On this page
Your open models still work, and so do the weights you’ve already downloaded. Nvidia’s deal to buy Hugging Face hasn’t closed, and nothing Nvidia has said since Jensen Huang resigned as CEO on 12 November changes it. What’s uncertain is the Hub after the deal closes: who sets its rules, and what happens to pricing, access and terms.
Below is what’s confirmed, what isn’t, and how to keep your access to open models whoever owns what.
We make Off Grid AI, including the free apps and Off Grid AI Boost mentioned below, so weigh that. Every fact links to its source.
What happened
Nvidia is buying Hugging Face. Nvidia signed the deal on 2 September 2026 and announced it the next day, for $12.93 billion (Nvidia, SiliconANGLE). Nvidia’s filing says it expects the deal to close in the first half of 2027, once regulators approve it (SEC 8-K). Until then, Hugging Face is still its own company.
Jensen Huang resigned as Nvidia’s CEO. An 8-K dated 12 November 2026 says he left as president, CEO and board member, effective immediately (SEC 8-K). CFO Colette Kress is interim CEO and stays on as CFO while the board looks for a permanent CEO, inside and outside the company. Harvey C. Jones, who has been on the board since 1993, is its first-ever chair (CNBC).
The board says it expects operations, customer commitments and the product roadmap to carry on without interruption. The filing doesn’t mention Hugging Face, and it keeps Nvidia’s quarterly results on 17 November.
What Nvidia says about the Hub
Nvidia’s position is that the Hub stays open. In the deal announcement, Huang wrote that “Hugging Face will remain an open platform for the entire AI ecosystem,” and that Nvidia’s own chips won’t be required to build on or deploy through it (Nvidia). The same post says Hugging Face keeps supporting many clouds and chip makers, and keeps its brand.
Huang made those promises, and he has now left. We found nothing from Nvidia that withdraws or changes them.
What still works
- Weights on your disk. A model file you’ve downloaded runs without the Hub. Nothing in the deal reaches your computer.
- Licenses. The license comes from the model’s maker, like Alibaba for Qwen or DeepSeek for DeepSeek, not from the site that hosts the file. An Apache 2.0 model you downloaded stays Apache 2.0.
- The Hub today. It’s open, with the same plans and the same brand, and Nvidia doesn’t own it yet.
What’s uncertain
- Who sets Hub policy after closing. Nvidia hasn’t said who will run Hugging Face or how its decisions will get made. Hugging Face had about 750 staff when the deal was announced, and Nvidia hasn’t said how many will join it (Implicator).
- Pricing. Nvidia’s announcement doesn’t mention prices for Hugging Face’s paid plans, like PRO at $9 a month (Hugging Face). There’s no announced change, and no promise either way.
- Access and terms. Gated models, rate limits, download rules and the terms of service are all Hub decisions. Nothing published fixes them after closing.
- Ranking and search. The promise covers which chips you can use. Sanchit Vir Gogia of Greyhound Research pointed out that it says nothing about how models get ranked, searched or routed (Implicator).
- Regulators. Their approval is a condition of closing, so the deal’s terms could still change.
Nvidia hasn’t announced a change to any of these. It also hasn’t published rules that lock them in after closing.
How to keep your access no matter who owns what
1. Keep local copies of the weights you rely on
Download the models you depend on and keep them somewhere you control. A change to any site’s terms can’t take back a model you already have.
The free Off Grid AI apps run open models on your own phone and computer, with no account. Off Grid AI Desktop runs on Mac, Windows and Linux, and Off Grid AI Mobile runs on iPhone and Android. Your prompts stay on your device. The model browser in both apps pulls from Hugging Face today, and both apps import a .gguf file you already have, so the copies you saved keep working whatever happens to the Hub. Both apps are open source, MIT on mobile and AGPL-3.0 on desktop (GitHub).
2. Use an API where you hold the key and the serving code is portable
Some jobs need a bigger model than your device can run, or an API your app can call. For those, pick an API where you hold the key and could move the serving code somewhere else.
That’s what we’re building with Off Grid AI Boost. It’s hosted inference for open models, including Qwen 3.8 and DeepSeek V4.1 Flash, on our own GPUs. One API key works for every model on Boost, at one fixed monthly price. We don’t keep your prompts or train on them. The serving code will be open source, so you’ll be able to run the same thing on your own hardware if you leave.
Boost is a pre-order now, API key by January 2027. The first cohort pays from $4.99 a month, half the $9.99 regular price, for as long as they stay. Limits not yet published. Boost runs on GPUs like every other host, and we don’t claim it avoids Nvidia hardware. See the Boost plans, or read more in a flat-price, private API for open models.
3. Know the other hosts
You don’t need one site for everything. These all serve or host open models today. Checked 16 November 2026.
| Where | What you get | Price | Your prompts |
|---|---|---|---|
| Ollama | Free open-source app to run models on your own computer, plus a cloud | Local free; Cloud Pro $20 a month with $60 of credit | Cloud: “never logged or trained on” |
| DeepInfra | Per-token API for many open models | DeepSeek V4.1 Flash $0.098 in / $0.294 out per 1M tokens | Not stored or trained on (docs) |
| Together AI | Per-token API for open models | DeepSeek V4.1 Flash $0.30 / $1.20 per 1M tokens | Stored by default; zero retention is a setting (docs) |
| OpenRouter | One key for 600+ models from many providers | Each model’s price, plus 5.5% when you buy credits | Not trained on; zero retention is a setting |
| Hugging Face Inference Providers | 200+ models through partner providers | PRO $9 a month with $2 of credit; providers’ own prices | Follows each provider |
| ModelScope | Alibaba’s model community, for downloads | Free to download | n/a for downloads |
| Kaggle Models | Google’s model library, for downloads | Free to download | n/a for downloads |
We compare all of these, plus Replicate and Civitai, in the best Hugging Face alternatives in 2026.
FAQ
Will my Hugging Face models stop working?
No. Weights you’ve downloaded run without the Hub, and the Hub is still open. Nvidia says it will stay open after the deal closes.
When does Nvidia’s Hugging Face deal close?
Nvidia’s filing says the first half of 2027, subject to regulators and the usual closing conditions.
Does Jensen Huang’s resignation change the Hugging Face deal?
Nvidia hasn’t said it does. The 8-K about the CEO change doesn’t mention Hugging Face, and the board expects the product roadmap to continue without interruption.
Will I need Nvidia GPUs to use Hugging Face?
Nvidia says no. Its announcement says Nvidia compute won’t be required to build on or deploy through Hugging Face.
Will Hugging Face prices change?
Nobody has announced a change. Nvidia’s announcement doesn’t mention pricing for Hugging Face’s plans.
Is Off Grid AI Boost live?
No. You can pre-order now, and your API key arrives by January 2027. Your plan starts then.