Lin Qiao argues companies should host and control their own AI models instead of licensing proprietary APIs
Story Overview
Lin Qiao, CEO of Fireworks AI, is urging enterprises to run and fine-tune their own models on controlled infrastructure rather than paying for access to closed APIs, citing gains in privacy, customization ownership, and long-term cost control. Hugging Face co-founders Thomas Wolf and Clément Delangue have publicly amplified the same stance through shares and quotes that highlight open-source progress as a practical path forward.
Open-source leaders throw weight behind ownership
Wolf reposted a deep dive on open-source model infrastructure while Delangue quote-tweeted Qiao's framing of owning versus renting intelligence, aligning with Hugging Face's long-running emphasis on community-driven models over proprietary clouds.
Enterprise data advantages stay partly unproven
The pitch rests on claims that most relevant data stays private and that self-hosting cuts dependency risks, yet exact cost comparisons, latency benchmarks, or adoption metrics for companies like Uber or DoorDash remain unspecified in the current discussion.
Positive users back Hugging Face's push to own AI via open weights as it avoids dependency on rented closed models, while negative users reject closed rental systems with recurring fees.
Most Activity
Love this framing from Lin: it’s time for all to own intelligence instead of renting it!
http://x.com/i/article/2066399704347463680
@lqiao
@lqiao The unsettling part isn’t that it was shut down per se (and, that is unsettling). It’s that “your” intelligence could be shut down by someone who was never in the room. Renting was always a bet that the landlord’s interests would stay aligned with yours. We got an answer.
@lqiao
@lqiao Agreed
@lqiao Fable 5 has definitely one distinction of being world's most hyped Model
My interaction with it wasn't any better than Gemini3.1 Pro or Chatgpt5.5
However the hype surrounding itself was blown out of proportion
@lqiao Some frontiers are discovered. Others are created.
@lqiao https://open.substack.com/pub/procurefyi/p/the-unbearable-lightness-of-the-flywheel?r=223ajc&utm_medium=ios
@ClementDelangue 😒 This is spot on. 🍲
@ClementDelangue The problem isn't owning the intelligence, but owning the hardware to run it.
@lqiao Good points! Apart from all that, if you use a closed model, isn't it fairly trivial for a lab to completely reconstruct your proprietary workflows (with data), and bake the capabilities into their models? Or offer custom harnesses per profession?
This is the next layer of tenancy.
Frontier APIs rent cognition directly.
Open models plus post-training let firms own more of the model surface.
But the intelligence environment still includes compute, chips, cloud, energy, inference, identity, compliance, procurement, and distribution.
The class divide is not API versus open model.
It is who owns the estate in which intelligence becomes usable.
@lqiao I think this aligns with @satyanadella
@lqiao Here’s the model!
@lqiao Own it with http://NexxusOS.live Desktop
@ClementDelangue bittensor:native
Yeah, it makes sense. I wrote about this recently from a different lens.
Most people see renting intelligence as a business model. I think the deeper question is dependency. If the intelligence you rely on isn’t yours, the threat may eventually become existential.
Here is the article: https://www.linkedin.com/pulse/brain-isnt-yours-do-you-have-business-rohit-gupta-lpcff?
@ClementDelangue open weights ftw
@lqiao A genuine competitive moat never hinges on how powerful a third-party model you can access via API calls Instead, it lies in cultivating a proprietary intelligence asset built on general foundation models—one that belongs solely to you and cannot be replicated or seized
@ClementDelangue Yes. I do not want another closed system with a monthly fee pretending to be progress.