Artificial Intelligence in Hong Kong

Hong Kong’s AI posture is easier to read in March 2026 than it was a year earlier. The blueprint, the AI+ budget package, AISC, AIRDI, and the expanding regulator toolkit now describe a concrete operating model rather than a generic innovation narrative.^{[1][3][8][11][17]}

The February 26, 2026 Budget matters because it added new line items and institutions to a strategy that had already been outlined in the 2022 Innovation and Technology Development Blueprint. It highlighted 5,000 PFLOPS of overall computing power in Hong Kong, said around 30 AISS applications had been approved, created a Committee on AI+, and earmarked HK$100 million for AI in public services.^[1][8]

That makes Hong Kong’s AI story more specific than the old “innovation hub” pitch. The city is building a compact applied-AI system that links public compute, public-service demand, finance, and a governance layer that regulators are visibly willing to use.^{[5][6][8][11][14]}

The Greater Bay Area context is also becoming clearer. HKMA said in September 2025 that its Fintech Connect network already linked more than 200 banks and technology providers in Hong Kong and Qianhai, with 40 matching partnerships completed. That gives Hong Kong a concrete cross-boundary finance-and-technology lane, even if its role is more about capital, regulation, and commercialization than sheer industrial scale.^[13]

The best evidence that Hong Kong wants local AI capacity rather than only imported services is the combination of AISC, AISS, AI Lab, and AIRDI. Together they answer three practical questions: where the compute sits, who can afford it, and how research is supposed to turn into deployment.^[3][8][17]

AISC moved from plan to operating asset on December 9, 2024, when Cyberport said the first phase had commenced at 1,300 PFLOPS and would scale to 3,000 PFLOPS the following year. The February 2026 Budget then described Hong Kong’s overall computing power as 5,000 PFLOPS, which is a stronger signal than the earlier promise alone.^[8][17]

AISS is the difference between symbolic infrastructure and usable infrastructure. When the government opened the scheme on October 7, 2024, it said eligible users could generally receive subsidy support of up to 70% of AISC service list prices. By the February 2026 Budget cycle, the government said around 30 applications had already been approved.^[3][8]

The institutional layer matters too. AIRDI is meant to begin operating in the second half of 2026, and Cyberport’s own framing of AISC links the facility to model building, model-risk assessment, governance, and cross-industry application support rather than to raw hardware alone.^[8][17]

Hong Kong’s public-sector AI story is no longer only about principle-setting. The governance layer now sits alongside named tools, departmental targets, and budgeted implementation support.^{[5][6][8][12]}

The Digital Policy Office’s Ethical AI Framework and local generative-AI guideline matter because they are written for operational use, not just for rhetorical positioning. The documents focus on lifecycle management, information security, accountability, and accuracy in a form that departments and contractors can actually work from.^[5]

That governance layer is being matched with deployment. The government has been trialing HKPilot since 2024 for drafting, translation, and summarisation, and on February 7, 2026 the Digital Policy Office said the AI Efficacy Enhancement Team was steering bureaux and departments toward AI tooling for 100 public-administration procedures by end-2026.^[6][12]

The February 2026 Budget deepened the point by reserving HK$100 million for AI use in government departments. That is important because it puts money behind administrative modernization instead of treating it as an aspirational digital-policy theme.^[8]

Hong Kong’s financial AI story is now stronger than a simple adoption survey. The market has public evidence of interest from institutions, regulator-backed testing lanes, and a growing set of named use cases that matter in revenue and risk terms.^{[9][10][11][13][14]}

The HKIMR’s April 2025 report said roughly three quarters of surveyed institutions were already piloting or using GenAI. That tells us the demand side is real. What matters more now is that the official record increasingly names the underlying jobs to be done.^[9]

When HKMA and Cyberport launched the GenA.I. Sandbox in August 2024, they highlighted risk management, anti-fraud, customer services, and process re-engineering. By September 2025, HKMA was publicly pointing to examples from sandbox cohorts that included augmenting credit assessment, fraud detection through automated processing of unstructured data, more personalized customer-service workflows, and deepfake detection.^[13][14]

Insurance is moving in parallel. The Insurance Authority’s AI Cohort Programme asked core participating insurers to build Hong Kong-based centres of excellence, support AI talent development, and share knowledge with industry and regulators. The participating institutions are not anonymous start-ups but major insurers operating at regional scale.^[10]

The March 5, 2026 joint circular is what turns these examples into something more strategic. A cross-regulator sandbox is better aligned with how large financial groups actually work, especially when they span banking, insurance, pensions, and securities.^[11]

• Publicly named banking use cases now include credit assessment, anti-fraud work, personalized customer service, and process re-engineering.^[13][14]
• Insurance is being pushed toward more organized capability building through centres of excellence, talent programmes, and structured knowledge sharing.^[10]
• Hong Kong’s Qianhai-linked Fintech Connect network suggests the city’s regulated-finance AI story is also increasingly cross-boundary within the Greater Bay Area.^[13]

Talent policy in Hong Kong is no longer an abstract promise to “attract experts.” The official record now points to specific mechanisms that cover researchers, students, teachers, and civil-service users.^[6][15][16]

For advanced talent, the government’s February 2025 Legislative Council reply pointed to the Research Talent Hub and STEM Internship Scheme as existing channels for AI and innovation talent development. That matters because Hong Kong is trying to retain and train technical workers through operating programmes, not only visa rhetoric.^[16]

At the school level, the same reply said the Education Bureau had already launched the “Module on Artificial Intelligence for Junior Secondary Level” in 2023, while the bureau later launched the AI for Empowering Learning and Teaching Funding Programme with a general HK$500,000 block grant for publicly funded schools. The goal is not simply coding familiarity; it is to normalize AI use in everyday teaching and learning.^[15][16]

The February 2026 Digital Policy Office speech connected these efforts back to the wider state agenda, linking school modules, internship opportunities, and AI literacy to the same AI+ push that is also reshaping government workflows. Hong Kong’s talent strategy is therefore becoming more layered: flagship research, practical departmental capability, and much earlier pipeline formation.^[6][15][16]

By March 18, 2026, Hong Kong’s AI story is materially stronger than it was even a year ago. The city can now point to a live compute asset, a subsidy regime, an impending R&D institute, public-service deployment targets, and finance-sector testing lanes with named use cases.^{[3][6][8][11][13][17]}

That is enough to make Hong Kong credible as an applied-AI jurisdiction. It is especially credible in places where trust, supervision, multilingual documents, and institutional density matter more than sheer territory or domestic market size.^{[5][9][10][11][13]}

But the pressure point is obvious: throughput. If AISC, AISS, HKGAI, HKPilot, AIRDI, and the expanding financial sandboxes stay as parallel tracks, Hong Kong will still look busy without fully compounding. If the city starts producing a larger volume of deployed tools, approved projects, cross-boundary partnerships, and commercially durable case studies, then its compactness becomes a strategic strength rather than a scale disadvantage.^{[8][11][13][17]}