OpenAI's $1-a-Year Government Deal Is Over — and That's the AI Cycle Turning From Land-Grab to Cash

The most telling number in this week's reporting on OpenAI's federal business is not the "50% off." It's the "$1." That was the old price — a token dollar per year for the U.S. federal government to use ChatGPT at scale. The largest, best-credit customer in existence was paying less per year than a vending-machine snack.

The reported change replaces that arrangement with a 50% discount off list prices. Step back from the discount framing and read the move for what it is: a sharp increase in what the government now pays. The reason it matters has nothing to do with discount math and everything to do with where the AI business cycle happens to sit.

A dollar a year was a billboard, not a price

For most of this boom, the leading labs priced like a startup in its seed round: hand over the product, win the account, sort out revenue later. The federal government was the ultimate land-grab — a massive, procurement-sticky, multi-year customer carrying national-security prestige. The prize was the inside track against Anthropic and Google, not the invoice. At a dollar a year the unit economics were meaningless by design. That contract was a billboard, and the budget line for "revenue earned" was never the point.

Against that backdrop, the shift to 50%-off is the cleanest evidence I have seen that the giveaway phase is closing. A vendor does not start charging its most favored customer halfway to list until it believes it no longer has to buy the relationship. The government already bought into ChatGPT — thousands of civil servants now use it, and switching an agency to a rival's tool is a slow, expensive, compliance-heavy ordeal. That switching cost is the moat, and OpenAI is now monetizing it. This is the software-layer part of the cycle coming into view: the build-out wins the installed base, and then the pricing lever starts to turn.

The discount cuts both ways

The analyst in me, though, does not let the "off" in "50% off" pass unexamined. A half-price ceiling is still a ceiling. OpenAI is collecting far more from the government than a dollar a year, but still meaningfully less than it gets from its best commercial customers. So this is an improvement to monetization, not a validation of full-rate pricing.

It is also a real cost step for the customer. Agencies work on fixed, congressional budgets and are trained to re-bid as costs rise. A jump from nominal to half of list is exactly the kind of line-item that invites a recompete — and the government has options: Anthropic and Google both court the same agencies, and commodity and open-weights models have driven inference costs down year after year. The result is the same tension the industry has in supply commitments: the demand signal is strong, but the pricing and leverage signals now carry risk that did not exist when the product was free. Adoption that OpenAI bought cheaply is not adoption it owns forever.

What it means if you can only buy the public names

OpenAI is private, so none of this lands directly on a brokerage statement. The signal does travel. Microsoft — the company that hosts ChatGPT Gov on Azure and is OpenAI's largest strategic backer — is the most direct way to own the monetization story, and it does so at a market value near $3.65 trillion on a trailing P/E of about 27 times. Palantir is the other large federal-AI software exposure, though at roughly 135 times trailing earnings the market has already paid up handsomely for the defense-AI boom.

Read the headline at the portfolio level, and the conclusion is not about Microsoft or Palantir specifically. It is about the shape of the cycle: the height of the give-the-product-away phase is behind, and the biggest vendors have begun extracting revenue even from their most favored customer. That is a net positive for the margin and cash-flow side of the AI complex, and it argues against the tired "this is all subsidies with no revenue" bear case. The judgment to carry is the timing one: a pricing step like this is bullish for monetization but it places a "will they actually pay" question on the table that a standing free deal never asked.