With so much capital is focused on a small select group of companies like OpenAI and Microsoft, we’ve created our own form of central planning, just with corporate overlords instead of government ones, says Lewis Liu
The US stock market continues to march toward record territory, with everyone from Big Tech CEOs to Goldman Sachs speculating about an “AI bubble”. Most conversations focus on the so-called circular dealings between Big Tech firms: Nvidia investing in OpenAI, which spends money on Oracle data centers, which spend money on Nvidia chips, to the tune of hundreds of billions of dollars and valuations mark-ups into the trillion-dollars, the “infinite money glitch” as some memes call it. Moreover, there is an underlying assumption that for these investments to pan out, the US AI ecosystem must achieve some kind of AGI.
However, I think there is a far more “uncomfortable” conversation to be had about the US stock market, considering that many economists think that if it weren’t for AI, US GDP would be flat. And that conversation is about China.
See, I spent last week in Singapore, chairing a panel at the Singapore FinTech Festival on the use of AI digital twins and other AI tools for banking regulation use cases with a number of central bankers. During this time, I also spoke with Singaporean founders, lawyers, and investors, attended a reception hosted by the German Ambassador, partied with some Asian crypto bros, and even got to hang out with my old elementary school friend! All this is to say: I surveyed a broad spectrum of the Singaporean business community.
Singapore at a knife-edge
If Singapore balances itself at a knife-edge between the US and China, then there is one thing I learned loud and clear from all these conversations: China’s position in the global AI race has accelerated exponentially, underpinned by one primary thing: open-source models.
To contextualize this, it’s useful to understand what has made Singapore so successful as a city-state, and this is something I think we can deeply learn from. That thing is openness. Most Singaporeans I’ve met have a voracious curiosity toward learning from other systems and are generally extremely practical about applying the best technology or thinking to their own context. As such, their perspective on the US and China question regarding AI really stuck with me.
Let’s start with model supremacy. According to the latest Hugging Face leaderboard, as of the beginning of this month, China has 14 of the top 20 models in terms of performance, with the majority being open source. In fact, there are no US open-source models in the top 20. According to a recent post by Andreessen Horowitz (a prominent Silicon Valley venture fund) partner Martin Casado, “80 per cent of [Silicon Valley] AI start-ups applying for VC funding with Andreessen Horowitz are using Chinese open-source AI.” Moreover, it has become apparent that Cursor, the “supernova” vibe-coding startup, uses Chinese models as its core reasoning engine. One hypothesis I’ve been nursing is that one of the reasons Chinese models are so efficient and good at reasoning is that the Chinese language is far more logical than English – it has almost zero exceptions to its grammar, and its tens of thousands of characters are far more precise.
This is profound. If the application layer is being built on open-source Chinese models that are significantly more cost and compute efficient than Western models, how does this flow back into the revenue streams of OpenAI and others, and thus how does this impact the existing AI-fueled valuations in the equity markets? Is the demand as high as we expect for data centers? For example, if the market shifts to Chinese models that are orders of magnitude more efficient, what happens to demand for compute? We are at the stage where there is more investment in data centers than in physical offices, a signal of AI replacing humanity, but what if those data centers are not needed? And if, as mentioned, US GDP growth is solely due to data centers, what does this mean for the underlying American economy and American exceptionalism?
In terms of applications, there is a general acceptance that whilst the US is better at foundational generalist models (though that itself is being questioned), the Chinese are much better at use cases, unit economics, and adoption. I felt this very strongly when pitching my new start-up to Singaporeans. One of the key questions I get asked is whether I’ve looked at the various tools and AI apps the Chinese have built and whether we are getting inspiration from them, a question I never got ten years ago. Clearly people are looking at both sides and trying to find the best solution.
AI robots
However, what is more profound is that, in addition to pure software where the US shines, I have learned that the Chinese are also super focused on applying AI to hardware and robotics, a whole layer of AI applications that are only just coming into focus here in the US. For example, Bezos just raised $6.2bn to build AI for manufacturing, something the Chinese have already been working on for years. Perhaps one explanation for the divergence is that the US needs its AI to have AGI due to its service-oriented economy, whilst China is more manufacturing-based, so therefore its AI is more application-focused.
As such, what I have learned from the Singaporeans is that they are, like the smartest Silicon Valley entrepreneurs, hedging their bets with both Chinese and American models. DeepSeek is as much a part of their vocabulary as OpenAI. This is key. Just as the Singaporeans have such openness in learning, we need to have honest conversations about the genuine state of the art – and not be blinded by just one or two companies.
What I mean by this is that in the US, so much capital is focused on a small select group of companies like OpenAI and Microsoft. As a venture capital investor, the question I constantly hear is “Will OpenAI do this? If so, I won’t invest.” This is exactly the opposite of the disruption-based thinking that powered Silicon Valley’s success in the past. We’ve created our own form of central planning, just with corporate overlords instead of government ones.
In China, by contrast, the vast diversity of applications and adoption based on open-source models means there is far more experimentation with business models and use cases that aren’t bottlenecked by worrying about future “national winners.”
The AI race is going to be global, and the trick is to make sure AI benefits humanity. Right now, ironically, it is Chinese models that are making cutting-edge AI available for anyone to build with for free. We need to up our game in the West by being more open with AI development and diversifying our thinking about AI applications. Otherwise, we become victims of our own central planning dressed up as market capitalism.