Hugging Face just quietly retired the paid tier of its Inference API, a move that’s been visible on their pricing page for anyone paying attention. The free tier isn’t going anywhere — you can still hit the API without dropping a credit card. But if you need enterprise-grade speed and reliability, the company is now pointing you toward Inference Endpoints, a newer service built for exactly that use case.
The shift makes sense when you look at how Hugging Face has been building out its monetization strategy. Alongside Endpoints, they’ve rolled out hardware upgrades for Spaces, so ML engineers can run demos on GPUs or other specialized hardware without a subscription. Just add a payment method and go. The company’s positioning is refreshingly direct: they monetize by selling compute access for AI, not by gatekeeping models or basic API calls.
Billing now funnels through a single dashboard where you can manage PRO subscriptions, swap payment methods, and see three months of usage history. Charges hit at the start of each month, with consolidated invoices for your records. For teams already juggling multiple Hugging Face services, this consolidation is overdue.
What’s notable here is the broader pattern. Hugging Face has spent years as the neutral ground for open-source AI, but neutral doesn’t pay for GPU clusters. The push toward Inference Endpoints and paid Spaces hardware suggests they’re betting that developers who train on the Hub will eventually deploy there too — and pay for the privilege. Whether that bet pays off depends on how competitive their pricing stays against AWS SageMaker and Replicate, both of which are circling the same customers.
