Singapore vs. Johor Bahru: Southeast Asia's Data Center Rivalry
Singapore's moratorium is forcing a complete rethink of Southeast Asian data center strategy, with Johor Bahru emerging as the unlikely winner that's rewriting the rules of proximity and efficiency.
Everyone keeps asking when Singapore will lift its data center moratorium. That's the wrong question. The right question is whether it even matters anymore, because Johor Bahru just hit 98.9% occupancy and is building at densities that Singapore's constrained urban footprint simply cannot match.
We've been watching this shift accelerate since 2020, and what started as Singapore's overflow problem has become Southeast Asia's most interesting market dynamics story. Our dataset shows Singapore with 85 tracked facilities totaling roughly 1,222 MW, built over decades of careful planning and premium positioning. Johor Bahru has just 5 tracked facilities with approximately 235 MW, but here's the kicker: their average facility size runs at 46.9 MW compared to Singapore's 14.4 MW, and they're achieving a 1.34 PUE while Singapore averages 1.46.
The numbers tell a story that the press releases don't. This isn't just about Singapore running out of room. It's about a fundamental rewriting of what proximity means in the age of hyperscale AI workloads.
The Constraint Reality No One Talks About
The SIJORI Scale Gap
Singapore's total operational capacity vs Johor and Batam pipeline commitments as of mid-2025
DC Atlas proprietary data combined with publicly available industry reports, government publications, and market analyses (2024–2025).
Singapore's moratorium gets all the headlines, but the real constraint isn't policy, it's physics. Based on our tracked facilities, Singapore averages roughly 15,300 square feet per MW while Johor Bahru operates at approximately 920 square feet per MW. That's a DC Atlas-derived ratio of about 17:1 in space efficiency. When you're trying to deploy 100 to 200 MW AI clusters, that density advantage isn't a nice to have, it's make or break.
The land economics tell the deeper story here. Singapore operates under what we call the "premium density trap." Every square meter of data center space competes directly with residential towers, commercial development, and industrial facilities in one of the world's most expensive real estate markets. Land costs in Singapore's data center zones range from SGD 800 to 1,200 per square foot, while comparable industrial land in Johor's Iskandar region trades at MYR 25 to 40 per square foot. That's roughly a 20:1 cost differential that no amount of operational efficiency can overcome for space hungry workloads.
We've tracked every major hyperscaler move since 2020, and the pattern is clear. Oracle committed USD 6.5 billion to Johor, Google USD 2 billion, Microsoft USD 2.2 billion SIJORI Growth Triangle Market Analysis. These aren't backup plans anymore. They're primary strategies.
But here's what makes this particularly interesting from an infrastructure perspective: these hyperscale commitments aren't just about land costs. They reflect a recognition that modern AI workloads operate fundamentally differently than traditional enterprise computing. Training large language models requires sustained, parallel processing across thousands of GPUs. That means continuous power draw at unprecedented densities, massive cooling requirements, and networking architectures that prioritize east west traffic over north south connectivity. Singapore's urban constraints make deploying these architectures prohibitively complex, while Johor Bahru's greenfield sites can build purpose designed infrastructure from the ground up.
The causeway between Singapore and Johor Bahru is exactly one kilometer. The latency difference for most workloads is negligible thanks to multiple submarine cable landings that give Johor Bahru connectivity parity with Singapore. But the cost difference for land is 50%, and the regulatory approval time has dropped to 9 to 11 months using prefabricated modules from Chinese supply chains.
What's particularly compelling about the connectivity story is how it challenges traditional assumptions about geographic proximity. Singapore's advantage was always its position as the region's internet exchange hub, with direct connections to submarine cables linking Southeast Asia to the broader internet backbone. But modern hyperscale deployments increasingly prioritize bandwidth over latency, and bulk data transfers between regions can easily be routed through Johor Bahru facilities without meaningful performance impact. This means Johor Bahru gets Singapore's connectivity advantages without Singapore's space constraints.
The Growth Trajectories Tell Everything
Singapore's growth curve has flattened to what you'd expect from a mature, constrained market. The trajectory shows steady but limited expansion, with most new capacity coming from efficiency improvements and densification rather than greenfield development. Meanwhile, Johor Bahru's curve looks like a startup's hockey stick, going from essentially zero in 2020 to a projected dominance that will host 60% of Malaysia's total data center capacity by 2030.
The shape of these curves reveals something critical about market maturation that gets lost in discussions about individual projects. Singapore's flattening growth isn't a policy failure or regulatory misstep. It's the natural evolution of a market that has reached its carrying capacity given urban density constraints. Every additional megawatt in Singapore requires retiring something else, optimizing existing facilities, or accepting trade offs in other urban systems. This creates a natural ceiling that has little to do with demand and everything to do with physical limitations.
Contrast this with Johor Bahru's exponential trajectory, which reflects the advantages of building hyperscale infrastructure in purpose designed industrial zones. The Malaysian government designated the Iskandar region specifically for heavy industrial development, with power generation and transmission infrastructure scaled to support energy intensive manufacturing. This means data centers aren't competing with urban systems for power, water, or land use. They're building on infrastructure designed to support exactly these types of facilities.
What's particularly telling is the pipeline composition. Singapore's pipeline represents careful, incremental growth within existing constraints. Johor Bahru's pipeline represents a fundamental scaling up. Knight Frank reports aggregate supply nearly doubling to approximately 5.8 GW by Q2 2025, including 2.0 GW of new announcements.
The customer composition driving these pipelines tells an equally important story. Singapore's take up is dominated by financial services, enterprise software, and latency sensitive applications that justify premium pricing. Johor Bahru's take up is driven by social media platforms, content delivery networks, and AI training workloads that prioritize scale and cost efficiency over millisecond latency optimization. This isn't market cannibalization as much as market segmentation, with different infrastructure optimized for different use cases.
Singapore's Careful Calculation
Singapore's approach reflects the constraints of operating in one of the world's most land scarce environments. With 89% of capacity already operational and just 11% in the pipeline, every megawatt has to earn its real estate. The result is a market optimized for premium workloads where latency and reliability justify the premium.
But Singapore's constraint driven optimization has produced some fascinating innovations that other markets are now studying. Liquid cooling adoption in Singapore runs significantly higher than global averages, not because the climate demands it, but because space constraints make air cooling inefficient. Similarly, Singapore operators have pioneered edge optimization techniques that maximize compute density per rack, achieving power utilization effectiveness that would be unnecessary in markets with abundant space.
Amazon AWS operates 173 MW in Singapore, Google has 160 MW, and Meta runs 150 MW. These aren't distributed across massive campuses but concentrated in highly efficient, network dense facilities that maximize connectivity per square foot. It's a different model entirely from what we're seeing evolve in Johor Bahru.
The operational model in Singapore has evolved toward what we call "connectivity maximization." Rather than optimizing for raw compute capacity, Singapore facilities optimize for network density, with more submarine cable terminations, internet exchange connections, and cross connects per megawatt than almost any other market globally. This creates a network effect where Singapore's value proposition compounds over time. Every new connectivity relationship makes the existing infrastructure more valuable, even as space constraints limit absolute growth.
The facility size distribution in Singapore tells the constraint story perfectly. Sixty percent of facilities fall in the 5 to 20 MW range, reflecting the practical limits of urban deployment. Only 18% reach the 20 to 50 MW scale that's becoming standard for hyperscale deployments elsewhere.
This size distribution isn't accidental. It reflects deliberate urban planning that prioritizes mixed use development over single purpose industrial zones. Singapore's data centers exist within the urban fabric rather than separate from it, which creates operational efficiencies around workforce access and service delivery but imposes hard limits on facility scaling. The 85 facilities scattered across Singapore represent decades of careful integration with urban systems, from power grid connections to waste heat recovery programs that feed district cooling networks.
What's particularly interesting is how this constraint has driven innovation in modular facility design. Singapore operators have become experts at maximizing capacity within fixed footprints, using techniques like stacked cooling systems, vertical cable management, and space optimized backup power that other markets are now adopting. The constraint has become a competitive advantage in facility engineering, even as it limits absolute scale.
Johor Bahru's Hyperscale Bet
Singapore's Controlled Capacity Release
From moratorium to measured expansion — new data center capacity awarded under DC-CFA rounds
DC Atlas proprietary data combined with publicly available industry reports, government publications, and market analyses (2024–2025).
Johor Bahru's development pipeline shows a market built from the ground up for hyperscale. With 87% operational and 13% under construction, they're building fast and building big. The market reached 99% occupancy with just 1.1% vacancy in H1 2025 Johor Data Centers 99.9% Taken Up, creating the kind of supply constraint that would take years to develop organically.
The speed of development in Johor Bahru reflects advantages that go beyond land availability. Malaysian regulatory frameworks allow for parallel processing of environmental impact assessments, construction permits, and utility connections in ways that would be impossible in Singapore's more complex urban environment. This has compressed development timelines from the typical 18 to 24 months for greenfield data centers to 12 to 15 months for comparable facilities in Johor Bahru.
What's driving this isn't just Singapore overflow. It's the recognition that AI and machine learning workloads need space to breathe. When you're training large language models or running inference at scale, the 200 to 300 MW campus model makes operational sense. Johor Bahru can deliver that. Singapore fundamentally cannot, regardless of policy changes.
The AI workload angle is particularly compelling because it represents a fundamental shift in how we think about data center economics. Traditional enterprise workloads prioritize reliability and low latency over absolute scale, which plays to Singapore's strengths. But AI training requires massive parallel processing across thousands of accelerators, which demands enormous amounts of space, power, and cooling. A single AI training cluster can require 20 to 40 MW of continuous load, the equivalent of a medium sized Singapore facility. Johor Bahru can host multiple clusters within individual facilities, creating the kind of economies of scale that AI companies require.
The efficiency gains are real and measurable. Bridge Data Centres and DayOne collectively hold 58% of built capacity and 41% of under construction capacity, suggesting a market dominated by operators who understand hyperscale economics rather than traditional colocation players trying to adapt.
This market concentration reflects something important about how hyperscale infrastructure develops. Rather than the fragmented, multi tenant model that characterizes mature markets like Singapore, Johor Bahru is developing around purpose built, single tenant or wholesale facilities designed for specific hyperscale workloads. This allows for architectural optimizations that would be impossible in mixed use facilities, from custom cooling designs to specialized power delivery systems optimized for AI accelerator loads.
The SIJORI Reality
The Singapore Johor Riau Growth Triangle isn't just a policy framework anymore, it's an operational reality. We're tracking facilities that treat the causeway like an extension cord rather than an international border. Workforce mobility via the Rapid Transit System means Singapore's talent pool works in Johor Bahru facilities. Dark fiber connections mean Singapore's connectivity ecosystem extends seamlessly north.
The operational integration between Singapore and Johor Bahru has reached the point where some facilities operate as distributed campuses across the border. Network operations centers in Singapore monitor and manage compute resources in Johor Bahru in real time. Technical staff commute daily between facilities. Supply chains flow seamlessly across the causeway. This level of integration was impossible even five years ago but has become routine as both governments recognize the economic benefits of treating the region as a single operational zone.
The total planned capacity across the SIJORI triangle is estimated at roughly 5.8 GW based on DC Atlas tracking and public announcements, with the vast majority concentrated in Johor Bahru rather than the Singapore or Riau components. This represents a fundamental shift in how Southeast Asia's digital infrastructure is organized, with Singapore evolving into the high value network hub while Johor Bahru handles the compute heavy lifting.
What makes this integration particularly effective is the complementary nature of the specializations. Singapore excels at connectivity intensive workloads that require multiple network relationships, complex cross connects, and direct access to submarine cable infrastructure. Johor Bahru excels at compute intensive workloads that require massive amounts of space, power, and cooling but can operate effectively with more limited network diversity. The result is a regional ecosystem where workloads can be optimally placed based on their specific requirements rather than forced into one size fits all infrastructure.
The workforce integration aspect is particularly interesting because it challenges traditional assumptions about where technical talent needs to be located. Singapore's deep pool of data center engineers, network architects, and facility managers can now support operations across the broader SIJORI region without requiring relocation. This effectively extends Singapore's human capital advantages across a much larger geographic area, creating operational efficiencies that neither market could achieve in isolation.
Regional Context: Singapore's Shifting Position
Within the broader APAC context, Singapore ranks fourth by total capacity behind Sydney's 1,767 MW, Hyderabad's 1,544 MW, and Tokyo's 1,260 MW. This positioning reflects Singapore's role as a premium, connectivity focused market rather than a hyperscale destination. Meanwhile, Johor Bahru isn't even visible on this regional ranking yet, but the trajectory suggests that will change rapidly.
Singapore's position in the regional hierarchy is particularly interesting when you consider market maturation curves. Sydney and Tokyo represent mature markets with diverse workload portfolios, while Hyderabad represents India's hyperscale manufacturing approach. Singapore represents something different entirely: a premium, highly connected market that optimizes for network density over absolute scale. This creates a unique value proposition that becomes more valuable as regional connectivity requirements become more complex.
The power dynamics are particularly telling. Singapore's electricity costs have remained relatively stable but premium, while Johor Bahru faces 10 to 14% electricity cost increases as demand strains the grid Malaysia Data Center Market Report. Yet the take up continues, suggesting that even with rising power costs, the total cost of ownership for hyperscale deployments favors Johor Bahru's model.
What's particularly noteworthy about the power situation is how it reflects different infrastructure development philosophies. Singapore's power grid was designed for urban density and reliability, with multiple redundancy layers that create premium pricing but exceptional availability. Johor Bahru's industrial grid was designed for manufacturing loads, with lower costs but less redundancy. As data center demand has grown, Johor Bahru has had to invest in grid upgrades that are pushing power costs higher, but they're still starting from a significantly lower baseline than Singapore's urban premium.
The regional competitive dynamics are also shifting as other APAC markets respond to these developments. Indonesia's Batam Island is positioning itself as an alternative to Johor Bahru with similar cost advantages but different regulatory frameworks. Thailand's Eastern Economic Corridor is targeting hyperscale development with government incentives. Vietnam is opening its market to foreign data center investment. This means Johor Bahru's window of opportunity as the obvious Singapore alternative may be more limited than current growth rates suggest.
The Regulatory Arbitrage
Johor's Demand Profile: H1 2025
260 MW of new take-up at 98.9% occupancy — one of the lowest vacancy rates in Asia-Pacific
DC Atlas proprietary data combined with publicly available industry reports, government publications, and market analyses (2024–2025).
Malaysia approved 42 new data center projects in Johor during Q2 2025 alone, with RM 15 billion in committed investment. The acceleration in approvals reflects a government strategy that views data center development as economic development, not urban planning constraint. Singapore's approach prioritizes sustainability and urban density over raw capacity expansion Johor Fastest Growing South East Asian Data Centre Hub.
The regulatory environments reflect different priorities entirely. Singapore's moratorium isn't just about land use, it's about maintaining the city state's carefully calibrated balance between economic growth and livability. Johor Bahru's accelerated approvals reflect a state government that sees data center development as transformational economic opportunity.
But the regulatory arbitrage runs deeper than just approval processes. Malaysia's investment incentives for data centers include pioneer status tax exemptions, import duty waivers on equipment, and accelerated depreciation schedules that can reduce total project costs by 15 to 20%. Singapore offers different incentives focused on innovation and skills development rather than capital cost reduction. These different approaches reflect different economic development priorities, with Malaysia prioritizing rapid capital formation and Singapore prioritizing sustainable, high value economic activity.
The environmental regulatory frameworks also diverge significantly. Singapore requires comprehensive environmental impact assessments that consider urban air quality, waste heat impacts on urban systems, and water usage efficiency within a constrained supply system. Johor Bahru operates under industrial environmental standards that are less stringent for greenfield development in designated industrial zones. This creates a meaningful difference in project complexity and timeline that goes beyond simple approval processes.
What's particularly interesting is how these regulatory differences are driving innovation in different directions. Singapore's stringent requirements have pushed operators toward cutting edge efficiency technologies and urban integration solutions. Johor Bahru's more flexible framework has enabled rapid scaling and cost optimization. Both approaches are producing valuable innovations, but optimized for different market conditions and development priorities.
What This Means for the Next Five Years
We think the Singapore versus Johor Bahru dynamic settles into a complementary specialization rather than direct competition. Singapore remains the premium network hub for latency sensitive workloads and financial services. Johor Bahru becomes the hyperscale compute destination for AI training, content delivery, and workloads where scale matters more than milliseconds.
The specialization model makes sense when you consider how different workloads have different infrastructure requirements. High frequency trading, for example, requires proximity to financial markets and sub millisecond latency optimization that justifies Singapore's premium pricing. AI model training, by contrast, benefits from massive parallel processing capacity and can tolerate higher latency in exchange for lower costs and greater scale. Real time gaming falls somewhere in between, requiring good connectivity but benefiting from the cost efficiencies of hyperscale infrastructure.
The challenge for operators is that this isn't a temporary arbitrage opportunity. It's a structural shift that requires different operational models, different customer relationships, and different approaches to infrastructure planning. Companies that treat Johor Bahru as "Singapore lite" miss the point entirely.
This structural shift has implications for how data center operators think about regional strategies. Rather than building similar facilities in multiple markets to serve similar customers, the Singapore Johor Bahru dynamic suggests a model where different facilities serve different workloads optimally. This requires more sophisticated customer segmentation, more specialized facility designs, and more complex operational coordination across borders.
The real opportunity is in understanding that Southeast Asia's digital infrastructure is evolving from city state centralization to regional distribution. Singapore's constraints aren't a bug, they're a feature that's forcing more efficient regional allocation of compute resources. Johor Bahru's rapid growth isn't just overflow, it's the emergence of a purpose built hyperscale ecosystem.
Looking ahead, we expect this model to influence how other regional markets develop. Rather than every major city trying to build comprehensive data center ecosystems, we're likely to see more specialization based on natural advantages. This could mean connectivity hubs in strategic locations, hyperscale campuses in areas with abundant land and power, and edge facilities in population centers, all connected by high capacity networks that make the geographic distribution transparent to end users.
Our take is that by 2030, asking whether to build in Singapore or Johor Bahru will sound as dated as asking whether to build in Manhattan or New Jersey. The answer increasingly is both, but for completely different reasons.
The evolution toward regional specialization also suggests that traditional metrics for comparing data center markets may become less relevant. Rather than ranking markets by total capacity or facility count, the more important question becomes how well each market serves its specialized use cases. Singapore's value isn't diminished by having less total capacity than Johor Bahru if it provides superior connectivity and reliability for the workloads that require those characteristics. Similarly, Johor Bahru's value proposition doesn't depend on matching Singapore's network density if it can provide the scale and cost efficiency that hyperscale workloads require.