Here is a sentence that should make every federal CIO sweat: The Trump administration just cut deals to give government agencies access to ChatGPT for $1, Google Gemini for 47 cents, and Grok for 42 cents.

Sounds like the deal of the century. It is not.

ProPublica just published a devastating investigation showing exactly how this playbook works, because the federal government already fell for it once. In the early 2020s, Microsoft offered the government $150 million in "free" cybersecurity upgrades after a wave of Russian and Chinese hacks left agencies reeling. Satya Nadella looked like a hero. The press releases wrote themselves.

What actually happened? After installing the upgrades, federal customers were locked in. Switching to a competitor would have been so expensive and painful that they had no choice but to pay the full subscription fees once the free period ended. One former Microsoft salesperson told ProPublica it was "successful beyond what any of us could have imagined."

Now the same pattern is playing out with AI. OpenAI, Google, and xAI are racing to get their tools inside every federal agency, and the pricing is so low it barely qualifies as a business transaction. The agencies get to say they are "adopting AI." The companies get something far more valuable: integration into government workflows that will be nearly impossible to rip out.

ProPublica lays out three cautionary tales from the cloud computing transition, and the parallels are uncomfortable. The government rushed to adopt cloud technology with nearly identical rhetoric: it would make America more prosperous, more efficient, more secure. The Obama administration used those exact words. The Trump administration is using them again, just with "AI" swapped in for "cloud."

The core problem is vendor lock-in. Once ChatGPT is processing classified briefings, once Gemini is embedded in procurement workflows, once Grok is generating intelligence summaries, switching costs become astronomical. The 42-cent price tag becomes a multi-billion dollar dependency. And at that point, the company sets whatever price it wants.

There is a deeper problem too. These AI systems are black boxes. When the government adopted cloud computing, at least the data was still stored in formats that could theoretically be migrated. AI workflows are different. The prompts, the fine-tuning, the institutional knowledge baked into custom models: none of that transfers cleanly to a competitor.

The government is not buying AI tools. It is signing adoption agreements that will determine which three or four companies control the infrastructure of American governance for the next decade. And it is doing it for the price of a candy bar.

The companies will call it public service. The government will call it modernization. ProPublica, correctly, calls it a trap. One we have already walked into before.