The Compute Corridor: How a Malaysian Border Town Became Asia’s Fastest Data Centre Market

Christian Brooks – SeaPRwire – Johor Bahru spent most of its modern history as Singapore’s industrial periphery — a place where palm oil estates and light manufacturing absorbed overflow labour from across the strait without ever becoming anything like a destination. Commuters knew it as a traffic jam and a weekend shopping trip. Nobody called it a digital infrastructure market.

Then a regulatory decision made eighty kilometres away, in a jurisdiction with no say over Malaysian planning policy, turned it into one of the fastest-growing data centre markets on earth.

This is the story of how that happened, what the numbers actually look like in 2026, where the real constraints now sit, and why the answer to “should we build there” is much less obvious than the growth charts suggest.

One moratorium redirected an entire industry

Between 2019 and 2022, Singapore stopped approving new data centres. The city-state cited the pressure energy-intensive facilities placed on already constrained land, water, and electricity systems. It was a reasonable decision for Singapore, and it did nothing whatsoever to reduce regional demand for compute.

Demand went looking. Johor was the only place close enough, cheap enough, and politically willing enough to absorb it.

What that meant concretely: Singapore did not just pause approvals, it created a backlog of capital already committed to the region with nowhere to land. Hyperscale cloud providers, colocation operators, and Chinese technology firms seeking a Southeast Asian foothold all hit the same constraint at the same time. Johor offered land at a fraction of Singapore’s price, a government actively courting investment rather than restricting it, and physical proximity close enough to preserve the low-latency connectivity that made Singapore valuable in the first place.

None of this happened because Malaysia out-competed Singapore in a contested bid. It happened because Singapore closed a door and Johor was standing on the other side of it with the lights on.

The timing mattered enormously. Johor’s expansion began as an overflow market for conventional cloud and enterprise workloads. Then ChatGPT launched in November 2022, and AI-specific demand layered directly on top of spillover demand rather than arriving as a separate wave. An open door, a pre-existing connectivity advantage, and an AI demand curve arriving at exactly the right moment is what produced growth rates no other Asia-Pacific market has matched.

The numbers, stated carefully

Growth figures in this market get quoted loosely, so here they are with their sources attached.

Both figures can be true at once: multiple consultancies track this market with different definitions of what counts as “live.” Note also that capacity figures measure electrical supply commissioned for IT load, not necessarily load that is currently drawing power — a distinction that matters enormously for anyone modelling revenue rather than headlines.

For context on regional scale, Southeast Asia now accounts for roughly half of all data centre capacity under construction across Asia-Pacific, up from 31 percent at the end of 2025. Malaysia leads with 1,039 MW under construction. Johor alone lifted its under-construction capacity to 602 MW and pushed its total pipeline to 3,088 MW in the first half of 2026 — and Johor retained top spot in Cushman & Wakefield’s regional maturity index across thirty markets.

Who is actually building it

This is not an organic market assembled by dozens of entrants. It was built fast by a small number of very large operators.

As of December 2024, Bridge Data Centers — a subsidiary of Bain Capital-backed ChinData — operated about 32 percent of Malaysia’s built IT capacity, with DayOne at 29 percent and AirTrunk at 13 percent. Three operators, roughly three-quarters of the market. That concentration is typical of hyperscale-led buildouts, and it has consequences: bargaining power, land assembly dynamics, and a market where a handful of decisions explain most observed outcomes.

The tenant list reads like a roll call of the AI economy. YTL Power’s Kulai campus, built in partnership with Nvidia, was among the first to break ground at scale in Johor rather than across the causeway. Microsoft has acquired land and planned a Johor cloud region. Oracle committed roughly USD 6.5 billion to a cloud region. Google put USD 2 billion into its first Malaysian data centre in 2024, followed by a sovereign cloud partnership with Dagang NeXchange. Alibaba Cloud opened a Johor cloud region in mid-2026 — its largest digital infrastructure footprint in Southeast Asia to date.

Capital markets noticed. Blackstone completed its A$24 billion acquisition of AirTrunk in December 2024. A KKR-led consortium with Singtel agreed in February 2026 to acquire the remaining 82 percent of ST Telemedia Global Data Centres at an enterprise value of S$13.8 billion — the largest transaction in Singapore digital-infrastructure history — taking KKR to 75 percent of a platform with 2.3 GW of design capacity across twelve markets.

The strategic sorting: Singapore keeps what it is good at

It is tempting to read this as Singapore losing its industry. That reading is wrong, and operators who believe it make bad siting decisions.

Singapore is deliberately not trying to be the region’s scale-out location. Since lifting the 2022 pause it has released capacity selectively and conditionally — the IMDA and EDB capacity call launched in December 2025 offered at least 200 MW conditioned on green energy commitments, with applications closing in March 2026. Rationale: latency-sensitive financial and enterprise workloads, hyperscaler regional headquarters, and the most mature regulatory frameworks stay on the island. Raw training capacity goes where power and land are cheap.

Microsoft’s USD 5.5 billion Singapore commitment, announced in early 2026 as part of a broader USD 6.5 billion Southeast Asian package, is the largest technology infrastructure investment the country has ever received. Amazon Web Services had already committed USD 9 billion in 2024. These are not exit signals.

The clearest illustration is Singtel Nxera’s DC Tuas: opened in early 2026 at 58 MW with a design PUE of 1.23–1.25, large-scale direct-to-chip liquid cooling, an integrated cable landing station, and more than 90 percent of capacity pre-committed before opening. That is a facility designed for governance and interconnect value, not for cheap megawatts.

The result is one functional market spanning three jurisdictions — Singapore supplying capital, connectivity, governance and demand; Johor supplying land and power; Batam emerging as the third leg with a 463 MW pipeline and the Firmus/Nvidia/DayOne AI factory campus targeting 360 MW and up to 170,000 GPUs through 2027–28. The Batam GPU farms get hard-wired into Singapore’s ecosystem by the Singtel–Telin INSICA cable with dedicated fibre pairs ready for service in late 2026.

Where it is actually breaking

Three constraints now bind this market, and they are not the ones a land-cost analysis would surface.

Power is first. Grid capacity has not kept pace with commitments. Tenaga Nasional Berhad has told several operators that new substation capacity in parts of the Kulai corridor will not be ready before 2028 — a timeline mismatch now shaping where inside Johor new campuses get sited, pushing some proposals toward Iskandar Puteri where transmission headroom is less committed, and further out to Pontian and Batu Pahat. Malaysia plans up to eight gigawatts of new gas-fired power by 2030 specifically to meet data centre demand, and has revived a nuclear programme with a 2031 target largely because of AI electricity consumption.

Water is second. Regulators estimate approved facilities could demand 675 million litres of water a day against roughly 142 million sustainably suppliable. That gap produced an approximately eighteen-month pause on new water-cooled approvals from late 2025. Any operator whose design assumes evaporative cooling needs to have a different answer ready.

Cost is third and least discussed. Malaysia’s July 2025 electricity tariff rebasing lifted data centre power costs by an estimated 10 to 14 percent. The cheap power that originally justified the cross-border move is not as cheap as it was, and nobody should model 2023 economics into a 2028 business case.

Underneath all three sits a political constraint that has no spreadsheet equivalent: social licence. Malaysia saw its first-ever data centre protest in February 2026, when residents in Gelang Patah rallied against a ZDATA site over dust, pollution, and water concerns. Calls for a broader moratorium grew through mid-2026. Meanwhile the federal government has effectively stopped approving new projects not demonstrably linked to AI — a selective freeze disclosed by the Prime Minister in February 2026 after roughly two years of quiet operation. Community consent is now part of the cost base.

The sustainability collision nobody has resolved

Here is the tension that will define the next five years. Fossil fuels generate 81 percent of Malaysia’s electricity, with solar and wind providing roughly 2 percent. A standard AI data centre consumes as much electricity as 100,000 households, according to the International Energy Agency. Meanwhile Malaysia has decarbonisation commitments and a MYR 2.1 billion sovereign AI cloud allocation designed to keep data and model training within national borders.

These ambitions are on a collision course that has not been resolved. It is not hypothetical — it is visible in the nuclear timeline, in the gas buildout, and in the fact that every new hyperscale approval now carries explicit green energy conditions.

For operators, this means the compliance question has shifted. Three years ago the question was whether you could get power. Now it is whether you can credibly document where your power comes from, because hyperscaler tenants increasingly cannot sign contracts that put their own Scope 2 targets at risk. Budget for that documentation as seriously as you budget for substations.

A practical siting framework

If you are evaluating this corridor, the following questions separate real feasibility studies from optimistic ones.

Power before price. Ask Tenaga — or read aloud the answer — what date your specific substation is energised, not when the park promises energisation. Site selection has become an energy question first and a property question second. One operator was told there was no power available before 2028. Cheap land with no energised substation is not cheap.

Cooling before capacity. With water-cooled approvals paused, know whether your design works in this climate within available water allocations before you commit to a site.

Compliance before construction. Malaysia’s new approval regime rewards demonstrable AI linkage and punishes generic capacity. Applications framed as general colocation are not being approved.

Community before concrete. The Gelang Patah protest was not an anomaly. Engage the adjacent community before commencement, not after complaints.

Latency honesty. Workloads that genuinely need to be within milliseconds of Singapore’s financial district belong in Singapore. Workloads that need cheap megawatts belong in Johor or Batam. Pretending one site does both is how operators end up paying Singapore rents for a training cluster.

Questions people actually ask

Is the boom sustainable? The demand side is real and AI-driven. The supply side is now constrained by grid, water, and political consent — the same three factors that constrained Singapore in 2019, which is a useful warning.

Will Singapore reclaim the capacity? Partly, selectively, and only for high-value low-latency workloads. It has explicitly stopped competing on raw scale.

Can we rely on the developer’s power timeline? Treat any energisation date that does not come from the utility as a marketing estimate. The single most common reason projects slip in this corridor is a gap between when a developer promised power and when the substation was genuinely live. Get the utility’s answer in writing, dated, before you sign anything that carries a delivery obligation.

Is it too late to enter? For generic speculative capacity, yes — that window closed with the AI-only approval policy. For differentiated capacity with secured power and a credible sustainability story, there is still room.

What about Batam? Worth watching closely. The GPU campus scale and the dedicated cable are serious commitments, and its proximity to Singapore mirrors Johor’s original logic. It is roughly eighteen months behind Johor on the same curve.

What is the realistic downside? A grid delay slipping your energisation by two years while your anchor tenant’s contract has a delivery date is the scenario that actually kills projects here, not land cost or permitting fees.

Cooling has become a siting decision

Cooling used to be an engineering footnote decided after the site was bought. In Johor in 2026 it is a gating constraint, because the eighteen-month pause on new water-cooled approvals removed the default option most designs assumed.

The arithmetic is stark: approved facilities could require 675 million litres of water per day against roughly 142 million sustainably suppliable. Even allowing for conservative estimates on the demand side, the gap is roughly fourfold. No amount of engineering tweaks closes a gap of that size; only changing the cooling paradigm does.

Practically, operators now choose among three paths, each with a distinct trade-off. Air-cooled designs at lower rack density reduce water to near zero but sacrifice the power density that AI training clusters require. Closed-loop liquid cooling with dry coolers preserves density at higher capital cost and some efficiency penalty in a tropical climate. The third path — the one everyone asks about first — is to wait out the pause, which is a bet that regulation loosens rather than a plan.

What makes this a genuine siting decision rather than a design detail is that the answer varies within Johor. Sedenak Tech Park alone hosts over half the state’s live capacity, and its water allocation situation differs from newer proposals directed at Pontian and Batu Phat further from Singapore’s border. Sites further from the causeway trade latency for utility headroom, and for training workloads that trade is usually worth making.

What locating here actually costs

Published land prices are the least informative number in any Johor feasibility study. The costs that decide projects sit elsewhere.

The tariff change deserves particular attention because it quietly eroded the original premise. The reason to cross the causeway was cheaper power. Power is still cheaper in Johor than in Singapore, but the margin narrowed by double digits in a single year, and there is no institutional commitment that it will narrow no further. Model both spreads, not just today’s.

How Malaysia sits in the regional picture

Johor’s position looks very different depending on which comparison you make, and picking the wrong comparator produces bad strategy.

Against Singapore, Malaysia wins on land, power availability, and cost, and loses on latency, governance maturity, and connectivity density. That complementarity is exactly why the cross-border model works — it is not a competition, it is one supply chain.

Against Thailand, the picture is less comfortable. Bangkok posted Asia-Pacific’s fastest data centre expansion with a 148 percent jump in under-construction capacity, representing the whole of Thailand’s 859 MW of activity. Thailand offers large domestic demand, a stable grid, and no comparable water controversy yet. Operators who assume Johor is the only Southeast Asian option are not looking at the same maps their competitors use.

Against Indonesia and Batam, proximity to Singapore matters more than national policy differences. Batam’s dedicated cable and GPU campus commitment mirror Johor’s original logic almost exactly, eighteen months later. If you are making a decision today, Batam deserves evaluation on the same shortlist — its disadvantage is timing, its advantage may be a cleaner slate on grid and water planning.

Glossary

• Energisation date — The date a substation is actually live and able to deliver contracted power, as opposed to the date a park or developer promises availability.
• IT load — The electrical capacity available to computing equipment, excluding cooling and other facility overhead.
• PUE (Power Usage Effectiveness) — Total facility energy divided by IT equipment energy. Lower is better; 1.23 is strong, 1.5 is mediocre.
• Hyperscaler — A cloud provider operating at very large scale, typically the anchor tenant in a wholesale facility.
• Colocation — A facility leasing space, power, and cooling to multiple tenants rather than to a single anchor.

Forward calls, and how they could be wrong

Directionally, the corridor consolidates. Malaysia is heading toward 2 GW of operational capacity by 2028, and the strategic geometry — Singapore for inference and governance, Johor and Batam for raw training — looks stable for at least five years.

Three things could break it. First, if power timelines slip further while AI capital demands faster deployment, capacity chases reliability into other markets such as Thailand, where Bangkok posted the region’s fastest expansion with a 148 percent jump. Second, if water politics escalate, evaporative cooling becomes untenable and retrofitting costs land on operators who assumed otherwise. Third, if data localisation requirements harden in ways that conflict with cross-border compute patterns, the single-market logic of the corridor fractures.

The single most useful thing to track is not headline capacity but substation energisation dates. That is the variable every operator complains about privately and almost nobody publishes.

The honest bottom line

Johor’s rise was not a development programme. It was the unplanned consequence of one jurisdiction’s constraint meeting another’s willingness, amplified by the largest compute demand shock in a generation.

For operators, the practical takeaway is unglamorous: this market rewards those who secure power, water, permits, and community consent in that order, and punishes those who buy land first and discover later that power arrives in 2028. The growth charts are genuinely extraordinary. They are also a lagging indicator — the decision that shaped this market was made in Singapore in 2019, and the decisions shaping the next five years are being made right now in substation planning offices and municipal council meetings that almost no international coverage mentions.