Latest / The Emerging Markets Podcast with Fexingo: Developing Economies, Growth, and Investment / Why Chinese AI Models Are Gaining Ground with US Companies
Transcript
- Lucas: So there's a headline this morning that stopped me: 'Chinese AI models are gaining ground with US companies as OpenAI and Anthropic costs surge.' And the data backs it up. KWEB, the China internet ETF, is up over five percent in the last five days. Pinduoduo up nine and a half percent. Even Baidu, which has been struggling to get its AI story heard, is up two percent. Luna: That's a big shift. A year ago, US companies wouldn't touch Chinese AI with a ten-foot pole — national security, data privacy, the whole nine yards. What changed? Lucas: Cost. Pure and simple. OpenAI and Anthropic have raised prices multiple times. Training a frontier model now costs tens of millions, and inference costs are through the roof. Meanwhile, Chinese providers like DeepSeek, Alibaba's Qwen, and Baidu's Ernie are offering comparable performance at a fraction of the price. Luna: Comparable performance — that's the key. Are they really comparable, or is it a cheap knockoff? Lucas: Look, for many enterprise use cases — customer service chatbots, code generation, summarization — the gap has narrowed to within a few percentage points on standard benchmarks. For a startup watching its burn rate, a ten percent performance hit might be worth a fifty percent cost saving. Luna: And the yuan helps. It's at six point seven nine per dollar right now, which makes Chinese services cheaper for US buyers. Lucas: Exactly. The trade-weighted dollar index is at one hundred twenty point seven — still strong. That dollar strength means any yuan-denominated service gets cheaper every month. It's like an automatic discount. Luna: But what about the regulatory risk? US companies using Chinese AI could face scrutiny from the Committee on Foreign Investment, or CFIUS, or get caught up in export controls. Lucas: That's the tension. Some are using Chinese models indirectly — through a cloud provider's API hosted in Singapore or Ireland. Others are using open-source models from China, which they can run on their own infrastructure. That avoids direct data transfer to China. Luna: So it's a workaround. But is it sustainable? If the US tightens rules, those companies could be forced to switch back. Lucas: That's the bet. But here's the thing: once you've built your product around a model, switching costs are high. You've fine-tuned it, you've integrated it, your team knows it. If the performance is good enough, companies might lobby to keep access open. Luna: Interesting. So Chinese AI isn't just a cheap alternative — it's becoming an entrenched option. Lucas: Right. And it's not just startups. I've heard of mid-size tech firms in the US quietly testing Chinese models for internal tools. They don't advertise it, but the cost savings are too big to ignore. Luna: What about the model quality? I remember when DeepSeek came out, people were impressed but said it lagged on reasoning tasks. Lucas: That's still true on the hardest benchmarks — math, coding competitions, things that require multi-step logic. But for most business applications, the difference is negligible. And the Chinese models are improving fast. Alibaba's Qwen 2.5 is competitive with GPT-4 on several metrics. Luna: So we're seeing a bifurcation: high-end US models for cutting-edge research, Chinese models for cost-sensitive deployment. Lucas: That's one way to slice it. Another way is that the AI market is becoming more like the semiconductor market — specialization by use case and price point. And just like with chips, China is finding its niche. Luna: What does this mean for the broader emerging markets thesis? Is this a China-specific story, or can other EM countries get in on it? Lucas: China is unique because of its scale, its AI talent pool, and its government support. But I think India could follow — they have the engineers, and their IT services firms are already building AI solutions for global clients. Vietnam and Mexico are also trying to attract AI R&D centers. Luna: So the 'AI arbitrage' play is real. Buy cheap compute from emerging markets, sell to developed-world clients. Lucas: Exactly. And it's not just about compute. It's about data. Chinese companies have access to massive datasets from their domestic market, which helps them train models that generalize well to other emerging markets. That's a competitive advantage. Luna: Let's zoom out a bit. How does this affect the us china tech war? Is this a sign of cooperation or continued decoupling? Lucas: It's complicated. On one hand, US companies using Chinese AI shows that there's still a market for technology transfer. On the other hand, it's happening despite tariffs and export controls, which suggests the controls are leaky. The US might try to close those loopholes. Luna: But if they close them too aggressively, they risk pushing more innovation to China. It's a delicate balance. Lucas: Right. And in the meantime, investors are voting with their dollars. KWEB is up, Chinese tech stocks are rallying. The market sees this as a growth opportunity, not a risk. Luna: One more thing: what about the ethical concerns? Chinese AI models are trained under censorship rules. Can US companies trust that the models won't embed political bias? Lucas: That's a real concern. Some companies do their own fine-tuning to remove sensitive content. Others accept the bias as a trade-off for cost savings. It's a risk they're willing to manage. Luna: It's fascinating. We're watching the emergence of a multi-polar AI world, and emerging markets are right in the middle of it. Lucas: Absolutely. And I think this is just the beginning. As costs continue to fall and models improve, we'll see more cross-border AI adoption. Luna: If this episode made something click for you — about how the economics of AI are shifting — you know where to find us. It's buy me a coffee dot com slash fexingo. Just a small way to keep the show ad-free and independent. Lucas: Yeah, every bit helps us keep digging into these stories. So back to the point: I think the next frontier is in AI services — not just models, but full-stack solutions from emerging markets that package model, infrastructure, and support. That's where the real value will be. Luna: And that could reshape the global IT services industry, which is already centered in India. Lucas: Exactly. Indian IT firms are well-positioned to become AI integrators, taking Chinese or open-source models and customizing them for Western clients. That's a huge opportunity. Luna: So we're not just looking at a cost play — it's a strategic shift in how technology is sourced and deployed globally. Lucas: Right. And it's happening faster than most people realize. The headlines are about tariffs and chip bans, but the real story is in the quiet adoption of cheaper AI from unexpected places.