The datacenter industry will emit 2.5Bn tonnes of greenhouse gas emissions between now and 2030, three times more than if generative AI had not been developed. The industry accounts for about 2% of global emissions, similar to the airline industry today, and will more than triple to 7% by 2030. Inaction will put net-zero commitments made by governments, including that of Hong Kong, at risk.
Hong Kong is the world’s 4th biggest datacenter market. Yet, we do not have a clear plan to address sustainability as it continues to expand infrastructure to support AI. This is a pressing need waiting to be addressed.
UBS summarize AI’s value chain into three layers:
- Enabler: providers of infrastructure including semiconductors, datacenters, cloud providers and power supply
- Intelligence: technology companies that transform computing and energy resources into intelligence by developing large language models and datasets
- Application: end-users that embed tools from the intelligence level into applications
Hong Kong has a larger influence on the enabler and application levels as stakeholders at these levels are local. Accordingly, we look at approaches that influence datacenters and power supply (enabler), and end-users (application).
1 – Import renewable energy from mainland
Hong Kong has less than 1% of electricity that comes from renewable sources, compared to 4.4% in Singapore. Whilst Hong Kong and Singapore share challenges of a city such as dense population and limited land resources to self-generate sufficient renewable energy, Hong Kong benefits massively from bordering mainland China but is still lagging behind. In 2022, Guangdong received 200Bn kWh from western regions of China, more than 80% from renewable energy. This volume dwarfs the 15Bn kWh of electricity generated by the Daya Bay Nuclear Power Station in Guangdong that makes up 25% of Hong Kong’s fuel mix. As Hong Kong has an existing mechanism to import nuclear energy from mainland China, importing renewable energy that is already available in Guangdong is worth exploring.
2 – Mandate sustainability standards
Other countries have set mandatory sustainability standards in datacenters that enforce energy efficiency requirements. Singapore introduced the Green Mark DC Platinum Certification requirement for all new datacenter facilities starting 2023. In Australia, datacenters hosting federal agency workloads must achieve a five-star rating from the National Australian Built Environment Rating System (NABERS) starting mid-2025.
Hong Kong has its own certification, the Building Environmental Assessment Method (BEAM), but only adopted on a voluntary basis. Research conducted by Jones LaSalle indicated only 5% of Hong Kong datacenter operators prioritize sustainability, lower than other markets in Asia Pacific, “due to no stringent government policy on decarbonization”. As of October 2024, only 5 datacenters are assessed under BEAM, out of 122 datacenters in Hong Kong. A sustainability mandate for datacenters, consistent with international practice, is recommended for Hong Kong.
3 – Funding for AI-led decarbonization
AI could offset its emissions impact from its own workloads and achieve a net-zero outcome. BCG estimated that using AI for climate control can reduce 5-10% of total GHG emissions in 2030. There is a need to translate theory into practice but greenfield investment uncertainty hinders incubation. Government funding can lower investment risks. The UK government, for example, committed GBP4m funding into AI projects aimed at helping industries cut carbon emissions. One of the projects leverages AI to analyze weather patterns and optimize solar energy generation. The technology was subsequently adopted by the UK energy grid.
4 – Transparency of carbon emission of AI models
There is learning from other carbon-emitting industries, such as air travel. Studies have shown that publicizing air travel emissions data increases pro-environmental intentions of passengers on the demand side and rewards airlines on the supply side that offer fuel-efficient flights. In real life, the European Union launched an Environmental Labelling Scheme that informs passengers of the environmental impact of different flights.
A similar strategy can be explored for the use of AI. Just like comparing flights of different airlines, users could compare performance of foundational AI models. If emissions can be disclosed for each flight, emissions could be disclosed for each AI query. “Societal pressure may be helpful to encourage companies and research labs to publish the carbon footprints of their AI models… In the future, perhaps consumers could even use this information to choose a “greener” chatbot”.
Conclusion
Practices from abroad and other industries have provided good reference cases for Hong Kong. Importing renewable energy from the mainland is feasible as Guangdong has set a precedent. Mandating sustainability standards in datacenters is consistent with international practice. Incubating AI technologies for decarbonization and improving awareness of the carbon footprint of AI are approaches for longer-term outcomes. We look forward to more conversations for Hong Kong in pursuing a balanced approach in pursuing both AI and net-zero.
This article is written by a holder of the EFFAS Certified Environmental, Social, and Governance Analyst (CESGA). CESGA is a globally recognised qualification whose prominence continues to grow worldwide. CESGA has recently achieved a significant milestone as the first programme accredited by European standard setter EFRAG for compliance with the ESRS sustainability disclosure requirements in the EU (mandatory from 2025). For enrolment details, please visit https://bit.ly/40chuOR .
