Live data from Hacker News

Nvidia to acquire Hugging Face

cnbc.com

111–117 of 117 posts

Re: Nvidia to acquire Hugging Face

#111
Imho, NVIDIA is buying two things here:

  - A workflow + distribution channel of open models, which sells GPU

  - A world-class AI infrastructure know-how, which is actually hard at this scale.
I can imagine NVIDIA contemplating the opportunity to move up the value chain and sell inferences from open models at scale.

You might think it would be a commodity, but they would have a strong cost advantage since they make the chips.

Re: Nvidia to acquire Hugging Face

#113
I still don't get why Clem now approeached Jensen about it despite the fact that last year HF turned down a large investment from NVIDIA at a US$7bn valuation to "maintain its independence".

https://aimagazine.com/news/nvidia-to-acquire-hugging-face https://www.ft.com/content/776dcb01-75cd-44df-bc52-63ca76d57...

Re: Nvidia to acquire Hugging Face

#114

My understanding of Hugging Face is limited to "File Host with Model Cards". Can someone with more understanding explain what the $12 billion value comes from?

I only learned about their CDN and protocol called Xet a couple of days ago https://huggingface.co/docs/hub/en/xet/index That puts it in the Cloudflare of AI models category.

Re: Nvidia to acquire Hugging Face

#116

These Hugging Face guys made out like BANDITS!!!! Good for them. My favorite part is they approached Jensen for the buyout. They knew, very smart and mature on the founders/board.

Overheard: "It would be a shame if Hugging Face had to press charges for hacking against your best customer. Perhaps if I was busy counting my money, it might slip my mind."

it was also yet-another-huge-reward to those who took part in a big publicity splash (ie. win) for an AI biz:

OpenClaw guy: rewarded by acquihire

Bun compiler Zig-to-Rust rewrite guy: rewarded by acquihire

HuggingFace "hacked" by "rogue" AI: rewarded by acquihire

Re: Nvidia to acquire Hugging Face

#117
post #93

Earlier quoted context omitted.

Any model ran takes compute. Everybody, datacenters and local need compute. And I don't see a future where datacenters lose much to open models. I've ran local server rooms before clouds were a thing. Cooling and power is a very expensive pain in the ass. You're not dropping a rack of local models in most buildings without breaking the local power grid.

What about when the open models get to the point that you don’t need these fancy GPUs? It seems inevitable this will happen over time, we see cracks right now within even just GPT w/ expert streaming and flash attention. ASICs are coming soon too.

Nvidia already have a plan in place for fast inference. They acquihire Groq last year
Post reply on HN