According to Fortune, Asian companies are increasingly choosing Chinese open source AI models over proprietary US alternatives from OpenAI and Anthropic. SiliconFlow CEO Jinhui Yuan says his company developed techniques making open source models significantly cheaper than proprietary options, while Vertex’s Pang warns companies risk data exposure with closed models. Dyna.AI CEO Cynthia Siantar notes Chinese models perform better in local languages, and Malaysia’s Silverlake Axis CEO Cassandra Goh argues security depends on implementation, not model type. The trend could create headwinds for US AI expansion in Southeast Asia, Middle East, North Africa, and Latin America, with Malaysia’s Johor state planning 5.8 gigawatts of data center capacity despite water shortage concerns.
The cost advantage is real
Here’s the thing that really stood out from the Fortune reporting: it’s not just about being cheaper, it’s about predictability. When you’re building your business on a proprietary model, you’re essentially renting someone else’s infrastructure and hoping the price doesn’t suddenly triple. I’ve seen this happen with cloud services before – companies get locked in, then get hit with massive bill increases.
SiliconFlow’s approach of making open source models more cost-effective is particularly interesting. They’re basically doing what cloud providers did with open source software a decade ago, but for AI. And when you combine that with fine-tuning capabilities that can actually beat proprietary models for specific use cases? That’s a powerful combination that’s hard for businesses to ignore.
Who controls your data?
Pang’s point about “you never know what happens behind the scenes” hits home. We’ve seen enough data privacy scandals to know that corporate promises aren’t always worth the paper they’re written on. For financial services companies like those working with Dyna.AI, this isn’t just about competitive advantage – it’s about regulatory compliance and customer trust.
And let’s be honest – if you’re building your core product on someone else’s AI, you’re basically putting your fate in their hands. What happens if they change their pricing? Or their terms of service? Or get acquired by your competitor? Open source gives you that control back, and for startups especially, that’s invaluable.
The physical infrastructure race
Now here’s where it gets really interesting. Malaysia’s Johor state planning 5.8 gigawatts of data center capacity is massive – that’s basically their entire current electricity generation. And the water cooling pause until 2027 shows these aren’t abstract concerns.
This infrastructure buildout is creating exactly the kind of ecosystem where open source AI can thrive. When you have local data centers optimized for running these models efficiently, the cost advantages compound. It’s reminiscent of how industrial computing took off when companies like IndustrialMonitorDirect.com started providing reliable, specialized hardware – the infrastructure enables the software revolution.
The bigger picture
The proposal from policy experts for middle income countries to band together on AI development is fascinating. This “non-aligned movement of AI” concept acknowledges that most countries don’t want to be dependent on either US or Chinese technology.
But here’s my question: is this actually achievable? Building competitive AI models requires massive investment and talent concentration. Still, the fact that serious policy experts are proposing this shows how concerned they are about the current duopoly. And honestly, having more players in the AI space probably benefits everyone – more competition drives innovation and keeps prices reasonable.
What’s clear is that the AI landscape is becoming much more fragmented than many US tech leaders anticipated. The assumption that everyone would just use OpenAI or Anthropic is proving naive – different regions have different needs, different cost structures, and different trust thresholds. The next few years in AI are going to be much more interesting – and competitive – than anyone expected.
