APAC’s data center landscape is diverse, with each economy showcasing unique opportunities and strengths. To better understand what lies ahead for the region’s digital infrastructure industry, we caught up with Mikhail Jaura, JLL’s newly appointed Head of Data Center Research for Asia Pacific.
Q-1) In your opinion, what is the single most exciting aspect of APAC’s cloud and data center market?
Mikhail Jaura: What excites me most is that APAC is no longer one story. Five years ago the regional narrative was essentially Singapore, Tokyo and Hong Kong. Today we have a genuinely multi-speed region. Mature hubs are reinventing themselves around AI-ready, high-density capacity. Markets like Johor, Batam, Mumbai, Chennai and Seoul’s periphery are being built at hyperscale. A further tier of countries is putting in its first serious cloud infrastructure. Each market has a different mix of power, land, policy and demand. For investors and occupiers, that diversity is both an opportunity and a challenge.
Q-2) Despite exorbitant costs, Japan and Hong Kong remain strong digital infrastructure markets. But do you foresee a slowdown or a correction in the next few years due to other factors like labour shortages, difficulty in acquiring powered land, sustainability concerns etc.?
Mikhail Jaura: Demand fundamentals in both markets remain resilient. In Japan, demand is strong, driven by cloud and AI adoption across a large domestic enterprise base, and supply is not keeping pace. In Hong Kong, sentiment has improved this year on the back of demand from Mainland Chinese hyperscalers, as the market positions itself, through the Northern Metropolis, as the connectivity gateway to the Greater Bay Area.
The constraints are real, though. In Japan, grid connection timelines reaching ten years in Greater Tokyo and Osaka, construction cost inflation, and a construction labour market squeezed by overtime caps since 2024 and a shrinking workforce are lengthening delivery schedules. That is pushing development toward secondary regions such as Hokkaido and Kyushu, supported by government efforts to co-locate power and digital infrastructure.
In Hong Kong, powered land is the binding constraint, which makes government land allocation in the Northern Metropolis important. Absorption slowed in 2025 but has since rebounded, driven by AI demand from Mainland Chinese technology firms. Little new supply arrives before Sandy Ridge, which is over 200 MW, and targeting phased operations from mid-2027, and Goodman’s HKG10 in early 2028. The bigger risks are execution and concentration: projects slipping on power, demand sitting with a few large tenants, and exposure to changes in chip trade rules.
Q-3) China has always marched to the beat of its own drum. How do you see the AI story unfolding in China, and how will it impact the wider Asian data center landscape?
Mikhail Jaura: China is building a largely self-contained AI stack. That stack spans domestic accelerators, its own models, and national compute programmes such as “East Data, West Computing”. Launched in 2022, that programme designates eight national computing hubs and ten data center clusters, moving new capacity toward western provinces where land, renewable power and cooling are cheaper.
Chinese labs now produce many of the most widely used open-weight models. Alibaba’s Qwen family recorded more than three billion downloads on Hugging Face, and many large Chinese releases carry permissive licences that allow commercial use.
Further afield, Chinese cloud and internet platforms are among the groups expanding in Southeast Asia, alongside US and regional operators, particularly in Malaysia and Thailand. Chinese cloud providers are also building out regional capacity. For host markets, this investment sits alongside large commitments from US hyperscalers. Governments are welcoming both while working within global trade rules; last year, for example, Malaysia introduced permits for the export and transshipment of US-made AI chips.
Q-4) How do you see the hyperscale ecosystem evolving over the next few years in different APAC markets, especially South Korea, Malaysia, and Indonesia?
Mikhail Jaura: South Korea is moving from a Seoul-centric market to national AI infrastructure. Grid capacity in the capital region is increasingly constrained, and new transmission lines and local approvals are taking longer to secure. Policy is pushing new capacity south. The AIDC Special Act, passed in May 2026 and scheduled to take effect in Q1 2027, will fast-track permitting outside Seoul, and regional electricity pricing proposed in August 2026 would offer discounts of up to about 10% in the south.
Malaysia’s Johor capacity has grown more than 50-fold since 2020, with a further 1,800 MW under construction. The next phase is about quality and location. Greater Kuala Lumpur is becoming a cloud market in its own right: Microsoft’s Malaysia West region went live there in 2025, AWS operates a region, and Google is building its first Malaysian data center at Elmina. Malaysia approved over RM 30 billion of data center and cloud investment across dozens of projects in Q1 2026 alone. Power is the gating factor.
Indonesia has a large digital economy, data localisation rules and hyperscaler cloud regions in Greater Jakarta, and Batam offers a cross-border option alongside Johor for demand spilling over from Singapore. Grid reliability, renewable availability and regulatory clarity will decide how fast it scales. Earlier this year, CoreWeave entered the APAC market by announcing three new data center facilities in Indonesia, delivering 360 megawatts of contracted IT power.
Q-5) What exciting development trends have you observed regarding enterprise demand in APAC? What are some secondary and emerging markets you are watching keenly?
Mikhail Jaura: Enterprise demand is maturing toward hybrid and multi-cloud architectures. Regulated, latency-sensitive and AI inference workloads are staying in colocation near cloud on-ramps. Data residency is a major driver.
In India, vacancy fell to a record 2.8% in H1 2026, with Mumbai’s growth driven mainly by domestic demand, as residency rules bring workloads onshore. In Australia, sovereign hosting requirements keep government and regulated workloads in the country.
The markets I’m watching most closely include Mumbai, which is expected to take 35 percent of India’s 4.4 GW pipeline by 2029. Also, Chennai where colocation power costs are around 8.4 US cents per kWh, against 18.1 cents in Singapore. In Australia, I’m watching Western Sydney, Melbourne and Australia’s regional markets keenly. Brisbane, Perth, South Australia and Tasmania are attracting AI campuses where power is easier to secure.
I also have my eye on Greater Kuala Lumpur which is maturing into a cloud region as Johor takes the large-scale AI build. Then there are Japan’s GX regions: Hokkaido, Kyushu and Toyama. And let’s not forget Hong Kong, which is an established market that appears on the cusp of a rebound, as it repositions as the connectivity gateway to the Greater Bay Area, whilst still holding a strategic position by servicing enterprise demand from multinational corporations.
Q-6) While operators are focusing on acquiring AI chips, do you think they are also planning for their entire lifecycle? What happens after their accounting life ends?
Refresh cycles have compressed sharply. AI accelerators that used to be refreshed roughly every seven years are now refreshed in as little as 13 months. That doesn’t mean older chips are stranded. The newest chips go to frontier model training, and older generations shift to inference and fine-tuning, where cost per token matters more than peak speed.
Inference demand is growing fast enough to keep them booked. Earlier-generation accelerators are still in active service years after launch, which suggests their useful life can run longer than headline refresh cycles imply. For real estate, the pressure point is the building. A facility typically has 18 to 36 months before it needs a major power or cooling upgrade to host the next generation, so the priority is designing facilities flexible enough to support several chip generations over their lifetime.
Q-7) The “power struggle” is real! In your opinion, are captive power generation units going to become more common? If so, will we see more on-site generation and dedicated direct supply from such units, or will grid offsets become more popular?
Mikhail Jaura: I think captive and dedicated power will become more common, but policy is already setting the model market by market. For example, in Australia, new national and state frameworks may require hyperscale and AI operators to underwrite new renewable generation and storage at least equal to their added load, and to pay for the associated network upgrades. Meanwhile, in Japan, GX subsidies covering up to 50 percent of capex are available only to projects running entirely on decarbonised local power, effectively tying data centers to nearby generation.
In South Korea, regional tariffs and grid-impact assessments are steering large loads to where generation already sits. While in Singapore, the 700 MW low-carbon data center park on Jurong Island links new capacity allocations to low-carbon energy pathways. Then there’s India, where group captive and open-access renewable models are already well established.
The dominant model will be hybrid. The grid stays the backbone, supplemented by dedicated renewables, on-site generation for speed, and storage. Grid offsets still matter for ESG reporting, but they don’t solve physical power availability, and speed-to-power is now the deciding site selection factor.
Q-8) How is water access impacting leasing choices, given protests over data center water consumption in water-stressed markets?
Mikhail Jaura: Water has moved from an ESG footnote to a permitting criterion. For example, Johor has enforced tighter controls that restrict development of Tier 1 and Tier 2 facilities. In Australia, new frameworks restrict the use of potable water for cooling and require recycled or alternative sources. Developers are expected to fund the dedicated water infrastructure their projects need. Meanwhile in Thailand, water is one of the areas being reviewed by the subcommittees drafting national standards during the nation’s temporary pause on data center projects.
For leasing, hyperscale occupiers now ask detailed questions about cooling design, water usage effectiveness (WUE) and water sourcing at the RFP stage. There is a trade-off: moving away from evaporative cooling saves water but can increase power use. The shift to direct-to-chip liquid cooling for AI generally helps on both counts, particularly in closed-loop systems that recirculate the same coolant and need little or no make-up water. That is still not widely understood outside the industry. Sites that depend on contested municipal water will face longer approvals, while those with access to recycled or industrial water are better placed.
Water is also part of a wider question of social licence. Community pushback on data centers has become a contentious issue in the US, and that scrutiny is now reaching APAC.