Meta just launched a new AI model called Muse Spark, and in the release notes was something I haven’t seen from a major AI lab before: two prices for the exact same model, based entirely on whether you let them train on your data. Pay the standard rate, and your prompts stay private. Pay 92% less on input tokens and 95% less on output tokens, and Meta gets to use everything you send it.
Every enterprise AI contract has some version of a training-rights clause buried in it. Until now, that clause has always been free, a default nobody negotiated and nobody priced. Meta put a number on it and that it tells you what the chats, prompts, and workflows running through your AI tools are actually worth.
Two pricing models, colliding
For as long as software has been sold to businesses, there’s been an unwritten deal: you pay a license fee, and the vendor promises not to touch your data. That promise is baked into SOC 2 reports, confidentiality clauses, and every enterprise AI agreement your firm has signed.
Consumer tech has always run the opposite play. The product is free, and your behavior is the payment. That’s the old line “if you’re not paying for it, you’re the product” and it’s exactly how the free and flat-rate versions of a lot of AI tools have worked without anyone spelling it out.
Meta just merged the two. Instead of just bundling training rights into a free tier, they’re charging you for privacy directly, in dollars, per million tokens. Why now? The public web is basically tapped out. Every blog, forum, and book worth indexing already has been. The next real edge in AI comes from usage data: how people actually reason through problems, what they ask, how they correct the model. Meta would rather harvest that directly from its own users than keep paying an entire industry of data-labeling vendors to manufacture something close to it.
The number worth remembering
Blend Meta’s pricing at a realistic input-heavy ratio for agentic workflows, and the private tier comes out to about $1.35 per million tokens. The data-sharing tier comes out to about ten cents. That’s a spread of $1.24 per million tokens. Meta’s implied price for your data.
At a billion tokens a day (a real number for a business running AI across client work and internal operations) keeping your data private costs roughly $450,000 more a year than letting Meta train on it.
Think about it like cash versus a rewards credit card. Pay cash, and your purchase history stays yours. Swipe the card, and you get a discount, because the card company is selling access to your data. You’ve always had that choice with a credit card. Now you have it with AI.
What this means for your firm
If your organization has adopted AI anywhere you’re very likely sitting on a data gold mine. Depending on your provider, getting access to that data yourself ranges from straightforward to nearly impossible. On an enterprise plan, you can usually export it. On a personal or team-tier plan, you may have to fight to get logs.
That’s the first thing worth checking: not just whether your flagship enterprise contract protects your data, but whether every tier your team actually uses does too. A lot of firms assume “enterprise” coverage extends to a teammate’s personal $20-a-month subscription. It usually doesn’t.
The second thing is bigger picture. This pricing spread is the first real transaction evidence anyone’s had for what AI usage data is worth. Bring it into governance conversations, valuation conversations, and client conversations about AI policy.
Key Takeaways
- Meta’s two-tier pricing is the first standing, per-token price the market has put on AI training data — a 92%–95% discount for giving up training rights.
- The blended $1.24-per-million-token spread is Meta’s own valuation signal for what your usage data is worth.
- Personal and team-tier AI plans may not carry the same training-rights protection as your firm’s enterprise contract — check each one.
- Don’t let employee AI usage go unmonitored. Every token running through your systems now has a demonstrated dollar value.
- This is a real data point for governance, valuation, and client conversations — not a formal appraisal, but evidence.
Want the CPE credit? Take the full lesson on EverydayCPE and earn 0.2 CPE credits: https://everydaycpe.com/courses/technical-finance/whats-your-data-actually-worth/
Leave a Reply