The rise of open source AI

For as long as I can recall in my journalism career — 15 years and counting — open source technology has existed as a helpful but ultimately minor curiosity in terms of my coverage.

Back in the early 2000s, the top stories I saw at the outlets I worked, MSN, The Atlantic, and AOL, were all about Microsoft clashing with Apple over PCs and MP3 players — both of them offering proprietary operating systems and software.

Yes, of course, Linux existed and there was always some interesting stuff happening in the ecosystem, but in terms of what stories were most read by readers, it certainly wasn't the Linux ones.

The same dichotomy unfolded in mobile in the late 2000s and early 2010s, with Apple releasing the proprietary iPhone in 2007, challenging and dethroning Research-in-Motion's Blackberry phones and the Palm devices, only to be, in turn, challenged and for many years outnumbered by Google's open source Android mobile devices. Yet even Google offered a proprietary Google-controlled app store for downloading Android apps.

Now, as the AI era continues to blossom, it seems open source may finally becoming the star of the show.

Not only is Meta's Llama 3.2 and entire family of models finding incredible uptake — with more than 400 million downloads — but according to a new feature from VentureBeat founder and CEO Matt Marshall, both it and French startup Mistral's AI models are finding a large audience among enterprise customers sick of paying for API fees and the lack of ownership over the underlying tech.

Another notable story this week was the launch of Hugging Face's HUGS (Hugging Face Generative AI Services), a new platform designed to help enterprises and teams adopt open source even if they lack the hardware — such as expensive GPUs and servers — necessary to run it.

And in the synthetic media (AI generated media) side of things, a startup called Genmo launched an open source video generation model, Mochi 1, it says outperforms leading top proprietary rivals such as Lionsgate partner Runway and Luma AI .

The rise of open source AI not only threatens proprietary closed source leaders such as OpenAI and Anthropic — the latter turning heads with its new Computer Use mode released this week — but also stands to remake the tech industry as we know it.

If open source AI continues to advance, match or exceed the performance of closed source rivals, it will place more pressure on them to justify their value to users.

I think that's largely good for consumers and for technology writ large — competition theoretically reduces costs and results in better products and consumer experiences.

But it does raise the question of who will succeed in an open source AI dominated world, and how — what types of businesses?

Much of the chatter I've read recently has suggested that those businesses who can build "wrappers" atop AI models — open or closed — stand to be the big winners, as it is the application layer that can make the tech useful to consumers and provide something they are willing to pay for, unlike, say, a raw model that requires several GPUs to run.

What do you think? Let me know carl.franzen@venturebeat.com

Thanks for reading, writing, subscribing, sharing and just being you. Have a great weekend!

— Carl Franzen

Read More