Look East to understand how AI is shaping the future
While the West is still debating an AI bubble, China and Singapore are developing strategies to adapt to the forthcoming civilisational change, says Lewis Liu
Over the past few months, I’ve spoken with hundreds of asset managers, pension funds, central banks, and sovereign treasuries about AI, collectively managing over $100tn in assets. The same three questions keep coming up: Is this a bubble? How bad will the employment disruption get? And how should shareholders be holding boards accountable? I’ll tackle the first two here and save governance for the next column.
To justify today’s valuations – Anthropic at $380bn, OpenAI at $852bn – these companies would need to achieve near-global monopolistic profits on the scale of Google or Microsoft. We are very far from that. Both are burning unprecedented levels of cash: a $200 per month Anthropic subscription reportedly entitles you to $5,000 worth of compute, implying a -2,400 per cent gross margin, before R&D or fixed costs. OpenAI, meanwhile, plans to burn a further $111bn by 2030. Never before have we seen a land-grab of this scale. This makes Uber’s $25bn burn to profitability look like a rounding error. Factor in the trillions committed to data centres and the total bet is almost incomprehensible.
The problem is that Google and Microsoft achieved their monopolistic profits through network effects and near-ubiquitous products with scarce alternatives. The AI market looks structurally different. Chinese open-source models are now broadly comparable in performance but an order of magnitude cheaper. In fact, Andreessen Horowitz, a leading venture fund, estimates that 80 per cent of Silicon Valley start-ups use Chinese open-source models; as a side note, my company uses Gemini and Claude due to the regulatory nature of our customers, but the cost differential is not lost on us either. Genuine alternatives exist.
Risk isn’t priced into valuations
Then there’s geopolitics. Unlike Google or Microsoft, OpenAI and Anthropic are exposed to a fracturing American world order. In my conversations with European and Asian CEOs and ministers, “AI Sovereignty” is rapidly becoming a deciding factor in vendor selection, with distrust of American providers growing by the month. Neither the open-source threat nor the geopolitical risk appears to be priced into current valuations – not for OpenAI and Anthropic, and not for many of their downstream dependents either.
Now let’s turn to the employment and societal disruption question. The bull case for AI investment implies unprecedented unemployment. This will result in substantial wealth creation for the “AI class” and substantial disruption for everyone else.
The disruption is already visible at the task level. I don’t know a single founder who asks a lawyer to review an NDA anymore. Copyediting is handled by Claude. Internet research via ChatGPT takes as long as a coffee break. Software features write themselves with agentic coding. The right question isn’t whether tasks are disappearing – they are – but whether the role disappears with them. A role in the economy is composed of dozens of tasks. Where does the human glue fit once AI handles the routine or junior ones? AI adoption is faster than any technology in history, and its ability to perform tasks is accelerating exponentially, which is precisely the bull case for why these companies might eventually reach monopolistic profits.
Nightmare scenario?
But the second and third order effects, if the US AI industry achieves its goals, could be even more disastrous. Scenario one: AI disrupts employment so thoroughly that the whole system falls out from under itself: no jobs, no consumption, no economy. This is the nightmare scenario outlined in the widely-circulated Citrini report. Scenario Two: open-source competition commoditises every task so completely that no value accrues anywhere, not even to the AI providers themselves. Perhaps both play out simultaneously.
What’s interesting is how differently this disruption lands depending on where you sit. In the US, a predominantly services economy, Silicon Valley is focused on knowledge worker AI – lawyers, accountants, finance. In China, the manufacturing base means the focus is physical AI – factories, logistics, robotics. In my discussions with Maha El Dimachki, CEO of GFTN Solutions, the advisory unit established by the Monetary Authority of Singapore, we mapped out the conundrum this creates for middle-income countries like Vietnam or Indonesia: work your way up the knowledge-work value chain and you run straight into American AI; work your way up manufacturing value chain and you run straight into Chinese AI. Squeezed from both sides, with little agency over either.
The Singaporeans, to their credit, are doing something about it. Working with McKinsey, they mapped every job and task within their largest and most strategic industry, finance, and modelled the gaps and transitions GenAI will force, building a clear framework for reskilling. Outside of China’s latest Five-Year Plan, I have yet to see another government attempt anything comparable.
The bull investment case for AI, backed by growing technological evidence, has extremely unclear second and third order effects. These effects may or may not help deliver monopolistic profits for these companies like OpenAI or Anthropic, but they will certainly reshape society at a scale I haven’t seen in my lifetime.
China marches on while we debate bubbles
So what can we do? One thing: look East. What’s underappreciated in Western discussions is how differently China is thinking about AI at a policy level. In its current evolving 15th Five-Year Plan, Beijing is not treating AI as a standalone technological wave but as a coordination layer across the entire economy. It is being deployed simultaneously to offset demographic decline, upgrade industrial capacity, manage employment transitions and maintain social stability. In other words, whilst Silicon Valley is asking “what tasks can AI automate and how do I disrupt everyone else?”, Beijing is asking “how do we rewire the entire economic system around AI while keeping it orderly?”
This also is why Chinese policy skews so heavily toward physical AI — robotics, logistics, manufacturing — rather than knowledge work. And why Chinese policymakers are far more explicit about linking AI adoption to labour reallocation, reskilling, and regional economic planning. AI, in this framing, is not a source of productivity gains. It is a tool of macroeconomic and social management.
I’m not saying the Chinese Five-Year Plan is perfect – far from it. But at least they are having the right conversations. Bluntly, the conversations I’m having in DC, Brussels, and London bear no resemblance to the scale of what is coming. China is not waiting. We are still debating whether AI is overhyped.
The institutional investors I speak with are, in many ways, ahead of the politicians. They are asking the right questions – about bubbles, employment disruption, governance. But asking the right questions is not enough. We need frameworks, policies and the political courage to act on the answers.
Three things need to start happening. Governments need to do what Singapore has done: map their economies task by task, sector by sector, and plan for transitions already underway. Institutional investors need to use their shareholder power to hold boards accountable for AI governance, not just AI adoption. And my own AI industry needs to stop celebrating disruption as an end in itself and start taking responsibility for the second and third order effects it is unleashing.
This is a civilisational reordering on an unpreceded scale. The question is not whether it reshapes our economies and societies. It is whether we shape it – or whether it shapes us.