Hong Kong's AI Strategy: The Capital Pipeline Without a Compute Spine

The data point demands attention. From December to May, AI-related IPOs raised nearly HKD 100 billion. That is 55% of all capital raised on Hong Kong's exchanges in that window. Compare that to NASDAQ, where AI-linked listings typically account for 20-30%. Hong Kong has become an AI fundraising machine. But the machine runs on narrative. The core question is not whether Hong Kong can attract AI capital. It is whether the jurisdiction has the technical infrastructure to justify the valuations attached to that capital. This is a policy statement from Financial Secretary Paul Chan. It is not a technical document. It contains no architecture diagrams, no model specifications, no efficiency benchmarks. It is a signal of intent. Hong Kong wants to be an AI application hub. The government has deployed 30 efficiency projects across 13 departments. The narrative emphasizes economic empowerment, not technological breakthrough. This is the classic profile of a market that has chosen to be a consumer of AI, not a producer. And in this market, I see a structural risk that the capital markets have not yet priced in: the absence of a sovereign compute foundation. The Context: Hong Kong's Application-First Posture The Hong Kong AI push is an "application-first, efficiency-first" doctrine. The government is focused on deploying mature technologies in administrative settings. Document processing. Data analysis. Public service chatbots. This is not innovation. This is integration. The 30 projects across 13 departments are engineering tasks. They take existing models and adapt them to local workflows. This approach has a clear logic: it avoids the high capital expenditure and long cycle times of foundation model research. It avoids the uncertainty of frontier research. It puts Hong Kong in a position to benefit from AI without betting on the technology's direction. This is the rational choice for a small, open economy. But the approach creates a dependency. Hong Kong has no large-scale foundation model lab. It has no GPT-class architecture in development. It has no serious GPU cluster of its own. The technical route for the city depends on external suppliers. That means Alibaba's Qwen or DeepSeek from the mainland. That means OpenAI or Anthropic from the US. Hong Kong is a system integrator. It does not own the underlying intellectual property. This is not a criticism. It is a strategic choice. But the choice has a direct consequence: the city is permanently in a position of technical followership. It is a tenant in the AI value chain, not an owner. The capital market is not pricing this in. 55% of IPO proceeds are tied to AI narratives. The Hang Seng Index is incorporating AI-linked companies. This is a self-reinforcing loop. The index pulls in passive capital, which inflates valuations, which attracts more listings. But the underlying value creation is not happening in Hong Kong. The models are built elsewhere. The compute is built elsewhere. The patents are elsewhere. Hong Kong is the distribution point, not the factory. And when you are only a distribution point, your leverage is limited to the regulatory and financial system. That leverage can be replicated. It is not a moat. The Core: The Missing Compute Spine Let's examine the hard infrastructure. Hong Kong's AI ambitions require compute. The 30 government projects require inference runs. The financial sector's AI applications require processing. The SME adoption drive requires accessible compute. But Hong Kong has no large-scale AI data center. It has no national compute center. It has no GPU cluster of the scale that would support sovereign AI development. The physical constraints are real. Land is scarce. Power is expensive. The climate is hot and humid. Data centers need cooling. Hong Kong's geography is not ideal for massive compute infrastructure. The natural answer is to rely on mainland compute. The Greater Bay Area has resources in Shenzhen and Guangzhou. The logical architecture is "mainland compute, Hong Kong application." This arrangement works. But it introduces a problem: data governance. Government AI applications process sensitive data. Tax records. Identity information. Public service usage. Can that data cross the border? Under the current regulations, the answer is complicated. The mainland's data exit rules are strict. Hong Kong's Privacy Ordinance has its own constraints. The legal framework for cross-border data flows is not yet seamless. There is a strategic blind spot in this policy. The article does not mention compute infrastructure. The Financial Secretary did not announce a plan to build a computing center. This is a gap. Without sovereign compute, Hong Kong's AI application layer depends on external cloud providers. Alibaba Cloud. Tencent Cloud. AWS. This creates a supplier lock-in risk. It creates a security risk. And it creates a compliance risk. If government data is processed on third-party cloud infrastructure, the audit trail must be impeccable. The question of where data is stored is not just a technical question. It is a liability question. I have audited protocols where the team claims to be building a platform but is actually renting infrastructure from a third party. The result is always the same. The platform is only as secure as the third party's infrastructure. The same principle applies to Hong Kong's AI strategy. The city is renting the AI stack. It is building its entire application strategy on rented infrastructure. This is not sustainable in the long term. If the mainland supplier changes its model access policy, Hong Kong's AI applications are affected. If the US imposes export controls on AI chips, Hong Kong's supply chain is disrupted. The city is exposed to external variables that it does not control. The second blind spot is talent. Hong Kong's AI strategy is an application strategy. Application strategies require engineers, not just users. The city needs people who can adapt models, who can fine-tune models, who can build data pipelines. This is not frontier research. But it is not low-skill work either. The supply of such talent in Hong Kong is limited. The local universities are good. But they are not generating the volume of AI engineers that the market needs. The city must import talent. The current immigration and talent programs are not AI-specific. There is no dedicated AI talent visa. There is no tax incentive for AI engineers. There is no housing support. This is a bottleneck. The numbers back this up. The estimated HKD 650 billion economic benefit from SME AI adoption is the core opportunity. But this is potential value, not realized value. The SME adoption rate is currently low. The reasons are not trivial. SMEs lack the digital foundation. They lack the talent. They lack the budget. They lack the awareness. The government's role is to bridge this gap. The 30 government projects are a start. They demonstrate the use case. But they do not solve the talent problem. They do not solve the compute problem. The gap between the potential and the actual is wider than the policy suggests. The Contrarian View: Hong Kong Is Building a Casino, Not a Lab Here is the contrarian angle. Hong Kong is not building an AI ecosystem. It is building a financial instrument around AI. The 55% IPO share is not evidence of AI strength. It is evidence of AI speculation. The market is attaching a premium to the label. The question is whether the underlying companies are real. The "AI" label is broad. It includes fintech companies with a chatbot. It includes logistics companies with a predictive model. It includes companies that have merely changed their pitch deck. The market cannot distinguish between a real AI company and an AI-washing company. The 55% figure is not a sign of health. It is a sign of narrative concentration. And concentration is a risk. When the narrative breaks, the market breaks. This is a pattern I have seen before. The 2021 NFT explosion. The 2017 ICO mania. The market was full of projects with a label. The actual code was missing. The actual infrastructure was missing. The result was a crash. The same pattern is emerging in Hong Kong's AI market. The market is funding AI narratives. The infrastructure is not there. The talent is not there. The compute is not there. The market is funding a story. The story will eventually be tested. And the test will be the actual performance of the AI companies. If they cannot deliver real products, the valuations will collapse. The damage to the market will be broad. The second contrarian point is the concept of "Hai Nao" — the Chinese term for "Sea of Computing". The AI competition is not about applications. It is about infrastructure. The US has NVIDIA. The mainland has Huawei and Alibaba. Singapore has a national AI strategy. The UAE has a sovereign compute plan. Hong Kong has nothing. It is the only major financial hub in Asia without a sovereign AI infrastructure plan. This is a competitive disadvantage. The city is building its strategy on rented infrastructure. The landlord can change the terms. The tenant cannot. The final contrarian point is the ethical gap. The 30 government projects involve citizen data. There is no public framework for how these algorithms are audited. There is no public framework for how the bias is measured. There is no public framework for how citizens can challenge AI decisions. The government is moving fast. The governance is not moving at the same speed. This is a risk. If an AI system in the government makes a biased decision, the public trust in AI will be damaged. The adoption rate will drop. The economic benefit will not materialize. The Infrastructure Argument: A Call for Sovereign Compute The solution is not to stop the application strategy. The solution is to build the missing layer. Hong Kong needs a sovereign compute strategy. This is not about competing with mainland China. This is about having a fallback option. It is about having the ability to run sensitive workloads locally. It is about being able to audit the infrastructure. It is about being able to control the supply chain. This is not a luxury. This is a requirement for a jurisdiction that wants to be an AI hub. The government should consider a hybrid approach. The city needs a small sovereign AI data center. It needs to be modest. A few hundred GPU clusters, enough for government applications and sensitive private workloads. It does not need to be a massive cloud. It needs to be a secure enclave. The mainland could continue to handle the heavy lifting. The Hong Kong enclave would handle the sensitive parts. The model is similar to a bank's use of a private cloud. The public cloud handles the bulk. The private cloud handles the core. This is a standard pattern. The second policy is to create a clear AI talent visa. This is a specific program with tax breaks. The program should be targeted at AI engineers and researchers. The program should not be a general talent program. The program should be specific to AI. The program should include a fast track. The program should include housing support. The program should be aggressive. The competition for AI talent is global. Singapore is offering a fast track. The UAE is offering tax-free. Hong Kong is offering... nothing. The city is losing the talent war by default. The third policy is to establish an AI audit framework. The 30 government projects should be audited. The audit should be public. The audit should include bias testing. The audit should include accuracy testing. The audit should include data privacy testing. The audit should be independent. This would build public trust. This would also set a standard for the private sector. This is the "audit first, invest later" principle. The code executes, not the promise. The fourth policy is to force AI companies to disclose their AI intensity. The stock exchange should require AI-related IPOs to disclose the percentage of revenue from AI. The disclosure should include the technical capabilities. The disclosure should include the R&D spending. The exchange should separate "core AI" from "AI-enabled". The 55% figure is too broad. The market needs to distinguish. This is a compliance mechanism. The market will reward the real companies. The market will punish the pretenders. The Takeaway: The Code Executes, Not the Promise The Hong Kong AI strategy is a distribution strategy. The city is a capital channel. The city is a testing ground. The city is a regional headquarters. This is a valid position. But the position is not permanent. The position is conditional. The condition is that the city must build the missing layers. The compute layer. The talent layer. The governance layer. If these layers are not built, the city will be a consumer of AI, not a hub. The capital will eventually flow to the places that have the infrastructure. The capital will flow to Singapore. The capital will flow to the mainland. The capital will flow to the places that can actually execute. The data point from the article is clear. The AI IPO share is 55%. This is the sign of the market's appetite. But appetite is not proof. The market is eating the story. The story needs to be substantiated. The 650 billion SME benefit is a potential. The potential is not a promise. The government projects are a signal. The signal is not a system. The code executes. The code is the system. The infrastructure is the code. The talent is the code. The governance is the code. Hong Kong has not yet written that code. The next 12 months will be the test. The 30 projects will produce results. The results will show whether the government can execute. The IPO pipeline will show whether the market is disciplined. The talent will show whether the city can attract. The compute will show. If the projects are hollow, if the IPO market is full of fake AI, if the talent does not come, if the compute remains absent, then the Hong Kong AI story will fade. The city will be a footnote. The city will be a consumer of AI, not a hub. The capital will go elsewhere. This is not a prediction of failure. This is a prediction of a clear, measurable set of criteria. The criteria are infrastructure, talent, and governance. These are the criteria that separate a hub from a pass-through. Hong Kong has the potential to be a hub. The potential is real. The potential is the combination of the rule of law, the financial system, and the international connectivity. These are real assets. These assets are hard to replicate. But these assets are not enough. The assets must be combined with the technical infrastructure. The technical infrastructure must be built. The building process must start now. Zero knowledge, infinite accountability. The Hong Kong AI strategy needs to be accountable for its infrastructure. The strategy needs to be accountable for its talent. The strategy needs to be accountable for its compute. The strategy needs to be accountable for its governance. The city must be a leader in AI governance. The city must be a leader in AI application. The city must be a leader in AI compliance. The city must not be a follower in AI infrastructure. The city must not be a follower in AI talent. The city must not be a follower in AI compute. The city must build the missing layers. The clock is ticking. The market is watching. The code will execute.