Latest / The Emerging Markets Podcast with Fexingo: Developing Economies, Growth, and Investment / How Emerging Markets Are Building Sovereign AI Infrastructure
Transcript
- Lucas: India is building its own national AI cloud. Not on AWS, not on Azure — on Indian-designed chips, in Indian data centers, with Indian-trained models. And the government just committed about one point two billion dollars to make it happen. Luna: That's a serious sum for a sovereign cloud project. What's driving it — data security concerns, or the sheer cost of renting NVIDIA GPUs from the usual hyperscalers? Lucas: Both, honestly. The cost argument is hard to ignore. To train a frontier-level model on rented H100s from a US cloud provider, you're looking at tens of millions of dollars — and that's before inference costs. India's argument is: we're already spending that money on imports of cloud services. Why not redirect it into domestic infrastructure that we own? Luna: So it's a kind of import substitution for the AI era. I remember we talked about a similar dynamic with rare earths in Kazakhstan a few episodes back. Lucas: Exactly the same playbook. And India isn't alone. Look at Southeast Asia — Malaysia is emerging as a data center powerhouse. The state of Johor, just across the border from Singapore, has attracted over twelve billion dollars in data center investments over the past two years. Companies like GDS, AirTrunk, and even Google are building there. Luna: Right, and that's partly because Singapore put a moratorium on new data centers back in 2021 — the land and power constraints were too tight. So Johor became the overflow zone. Lucas: And now it's a full-blown hub. But here's the interesting twist — a lot of those data centers are designed to serve local AI workloads, not just latency-sensitive cloud services. Malaysia wants to be the AI compute engine for ASEAN. Luna: That's a big ambition given the hardware supply chain is still dominated by the US. But I've seen that Malaysia is also starting to assemble its own chips — not at the leading edge, but for inference workloads, you don't always need the latest three-nanometer process. Lucas: Exactly. Inference is increasingly done on specialized chips, not just general-purpose GPUs. And that opens the door for countries like India, Vietnam, even Indonesia, to design their own accelerators using open-source architectures like risc v. Luna: Vietnam is an interesting case. We touched on semiconductor outsourcing a few episodes ago. But are they moving beyond assembly into actual chip design for AI? Lucas: Slowly. The government has a national strategy to train fifty thousand engineers in semiconductor design by 2030. And companies like FPT are already building AI chips. But the real action right now is in data center construction. Vietnam just approved a one-billion-dollar AI data center near Ho Chi Minh City, backed by local conglomerate Vingroup. Luna: How do these sovereign projects square with the fact that most of the world's AI models run on NVIDIA hardware? Are they building their own CUDA alternatives too? Lucas: That's the billion-dollar question. India's national AI cloud will use a mix of NVIDIA GPUs and locally designed chips initially. But the long-term play is software stack independence. They're investing heavily in open-source frameworks like PyTorch and TensorFlow, and they're funding an Indian foundation to develop a cuda compatible layer for their homegrown hardware. Luna: So they're trying to avoid the trap of vendor lock-in from the start. Smart, because we've seen how hard it is for even big companies to break away from the NVIDIA ecosystem. Lucas: Right. And the geopolitics of this is fascinating too. The US has export controls on advanced AI chips to China, but those controls also affect China's neighbors. If you're a Vietnamese or Indonesian startup, you might find it easier to buy NVIDIA GPUs through a US cloud provider than to import them directly. So building domestic AI infrastructure is partly a way to bypass those supply chain bottlenecks. Luna: But doesn't that create a new dependency — on us based cloud providers who own the data center hardware? Or are these emerging market data centers actually owned by local entities? Lucas: It's a mix. In India, the national AI cloud will be owned by the government's Centre for Development of Advanced Computing. In Malaysia and Indonesia, many of the new data centers are joint ventures between local firms and global operators. But the trend is toward more local control. And that has implications for how data flows across borders. Luna: Speaking of data flows, let's talk about the Chinese tech stocks. We've seen a big rally this week — Alibaba up over fourteen percent in five days, the KWEB ETF up three percent. Is that connected to this AI infrastructure buildout? Lucas: I think so, in part. Alibaba reported strong growth in its cloud and AI segments. They're also building their own AI data centers across Southeast Asia. And the Chinese government just relaxed some rules on cross-border data transfers, which should make it easier for Chinese cloud providers to serve clients in the region. Luna: So we have two competing models: the US hyperscaler model, where AWS and Azure provide AI compute globally, and the sovereign model, where countries build their own. Plus Chinese providers like Alibaba Cloud and Huawei Cloud. It's a three-way tug of war. Lucas: Exactly. And the winners might be the emerging markets themselves, because they can play off the competitors. Look at Indonesia — they're hosting data centers from AWS, Alibaba, and local firms all in the same industrial park. They get the investment, the jobs, and the tax revenue, while keeping some leverage over pricing. Luna: That's a smart play. But it also raises questions about digital sovereignty — whose laws apply to the data stored in those centers? If an Indonesian user's data is on an Alibaba Cloud server in Jakarta, is it subject to Indonesian law, Chinese law, or both? Lucas: That's the frontier issue. Indonesia's new data protection law, similar to India's, requires that personal data of citizens be stored domestically and that foreign companies appoint a local representative. So in theory, Indonesian law applies. But enforcement is complicated. Still, the trend is clear: emerging markets are asserting more control over their digital infrastructure. Luna: And on the investment side, how should an EM investor think about this? Are there specific opportunities beyond the obvious hyperscalers? Lucas: I'd look at two buckets. First, the enablers — companies that build and operate data centers in emerging markets. There's a firm called GDS Holdings, listed in the US, that is the largest data center operator in China and expanding into Southeast Asia. Second, the chip designers — companies like India's MosChip or Vietnam's FPT are interesting, but they're smaller and more speculative. Luna: And the ETFs we track — like EEM and VWO — they give broad exposure, but they're heavy on financials and energy. To get pure AI infrastructure, you'd need something more targeted. Lucas: Right. And that's the challenge: the AI infrastructure story in emerging markets is still nascent. Most of the revenue right now goes to US companies. But over the next five years, as more local capacity comes online, that could shift. It's a long-term thesis. Luna: Before we wrap, I want to circle back to something you said about open-source chips. You mentioned risc v — what's the significance for emerging markets? Lucas: It's a game-changer. risc v is an open instruction set architecture, so no licensing fees. Countries like India and China are betting big on it for AI accelerators. India's VEGA series of chips is risc v based. If those chips can deliver competitive performance for inference workloads, it could dramatically lower the cost of building sovereign AI infrastructure. Luna: So you're saying a country that can't afford to design its own ARM or x86 chips could still build a functional AI cloud using risc v? That's a huge democratizing force. Lucas: Exactly. And it's not just theory. There's a startup in Bangalore called Tenstorrent — actually, it's a Canadian company with a big R&D presence in India — that's building risc v based AI accelerators. They're already shipping to customers in Southeast Asia. So the pieces are coming together. Luna: I like that. It reminds me of the whole 'China plus one' supply chain diversification. Now we're seeing a 'US plus one' for AI infrastructure. Emerging markets are building their own lane. Lucas: And that's the big picture. For decades, emerging markets were consumers of technology built elsewhere. Now they're becoming builders. It's a structural shift that will play out for the rest of this decade. Luna: If walking through the economics of AI infrastructure with us has made something click, you can support the show at buy me a coffee dot com slash fexingo. It's a simple way to keep this podcast ad-free and independent. Lucas: Exactly — small contributions from listeners make a big difference. It helps us cover topics like this without chasing sponsors. And we're grateful for everyone who's chipped in. Luna: Alright — let's get back to the data. We mentioned the Chinese tech rally. But the broader EM picture is mixed. The iShares MSCI Emerging Markets ETF, EEM, is down about one percent this week. So the AI infrastructure story is still a sub-theme within a larger complex market. Lucas: True. And the strong dollar is still a headwind for many EM economies. But interestingly, the dollar index has pulled back a bit — it's at 120.7, down from 121.15 last week. That's a small relief, but not a game-changer. Luna: We also have the yuan holding steady at 6.79 per dollar, while the yen weakened to 160.9. The Japanese yen weakness is making Japanese tech exports cheaper, but that's a different story. Lucas: Right. For emerging markets, the key takeaway is that the AI infrastructure buildout is a bright spot in an otherwise challenging environment. Countries that invest in digital sovereignty may be better positioned for long-term growth. Luna: And for investors, it's worth watching which companies and countries execute on these plans. The opportunity is real, but so is the execution risk. Lucas: On that note, I think we'll leave it. Thanks for listening — we'll be back next week with another deep dive. Luna: Take care, everyone.