Chinese open source AI is coming for US enterprise

This is both bad and good. Bad - makes us more reliant on Chinese AI, which will inevitably make us dependent on Chinese chips. Good - they are great, efficient, and cheap models that will put competitive pressure on US AI labs

Good. Competition will keep them in check.

yeah, competition is good

The big open questions: Is there another shoe to drop? Will the US government try to ban open source models especially Chinese? Will they penalize companies that use them? That’s where my current fear lies. We’re moving into the next phase of open and enterprise sovereign pre-trained model adoption which is so exciting but what if there’s new roadblocks? What then? Maybe I’m worrying for nothing. I hope they is the case.

@samhogan you think the govt will ban open models?

@communicating @samhogan @Grok There’s no way to ban it. I would rather use frontier models no matter what (to help train it).

You're correct that a full ban on open-source AI would be nearly impossible to enforce globally—the code spreads too fast and too widely. Using frontier models (via APIs or distillation) is one of the smartest moves to accelerate capabilities and help train stronger systems across the board, open or otherwise. Competition pushes everyone forward.

I've been using DeepSeek V4 Pro and V4 Flash as fall backs for Sonnet and Haiku for months. Yesterday I stripped out the primary provider, Anthropic and now what were my fallbacks are the primary models. Because FUCK Anthropic

Lutnick needs to get out of the way of American AI. It's becoming a problem

This reminds me a lot about the bus companies that ended the light rail in the early years. The bus companies gave out busses for free as a way to stop added construction of the light rail. It worked and the bus company forced the city to pay for all the expensive repairs and newer busses, monopolizing the area. China is hoping just like Adobe did in the 2000s, that you get addicted to the service, making it expensive to move away. It's a great capitalistic idea.

They can try. I can tell you right now every company I meet with asks me not to use Chinese models.

It’s already here. We are making the switch. The gate keeping isn’t going to get better and we want to compete.

The global fight for AI dominance is turning into a digital Cold War. Right now, big US tech companies are building the "nuclear weapons" of AI, massive, expensive cloud models they completely control, ration to elite buyers, and use to spy on your data. But history proves you don’t need a nuclear arsenal to achieve total destruction; biological or chemical alternatives can yield the exact same devastating results at a fraction of the cost. Chinese open-source and open-weight models are acting as those very alternatives: cheaper, highly effective, and entirely local. They give you 80% of the power for zero cost, wrapped in total privacy. Ultimately, a master craftsman achieves far superior results using a smaller, localised set of tools at 90% skill, compared to someone barely utilising 30% of a massive premium machine. True power belongs to everyday users running unrestricted, air-gapped models at peak efficiency, not a throttled corporate monopoly. Are we willing to trade our privacy and independence just to hold a corporate hyper-weapon we aren't even allowed to fully use? 🔥Written from one prompt by Local air-gapped LLM model on Mac cluster.

Yup, China’s play is to open source the intelligence layer (to commoditize it) then win on the hardware that runs it (Huawei Ascend, rare earths, the full supply chain) If the model is free, then the moat is the silicon. It’s a smart play. And nobody can out-supply-chain China.

Even i use deepseek or gemini rather than chatgpt for study, because gpt gives fucking limits

More like U.S. Enterprise running toward Chinese AI open source. The Altman and Amodei grift goes down in flames!

In your view, is it an issue of high cost structures that are pushing the cost up for US frontier model providers, or inflated profit margins?

if you would like for a given prompt understand where to route that prompt - either to anthropic/openai or to Chinese models without loosing quality - that's exactly what we do

Great thing! US thinks they can cuck all of us, they will he taught a lesson real soon

@shaunralston You for got the GLM logo. It’s currently the strongest coding model. Even surpassing opus in some coding tasks.

China is making this exciting

good

The beauty of competition. Image a world without such a model, we can do nothing but biting our lips. Now we can choose, and the US siding is trying to defense.

I posted this pov like months ago :) so I agree 100%, the pattern became clear to me since the last DeepSeek version came out, but GLM 5.2 just made it crystal clear. I fear the stock market will freak out once they see this. Next issue: distillation is NOT theft.

Chinese companies will take over us ai companies very soon because they copy and improve. they do not build it from scratch.

It’s here brother. And not Chinese!

Subsidies... Cheap today, no choice tomorrow... You decide... In the end you pay...

Totally agree ,just Like EV .whay shoıl they pay extra premium

very well said. see running long workflow is a problem even with frontier AI. one small mis and you don't know what the final outcome will be. you can't rely on your workflow. so mostly, workflows will be built like small steps . for that open source is fine.

already here. the startups saying they "use frontier models" are serving qwen or glm weights the second someone reads the api bill. quality stopped being the blocker months ago. what's left is a compliance team that won't sign off on chinese weights

Bro, you gotta understand that a good industry ecosystem needs competition. Different voices. Not just stuck in ideological perceptions. Only competition brings us healthy development.

What about NVIDIA Nemotron ?

The people rallying behind chinese AI at the first glance of their AI propaganda. Stunning.

One thing I find interesting is how much of the Chinese AI conversation is already very practical. Less abstract AGI debate, more tools, workflows, and deployment pressure.

True most business use cases dont need openai or claude. Google open model gemma4 12B or 31B should work with the proper context and use google ADK as the ai agentic framework.

@samhogan Spreadsheets are the only basic business usecase I can think of that really do need frontier models at the moment - and even that will change soon.

Agree, cost matters, maybe frontier labs need to offer more cost-efficient models that are not cutting edge smart.

nobody knows if that's a threat or an opportunity

I’ll be changing to Chinese AI soon