $5,000 within a few years of the IPO wave — through some blend of consumption growth, tier restructuring, and renewal repricing — and should treat anything better as upside. For a business running 15 percent margins, an unplanned doubling of a core operating line isn't an annoyance. For some, it's the margin.

VI. What You Do With This

The lesson of the $5 ride was never "don't take the ride." The subsidized decade of Uber was a genuine windfall for anyone who used it without organizing their life around its price. The mistake was made by everyone — riders, cities, businesses — who treated an investor promotion as a permanent condition.

So take the ride. The AI subsidy is the largest transfer of capability to small businesses in the history of software, and refusing it on principle just hands the windfall to your competitors. But take it the way a disciplined operator takes any promotional pricing. Underwrite at the real price: if your business model only works at today's AI costs, you don't have a business model, you have a subsidy with your logo on it. Keep the exits open: build workflows that can switch vendors and models without surgery, because the open-weight escape hatch only saves businesses that never welded their doors shut. And own what matters: your data, your prompts, your processes — the assets that make the intelligence useful — should live with you, not inside a vendor's walled tier.

The renewal notices have started arriving. The S-1s are filed. The memo about unit economics has already been written once, word for word, and the companies that will write it next have collectively promised their future shareholders close to a trillion dollars a year of infrastructure that somebody has to pay for. There's no mystery left in this plot — only timing. Everyone's ride was $5 once. The businesses that thrive in the next decade will be the ones who enjoyed the fare, checked the meter anyway, and never forgot where the bus stop was.