HuggingFace and the Future of AI Hosting
The other day, I listened to an AI podcast where HuggingFace's Head of Product discussed their partnership with Amazon, which has been in place for years and has recently become closer. As I understand it, Amazon provides all their hosting, storage, and bandwidth via AWS, and part of that partnership is that they receive significant discounts compared to a regular company.
According to the interview, HuggingFace already has many thousands of paying customers, and they're aiming to be the easiest or only solution for any company that wants to quickly and easily host some kind of AI solution. They are trying to be the "Github of AI," a free platform that becomes the de facto solution in its space that millions of developers/enthusiasts use for free and is the obvious (and perhaps only) option when a company wants to pay for that same service.
Despite the low costs of disk space and bandwidth, companies are still reluctant to host their own models, and no one can distribute larger models on GitHub or Google Drive. At HuggingFace's scale, they absorb significant costs to get brand recognition, find paying customers, and develop a good platform. Any class or tutorial will tell people to use HuggingFace. If you distributed a model somewhere else, people will ask for it on HuggingFace.
Due to general startup VC funds, ad-money, and subsidies from their pricing plans, their costs for hosting model binaries are likely easily taken care of. Even if they're running at a loss, this is an investment for them worth sinking $$$ into as they monopolize the market as the go-to AI "hub".
Tags: HuggingFace, AI hosting, Amazon, AWS, Github, machine learning
Source Links: HuggingFace Pricing