What the study measured
The Center for the Governance of AI examined 375 large language models released between June 2018 and May 2026. It asked a question that owners rarely see in the coverage of any single launch: when a new model ships, does it reach everyone at the same time?
The answer was no. At least 11 percent of advanced models were delayed or never made available in the European Union, against a lower rate in the United States, with the United Kingdom in between at 7 percent. Of the 68 clear cases of delay or non-release, 56 were attributed to regulation, and data-protection law was named as the primary driver.
Why the delay reaches your business
A delay measured in weeks sounds academic until it lands on your desk. One leading model arrived for European users 71 days after its launch elsewhere, and one major developer delayed or withheld 26 percent of its releases in the EU. A competitor in a lighter-touch market can build on a capability for two months before you are allowed to touch it.
That turns regulation into a timing question, not only a paperwork question. The rules exist for good reasons and they are not going away. But an owner who assumes their toolset matches a US rival's is planning on an availability that the data does not support.
What owners should do about it
Treat access timing as a real input to strategy. Ask your vendors, in writing, when a new model reaches the EU, not only when it launches globally, and factor that lag into any plan that depends on a specific capability. Where a delay is caused by a data-protection review, an early and clean compliance path is often what shortens it.
Then decide what you do while you wait. A second, EU-available model tested in advance keeps you moving when the one you wanted is held back. The goal is not to fight the regulation. It is to stop being surprised by it.
Read next: Your Confidential Messages Can Be Scanned Again · Europe Just Repriced Defence Tech
