By SavantCat
A few days ago I was thinking about something: for years we have said Chinese AI is catching up, but how far has it actually got? Only when this news came out on 3 September did I feel for the first time that the answer had changed.
HUMAIN, the AI company under Saudi Arabia's Public Investment Fund, formally released its Arabic large model, HUMAIN M3. Backed by a sovereign fund and aimed squarely at national self-reliance in AI. Yet the foundation of this much-anticipated model does not come from Silicon Valley; it comes from China's MiniMax.
My first reaction was that this is far more than a tech news item.
1. First, see clearly what actually happened
HUMAIN M3 is not borrowing Chinese technology for a moment — it is using a Chinese open-weight flagship model as the foundation of a national AI effort:
- The base is MiniMax M3, the open-weight flagship MiniMax released in June this year, a mixture-of-experts architecture with 428 billion parameters;
- The Saudi team did secondary training on top of it with more than 1 trillion tokens of Arabic data;
- After training it averaged 89.37% across seven authoritative Arabic benchmarks and took first place in five of them, beating several mainstream international models on the composite score;
- It is also planned to open the model weights and support local deployment, laying the groundwork for Saudi Arabia's own AI capability.
Put simply: the richest country in the Middle East wanted its own AI and chose not to build from zero, but to take a Chinese company's open foundation, reprocess it and turn it into a nationally branded Saudi model.
The heaviest sentence in the story comes from an industry interviewee: the definition of a model going global is being rewritten.
2. Why rewritten? Because going global has moved from selling finished goods to exporting foundations
In the past, Chinese AI going global followed two main routes:
- Selling APIs: packaging model capability as interfaces and charging by usage — selling water;
- Selling application products: shipping finished apps abroad for overseas users — still selling finished goods.
On both routes Chinese companies remained the side that gets called and gets used, with limited pricing power and replaceable at any time.
This time it is completely different. MiniMax opened its foundation, and Saudi Arabia took it, did secondary training, built its own model and plans autonomous deployment — the first time a Chinese model has been embedded into another country's AI infrastructure. The initiative and the voice it carries are on a completely different scale from selling APIs.
This is not an isolated case. The data started speaking some time ago:
The 2026 spring open-source ecosystem report showed that Chinese open-model platforms already account for 41% of downloads, placing them near the front of the global open-source ecosystem.
From the world mainly using overseas models to the world beginning to take Chinese models as a foundation — that curve was bought by the maturity of the open-source ecosystem. And once your model becomes someone else's bedrock, you become something like water, electricity and coal in the AI world: unremarkable day to day, impossible to do without.
3. What has this to do with those of us who do not build models?
You may think this is a matter between giants, and what has it to do with an ordinary person or a small shop?
It does. And more than you would think.
Layer one: when foundation capability spills over, the first to catch it are the people doing delivery.
If Saudi Arabia can take a Chinese open model and build national AI, the foundation is mature enough to be reprocessed and industrialised. So what about the countless small businesses in China? There is even less reason for them to build from zero. The valuable work is never making another model, but installing an already strong, mature foundation into a concrete industry and a real business process. That is exactly the bowl of rice the people doing AI delivery eat from.
Layer two: this is a cognitive reset — stop looking at domestic AI with a chaser's mindset.
I often hear small business owners say foreign AI must be better than domestic. But when a language as unfamiliar to us as Arabic can, on a Chinese foundation, take first place in five of seven tests, you should at least reassess once: in most Chinese-language business scenarios, the question with domestic models is not whether they are good enough, but why you have not already put one to work.
Layer three: do not wait for the strongest model; start from a foundation that is good enough.
Technology will always iterate, and you will never reach the day of the strongest. But an open foundation like MiniMax M3, already proven able to carry a national-level task, has long been more than enough for the shop downstairs. Borrow the capability first, use it first, get the first business process working — that beats everything.
What it comes down to for you and me — three sentences
First, do not treat domestic AI as a compromise; it is already a foundation that can carry national-level tasks. Shift the mindset and the opportunities become visible.
Second, if you are a business owner, stop agonising over which model to use and think about which part of my operation AI can take over. A model is water and electricity: plug in and there is power. The hard part is deciding which light to switch on first.
Third, if you work in services, do not build from zero; do the work of installing a mature foundation into a concrete industry. A country is already using a Chinese open model as its bedrock; your own industry is long overdue for being wired up too.
Chinese AI walked a long road to get from being called on by the world to being used by the world as a foundation. And only now that it has reached this point do ordinary people really begin to have a chance to borrow that strength and earn their own money — because the strongest foundation has been laid, and what remains is who builds the building first.
Does your industry have a link that should long ago have been transformed by AI but that nobody has done yet? Talk about it in the comments — maybe you will be the one to build the next building.
#AIDelivery #DomesticLLM #OpenSource #ArtificialIntelligence #SmallBusiness