Hook
Over the past seven months, AI-related IPOs have raised nearly HKD 100 billion in Hong Kong. That is 55% of all listing proceeds on the exchange. The Financial Secretary calls it a transformation. I call it a concentration risk. A market where a single narrative captures more than half of all capital is not a sign of strength. It is a signal of crowding. And crowded trades have a history of violent unwinds.
Context
Paul Chan's policy statement outlines a three-pronged strategy: government-led efficiency projects, capital market incentives, and SME adoption targets. The numbers are impressive on the surface. Thirty efficiency projects across thirteen departments. Export growth in high double digits for consecutive quarters. An estimated HKD 65 billion in economic value if SME adoption catches up to large enterprises by 2035.
But strip away the policy language, and the structural picture is different. Hong Kong has no foundational AI model developers. No homegrown GPT equivalent. No large-scale training clusters. The city is an application layer player in a market that increasingly rewards ownership of the base layer. This is not a criticism. It is a positioning statement. The question is whether that positioning is sustainable when the underlying dependencies are external.
Core
Let me break down the technical reality. The government's 30 projects across 13 departments are engineering adaptations of mature models. Document processing. Data analytics. Public service chatbots. These are integration tasks, not innovation tasks. The value creation comes from workflow optimization, not algorithmic breakthroughs.
This creates a specific risk profile. The entire AI stack relies on external model suppliers. Domestic open-source models like Qwen or DeepSeek. Overseas models like GPT-4 or Claude. The layer in between - the adaptation, the integration, the system deployment - is where Hong Kong creates value. That is a defensible niche, but it is also a dependency.
Based on my experience auditing protocols in 2017, I learned that supply chain risk is the first thing you check. When I examined Bancor's conversion logic, I found integer overflow vulnerabilities that would have allowed token creation at arbitrary values. The issue was not the code itself. It was the assumption that the underlying math was sound. Hong Kong's AI strategy makes the same assumption about its model suppliers. If a foundational model provider changes licensing terms, alters its API pricing, or faces regulatory restrictions, the entire application layer built on top becomes unstable.
The capital market data adds another layer of concern. HKD 100 billion raised at a 55% concentration suggests a narrative premium. History shows what happens when narrative outpaces fundamentals. The 2000 internet bubble. The 2021 DeFi summer. The 2022 Terra collapse, where I lost 65% of my portfolio before activating my emergency plan. The pattern is consistent: capital floods into a theme, valuations detach from fundamentals, and the correction is brutal.
Hong Kong's Hang Seng Index has added multiple AI companies to its components. This is a self-reinforcing loop. Passive funds must buy these stocks. Index inclusion drives inflows. Inflows drive valuations. Valuations attract more listings. The loop works until it doesn't.
The SME opportunity is real but misunderstood. The HKD 65 billion figure represents potential value, not guaranteed returns. For SMEs to adopt AI, they need three things: digital infrastructure, technical talent, and clear ROI. Hong Kong's SME sector lacks all three at scale. The government's role should be to close that gap through targeted subsidies and training programs. The policy signal is there, but the execution details remain unclear.
Contrarian
The conventional narrative is that Hong Kong's AI push is a success story. I see a different pattern. The 55% capital concentration is not evidence of strength. It is evidence of a market with limited investment options. Hong Kong's equity market has struggled to attract diverse listings. AI is filling a vacuum, not creating a new paradigm.
There is also a strategic blind spot in the policy framework: computing infrastructure. The statement makes no mention of GPU clusters, smart computing centers, or data center investments. This is critical. Without domestic compute capacity, Hong Kong's AI applications depend on cloud providers like Alibaba Cloud, Tencent Cloud, or AWS. This creates a double dependency: model supply and compute supply. Both are external. Both are subject to geopolitical and commercial risks.
Singapore is not making this mistake. Its National AI Strategy 2.0 includes explicit investments in compute infrastructure and talent development. Dubai is building similar capabilities. Hong Kong's "hub" positioning is being challenged on multiple fronts.
The government's AI projects also raise data governance questions that remain unanswered. Thirteen departments processing citizen data through AI systems requires a clear framework for data collection, storage, and usage. The article does not address algorithm transparency or bias mitigation. For a jurisdiction that prides itself on the rule of law, this is a gap.
Takeaway
Hong Kong's AI strategy is a classic "application layer + capital layer" play. It works in the short term. The capital inflows are real. The export growth is real. The government's execution speed is commendable. But the structural dependencies are building.
Watch for three signals over the next 18 months. First, the quality of AI-related listings. If the market continues to accept "AI-labeled" companies without core technology differentiation, the bubble risk compounds. Second, compute infrastructure announcements. The absence of a smart computing center plan is a red flag. Third, SME adoption data. The HKD 65 billion opportunity is the real test of whether this strategy creates lasting value.
Precision in audit prevents chaos in execution. Hong Kong needs to audit its AI dependencies before the market forces a resolution. The current trajectory is positive. The resilience of that trajectory depends on addressing the hollow core.
Position size dictates peace of mind. Hong Kong's position is large. Its peace of mind should be calibrated accordingly.