Navigating the US-China AI technology split

You can also listen to this podcast on iono.fm here.

SIMON BROWN: I’m chatting now with Dawie Scholtz, managing director and partner at Boston Consulting Group, BCG. Dawie, appreciate the early morning time.

A new report is out from the BCG Institute: ‘The great divide – how the US and China are splitting the AI world’. AI traces are really becoming geopolitical choices, and not just around which models are more powerful or better suited.

There is a lot hanging over a decision for either a company or a government, particularly across perhaps the sort of middle road, which is Africa.

DAWIE SCHOLTZ: Good morning, Simon. Thanks for having me. Great to be on. Yes. it’s fascinating. We’ve been looking into this in great depth. If you allow me to talk through it a little bit,  it’s interesting that a large entity like a government or a large company nowadays needs to think about deploying AI as a first critical asset for their business.

But let’s just think about it across a sort of tech stack of really four interrelated things. The first thing is the sort of core compute, the data centre capacity where the computer is coming from.

Read:
Quarterly market commentary: A global overview
How US-China rivalry, Taiwan and rare earths shape global investment risks
Tencent, Hon Hai ride AI wave
Investing in a world gone tech-crazy

Second is the sort of cloud capability of where the computer is going into the cloud and how the company is accessing it – the models that you then deploy in the cloud. So that’s where ChatGPT and Claude and all these things live.

And finally, the applications that they’re building themselves to generate value with these things.

What’s interesting is that the US seems to be building a fundamentally distinct cloud where if you want to use, for example, Microsoft Azure as your cloud provider, then you are sort of locked into mostly using the US stack – and the same on the Chinese side.

It’s becoming an increasingly interesting choice for governments and for companies to decide which one to lock into and participate in, because there are fundamentally different economics – and of course also fundamentally different geopolitical implications, especially in a world with increasing geopolitical tension. This is a super important choice for companies to take these days.

Read:
Alibaba releases AI model with performance rivalling Anthropic
Tech bubbles, AI, and emerging markets
AI is ‘new engine’ keeping Chinese economy from harder landing

SIMON BROWN: I want to delve into the geopolitical. But before we do, let’s quickly touch on that ‘economic’, because part of it is that the Chinese models are mostly open-weight, cheaper to run. You kind of don’t get – it’s like the Chinese cars – I suspect you get sort of 80% of the ability, but you get it at a fraction of the price.

That Kimi K3 is maybe 5% the cost of running a ChatGPT frontier model, so there is a huge price difference as well. This is a massive economic [challenge] before we get to the other challenges.

DAWIE SCHOLTZ: Yes, exactly. I think the Chinese car analogy is a very good one. It’s a bit like buying a car that’s about a third of the price – or maybe a fifth of the price if you look across the whole stack – but with 80-90% of the performance.

So I think the economics are attractive in many instances. But of course the geopolitics of this has to be considered carefully – the interoperability.

Even within the single stack different kinds of systems need to be considered. So the economics are a huge part of it, but not the only driving factor.

Read: Alibaba shares rise after unveiling upgraded flagship AI model

SIMON BROWN: The geopolitics  – I can see a world in which Donald Trump stands up and says, ‘You can’t deal with the American government if you use a Chinese open-weight model’.

The geopolitical is perhaps the markedly harder question, because there is a divide. And to be clear, maybe Xi Jinping does it. Maybe he says, ‘Don’t bring your ChatGPT or Anthropic here to Beijing’.

DAWIE SCHOLTZ: Yes, exactly. The implication is that large multinationals in particular need to think very carefully about interoperability between systems. Where does it make sense to take the cost to go interoperable? Where do we fully commit to a particular kind of stack?

And indeed you’re right. I think the risk that is emerging is a sort of geopolitical risk where access to your competitive advantage through AI could be cut off in somewhat challenging situations.

Read:
Tech chiefs at Davos highlight AI’s power as a geopolitical tool
Nvidia warns Trump curbs on China AI chips to cost $5.5bn

We’re seeing very interesting examples, like a very large European manufacturer that has a subsidiary in China locked into the American tech stack, but then at some point had to realise they had to create essentially a shadow version of this stack for their Chinese business – of course incorporating each number of costs.

And I think in the South African environment there are particularly companies that are in South Africa, but which are part of larger groups and need to think very carefully about this particular choice in their instance as it relates to the larger group of companies that they may be part of. And of course government also has crucial decisions.

SIMON BROWN: For government these are even harder decisions. I get your point. Absolutely.

But a quick last question. That resilience then becomes perhaps more about redundancy, perhaps more about modularity, so that you’re not locked into any one and you can be flexible – depending on whether you’re talking to Beijing or Washington.

DAWIE SCHOLTZ: Yes, exactly. Modularity of course, brings cost. So then there’s also this trade-off between cost and modularity. And I think companies are trying to minimise cost as much as possible, and control costs as much as possible.

There’s a lot of talk about token costs and all these things. So cost is absolutely on the agenda.

But at the same time, as you say, at the end of the day you can’t be subject to a sort of geopolitical development that breaks down your competitive advantage.

Read:
OpenAI models accessed cloud platform before Hugging Face hack
Warsh, war and AI wariness
Anthropic’s Amodei rejects open model ban but calls for testing
China state media says support for Open AI models has limits

SIMON BROWN: We’ll leave it there. Dawie Scholtz, managing director and partner at BCG, Boston Consulting Group, appreciate the early morning time.

#Navigating #USChina #technology #split