Idea check: licensing AI labs access to institutional data where it sits, instead of selling datasets. Real business or compliance fantasy? I will not promote

I’ve spent the past year talking to AI labs on one side and data-holding institutions (hospitals, government agencies, large enterprises) on the other, and I keep hitting the same wall from both directions. Labs have burned through the open internet. What they want now is proprietary domain data: clinical records, chemistry and drug discovery data, regional language speech, that kind of thing. The institutions that hold it will almost never sell it or hand over copies. Legal, privacy and IP fears kill the conversation before anyone even discusses price. So the idea I’m working on: an access layer where the data never leaves the institution. The lab’s training runs against the data where it sits, under a license, and the owner keeps full control the entire time. Licensing access, not transferring datasets. The closest existing analog I’ve found is MELLODDY, the pharma consortium where ten companies trained shared models on federated data without ever exposing it to each other. Where I am: mapped roughly 700 potential supply-side institutions and 200+ AI labs as buyers, early conversations running on both sides, starting with healthcare and drug discovery. Bootstrapped, no raise. Before I go deeper, I want this stress-tested by people who’ve actually touched either side: 1. If you’ve bought or sourced training data at an AI company: is in-place access genuinely workable for your researchers, or do teams quietly insist on raw copies for iteration speed? 2. If you work inside a hospital, bank or government body: who actually has the authority to say yes to something like this, and what kills it first, legal, IT, or plain inertia? 3. Pricing instinct: annual license, per-training-run, or revenue share? 4. Does the intermediary survive, or do labs just go direct once they know who holds the data? 5. Is synthetic data closing this gap faster than I think? Brutal takes welcome. I’d rather find out here than 18 months in.