The 55% Problem
Here's a number that should make every serious crypto and tech investor pause: from December to May, AI-related new stock listings in Hong Kong raised nearly HK$100 billion. That's 55% of all IPO capital raised in the city during that period.
Fifty-five percent.
Let that sink in for a moment. In a market that has historically been dominated by property developers, banks, and consumer goods companies, AI has suddenly become the dominant narrative. The Hang Seng Index is adding AI companies to its roster. The Financial Secretary is publishing policy essays about "comprehensive AI implementation." The government has launched 30 efficiency projects across 13 departments.
Code doesn't lie, but narratives do. And this narrative deserves a closer look.
I've spent years auditing blockchain projects in Bangkok, watching ICO mania and DeFi summer unfold. The pattern here is painfully familiar. When a single sector captures 55% of all capital formation in a major financial center, you're not looking at organic growth. You're looking at a feeding frenzy.
The Application Trap
Let me be clear about what Hong Kong is actually doing. This isn't a technical breakthrough story. This is a policy-driven application story.
Paul Chan's essay reveals a clear strategic choice: Hong Kong is positioning itself as an application-layer and ecosystem-layer player, not a foundation-model competitor. The 30 government efficiency projects across 13 departments aren't pushing the boundaries of AI research. They're deploying mature technology in government workflows—document processing, data analysis, public service chatbots.
That's engineering innovation. Combination innovation. Not architecture-level breakthroughs.
Hong Kong has no homegrown foundation model labs. No equivalent of Beijing's Zhipu AI, Shenzhen's Pengcheng Lab, or Hangzhou's DeepSeek. The territory's AI strategy necessarily depends on external model supply—Alibaba's Qwen, DeepSeek, GPT-4, Claude—and creates value through scenario adaptation and system integration.
This is a rational choice. Building foundation models requires billions in compute, years of research, and enormous uncertainty. Hong Kong's resource endowment doesn't support that game.
But here's the uncomfortable truth: application-layer dependence means permanent followership. Hong Kong will not set AI technical standards. It will not own core AI intellectual property. It will be a tenant in someone else's infrastructure.
I've seen this movie before in crypto. Remember all those "blockchain application" companies that raised money in 2017 by promising to put supply chains on distributed ledgers? They were building on Ethereum, on Hyperledger, on whatever protocol was hot at the time. When the infrastructure narrative shifted, their applications became obsolete.
The alpha hidden in the noise is this: application-layer bets in rapidly evolving technological stacks are structurally fragile. You're building on sand that the next foundation-model release will wash away.
The Capital Market Signal
Let's dig into the numbers because they tell a more complex story than the official narrative.
AI-related IPO proceeds hitting 55% of total capital formation is extraordinary. For context, AI-related listings on Nasdaq typically account for 20-30% of IPO volume. Hong Kong is running at nearly double that rate.
What does this tell us? Three things.
First, Hong Kong has become the listing venue of choice for AI companies, particularly from mainland China and Southeast Asia. The city's capital markets infrastructure, common law system, and international professional services ecosystem make it an attractive destination for AI firms seeking public listings.
Second, there's a self-reinforcing narrative loop at work. The Hang Seng Index adds AI companies → passive funds must buy them → valuations rise → more AI companies want to list → the index adds more. This isn't inherently fraudulent, but it creates a feedback mechanism that can detach prices from fundamentals.
Third, and this is the uncomfortable part: the "AI-related" label is doing a lot of heavy lifting. Based on my experience auditing whitepapers during the ICO boom, I can tell you that sector labels in frothy markets attract a lot of tourists. Companies with a chatbot on their website and a "machine learning" slide in their pitch deck suddenly become "AI companies." The definitional elasticity matters because it determines whether we're looking at genuine technological value creation or narrative arbitrage.
Trust is the new currency, and right now, the market is trusting a very broad definition of "AI."
The 650 Billion Question
The most interesting number in this entire policy document isn't the HK$100 billion in IPO proceeds. It's the HK$65 billion in potential economic benefits from closing the SME adoption gap.
According to the research cited in Chan's essay, if small and medium enterprises catch up to large enterprises in AI adoption by 2035, Hong Kong could unlock HK$65 billion in economic value. That's roughly 2.2% of Hong Kong's 2023 GDP.
Let me translate this into terms I understand from auditing DeFi protocols.
This is like finding a yield opportunity that could theoretically return 2.2% of GDP, but requires solving a multi-dimensional coordination problem involving infrastructure, talent, regulatory clarity, and cultural change. The potential exists. The path to realizing it is uncertain.
Why are SMEs lagging? The research doesn't say, but my experience in Southeast Asia gives me some hypotheses.
Cost is a barrier. AI tools have a learning curve and a budget line. For a trading company with 20 employees in Kwun Tong, spending HK$50,000 per year on AI tools that might save two hours per employee per week isn't an obvious ROI. The benefits are diffuse and delayed. The costs are immediate and concrete.
Talent is a barrier. You can buy AI software, but you can't buy AI literacy. Someone needs to understand what the tool does, how to prompt it effectively, how to validate its outputs. That person is expensive and hard to find. Hong Kong's AI talent pipeline is thin, and the competition from mainland cities and Singapore is fierce.
Trust is a barrier. This is where the crypto analogy gets sharp. In 2020, I watched DeFi protocols struggle to attract mainstream users not because the technology didn't work, but because people didn't trust it. The same dynamic applies to AI adoption. SMEs need to trust that AI won't hallucinate critical business documents, leak sensitive data, or make decisions that get them in regulatory trouble.
The HK$65 billion is real potential value. But potential is not the same as realized. The gap between the two is filled with execution risk.
The Infrastructure Blind Spot
Here's what's missing from the official narrative: compute infrastructure.
The policy essay is silent on GPU clusters, data center capacity, and supercomputing resources. This silence is deafening.
Hong Kong faces genuine physical constraints. Land is scarce. Electricity is expensive. The climate—hot and humid—is hostile to data centers that generate enormous heat. Building large-scale AI compute infrastructure in Hong Kong is technically challenging and economically questionable.
But the absence of a compute strategy creates a strategic vulnerability. If Hong Kong's AI applications depend on cloud services from Alibaba, Tencent, or AWS, then the territory's AI future is rented, not owned. Supply chain risks, data sovereignty concerns, and vendor lock-in all become structural constraints.
I've seen this dynamic in the crypto world. Projects that built on centralized infrastructure—whether it was a single cloud provider or a single oracle—eventually hit walls. Decentralization isn't just a philosophical preference. It's a risk management strategy. The same logic applies to AI.
Hong Kong's likely solution is the "mainland compute + Hong Kong application" model, leveraging the Greater Bay Area's infrastructure. This makes economic sense. But it introduces data governance complexity. Government AI applications that process citizen data will need to answer uncomfortable questions: Where is the data stored? Who has access? What happens under cross-border data transfer rules?
These questions aren't unanswerable. But they need to be answered before the applications scale, not after.
The Regulatory Tightrope
The policy essay says nothing about AI ethics or security. This is a notable omission for a government pushing AI adoption across 13 departments.
Government AI applications involve citizen data. Identity information, tax records, public service usage data. The privacy and security requirements for these applications are fundamentally different from commercial use cases. Citizens deserve to know when and how government systems are making decisions that affect them.
Hong Kong faces a unique regulatory challenge. Under "one country, two systems," it must navigate between mainland China's AI governance framework—which includes generative AI management measures and algorithm filing requirements—and international standards like the EU AI Act and OECD AI Principles.
This isn't an easy balancing act. The mainland's approach emphasizes state oversight and content control. The EU's approach emphasizes individual rights and risk classification. Hong Kong's own Personal Data (Privacy) Ordinance provides a baseline, but it predates the AI era and doesn't address algorithmic transparency or automated decision-making.
The risk is "deploy first, govern later." I've seen this pattern repeatedly in the crypto industry. Projects launch, users adopt, regulators scramble to catch up. Sometimes the catch-up is benign. Sometimes it's not.
Government AI adoption without a governance framework is like deploying smart contracts without an audit. The code might work. But you're assuming risk that you haven't quantified.
The Singapore Shadow
Every discussion of Hong Kong's AI ambitions eventually collides with reality: Singapore.
Singapore has a National AI Strategy 2.0. It has invested heavily in compute infrastructure. It has a proactive talent attraction program with favorable visa and tax policies. It is building an AI ecosystem that includes research institutions, startups, and multinational corporations.
Hong Kong's advantages are real but narrow. The capital markets edge is significant—no one can match 55% AI-related IPO concentration. The common law system and international professional services ecosystem are genuine differentiators. The "super connector" role between mainland China and global markets has unique value.
But capital markets advantages are easier to replicate than research ecosystems. If Singapore continues to build AI research capacity while Hong Kong remains an application and capital hub, the long-term trajectory favors Singapore. Hong Kong will have the listings. Singapore will have the technology. The listings are more visible. The technology is more durable.
I don't have a crystal ball, but I've watched enough market cycles to recognize when a hub is building on borrowed foundations. Hong Kong's AI strategy is a "borrowed infrastructure" play—borrowed models, borrowed compute, borrowed talent. That's not inherently wrong. But it's inherently fragile.
The Bear Case
Let me steelman the skeptics.
The 55% IPO concentration could be a bubble signal. If many of these "AI companies" are AI-enhanced traditional businesses rather than core AI technology firms, then the market is paying AI valuations for non-AI earnings. When the correction comes—and corrections always come—the damage to investor confidence could be severe.
The 30 government efficiency projects are nice, but they're a rounding error in the context of Hong Kong's HK$2.9 trillion GDP. Government AI adoption is a signal, not a driver.
The export growth is real but likely reflects global AI hardware demand flowing through Hong Kong's trade channels rather than Hong Kong-originated AI product exports. The value added is thin.
The SME opportunity is real but faces formidable barriers. Talent is scarce. Costs are high. Trust is low. And the timeline to 2035 is long enough for multiple technological shifts to render current assumptions obsolete.
What I'm Watching
Here's my practical framework for tracking whether Hong Kong's AI bet is working.
Three-month signals: Are the 30 government projects actually publishing results? Is the IPO pipeline maintaining its AI concentration? Is the Hang Seng Index continuing to expand AI inclusion?
Six-to-eighteen-month signals: Is there a concrete AI talent program? Visa changes? Tax incentives? Housing support? Is there any announcement about compute infrastructure? Data center investment? Smart computing partnerships? And critically—what's the SME adoption data? Is the gap narrowing?
Long-term signals: Is Hong Kong's AI industry value-add as a percentage of GDP actually growing? Is the HK$65 billion SME opportunity showing signs of materialization? And how is Hong Kong's global AI ranking evolving?
The Bottom Line
Hong Kong's AI strategy is a bet on application-layer value creation, capital markets intermediation, and regional connectivity. It's a rational strategy given the territory's constraints. But it's a strategy with structural limitations.
The alpha hidden in the noise is this: Hong Kong is becoming the world's largest marketplace for AI narratives, not AI technology. The city will facilitate capital formation, provide regulatory arbitrage, and serve as a regional headquarters. It will not be a center of AI innovation.
That's not necessarily bad. Marketplaces make money. But they're exposed to narrative risk. When the AI narrative shifts—when the next technology wave replaces the current one—Hong Kong's 55% concentration becomes a liability rather than an asset.
I've seen this pattern before. In 2017, Bangkok was full of ICO evangelists. In 2021, it was NFT artists. The platforms changed. The capital flows moved. What remained were the infrastructure builders who created durable value rather than narrative value.
Hong Kong's AI future depends on whether it can become an infrastructure builder rather than a narrative consumer. The current policy direction suggests application deployment and capital market development. That's a start. But it's not enough.
Code doesn't lie, but narratives do. And the narrative that Hong Kong is becoming an AI hub is only half true. It's becoming an AI application hub. Whether that's enough to sustain its economic ambitions is the question that will define the next decade.
The answer depends on execution, talent, infrastructure, and governance. The essay doesn't address any of these with sufficient depth. That's not necessarily a criticism of the policy—it's a limitation of the communication. But in a market where narratives drive capital flows, communication limitations become strategic vulnerabilities.
Trust is the new currency. And trust requires transparency about limitations, not just celebration of strengths.
