Malaysia is saying No the wrong types of Data Center
Malaysia has started pricing its geography like a strategic asset. For operators still in "watch and evaluate" mode, the window is closing.
On 24 February 2026, Prime Minister Anwar Ibrahim said something that most people glossed over. Malaysia had restricted new data centres unrelated to AI for nearly two years, deliberately, to protect the national power grid and water supply. A country that had spent years saying yes to anything with racks and a land title had started saying no.
That single decision is the key to understanding what Johor actually is right now.
The market still tells the old story. Johor is Singapore’s cheaper neighbour. The overflow yard. The place hyperscalers go when land across the Causeway gets too tight or too expensive. That story was useful once. It is no longer the real one.
What is emerging in Johor is not a spillover market. It is something far more specific: the physical compute annex of the Singapore system, purpose-selected for the age of AI. Land, power, water, network adjacency, and policy discipline are converging in one corridor at exactly the moment the world needs somewhere to put industrial-scale AI infrastructure. That combination is rarer than it looks.
The old edge was price. The new edge is scarcity, and who controls access to it.
The Strategic Re-Pricing of Infrastructure
For years, the pitch for Johor was simple: cheap land, room to expand, and enough infrastructure to attract hyperscalers fleeing Singapore’s constraints. That was sufficient in the first wave. It is not the real edge now.
The real edge is that Johor is becoming more valuable precisely because access is no longer indiscriminate. Malaysia has started to apply a scarcity filter. It may not use the phrase “AI compute reserve” in official language, but functionally that is where policy is drifting. If you are going to consume sovereign grid capacity and industrial water at scale, the state increasingly wants to know that the economic value density justifies it.
That is not a minor policy tweak. It is a strategic re-pricing of sovereign infrastructure.
There is a simpler way to understand what Malaysia has done. When Microsoft, Google, and AWS start competing for the same corridor, they have already done the analysis. The due diligence, the site selection, the infrastructure modelling. They know. And Putrajaya has read that conclusion directly from their behaviour. You do not get simultaneous commitment announcements from the three largest hyperscalers in the world by accident. That level of demand concentration is itself a signal, and Malaysia received it.
The moratorium on non-AI data centres is what happens next. It is not a conservation policy. It is not technocratic infrastructure planning. It is a government that has achieved information parity with the most sophisticated infrastructure operators on the planet, and is now pricing access accordingly. The seller finally understood the buyer’s urgency. The terms changed.
This matters because for years the hyperscalers held a structural advantage across Southeast Asia. They understood exactly how rare contiguous, gigawatt-scale, hyper-connected geography truly was, and their strategy was to secure as much of it as possible before host governments realised they were parting with a non-renewable strategic asset. That window has closed. Malaysia has stopped pricing its geography like a desperate emerging market and started pricing it like a chokepoint. The distinction is not rhetorical. It has direct consequences for who gets in, on what terms, and at what cost.
Why Geography Still Wins
There is a temptation to dismiss Johor’s physical advantages now that frontier AI training can be distributed across multiple data centres. NVIDIA has published work showing that large-model training can span multiple facilities, with its communications stack supporting cross-data-centre coordination and long-haul optimisation. The world has moved beyond the constraint that every serious training run must be confined to one building.
But that does not mean geography has stopped mattering. If anything, the more distributed the system becomes, the more valuable good geography becomes.
Frontier AI is not just about raw compute. It is about coordination costs, synchronisation overhead, latency predictability, and network topology. The more you stretch a training cluster across distance, the more you start paying taxes that do not show up in the original marketing deck. You can engineer around those penalties. NVIDIA and others are doing exactly that. But engineering around a penalty is not the same as eliminating it. Physical concentration still carries a premium, and that premium is rising, not falling.
Johor’s value is not that it defies physics. Its value is that it softens one of the hardest trade-offs in infrastructure: the choice between scale and adjacency.
Historically, hyperscalers had to pick one. Build at true scale in places with room to sprawl, or stay close to the connectivity, capital, and ecosystem density of a tier-one hub. Johor increasingly lets operators get a meaningful slice of both. TM and Singtel’s Nxera announced in 2024 an AI-ready campus in Iskandar Puteri, 16 km from Singapore, with an initial 64MW phase designed to scale to 200MW, built around liquid cooling and GPU-heavy workloads. Not generic colocation. Purpose-built AI compute, sitting next to one of Asia’s most connected cities.
“Spillover” is the wrong word for what is taking shape. Spillover implies passive overflow. This is more deliberate: Singapore’s network gravity extending into a geography that can absorb industrial-scale compute without the land, power, and water constraints that Singapore itself cannot solve.
Connectivity: Optionality, Not Sovereignty
Johor is not about to dethrone Singapore as the region’s submarine-cable capital. Singapore remains the dominant regional node. But Johor does not need to replace Singapore to become far more valuable than it was.
What Johor is gaining is direct cable optionality. Telekom Malaysia joined the Asia Link Cable system in 2023, with the cable landing station at TM Exchange Kuala Sedili in Johor, projected complete by 2025. TM has also tied the future CANDLE cable into the same connectivity story, with operations targeted for 2028. Johor is beginning to move beyond pure dependency on cross-border backhaul. Not into full connectivity sovereignty, but into something more strategically useful: redundancy, multiple routing paths, and a more direct role in international data flows.
The value of a compute corridor rises sharply once it stops looking like a dead-end estate and starts looking like a place with multiple exits.
The Geopolitical Layer: Neutrality With Caveats
Malaysia offers hyperscalers a neutrality premium, not geopolitical immunity. That distinction matters more than most commentary acknowledges.
Malaysia is not China. It is not a sanctions target. It is not politically frozen in ways that create deployment risk. In a world where AI infrastructure is increasingly entangled with export controls, alliance structures, and the security anxieties of great powers, that makes it genuinely attractive. For operators managing global infrastructure footprints, a jurisdiction that is politically stable, commercially predictable, and outside the most acute geopolitical fault lines commands a real premium.
But no serious operator should model Johor as sitting outside the Silicon Cold War. In July 2025, MITI imposed strategic-trade permit requirements on the export, transshipment and transit of high-performance AI chips of US origin. Read that carefully. Malaysia is not resisting Washington’s chip architecture. It is actively managing compliance with it. That is a meaningful signal about how Putrajaya intends to navigate the AI infrastructure era: not as a defiant non-aligned actor, but as a jurisdiction that understands the ruleset and is willing to enforce it.
For Western hyperscalers and their home offices, that is likely a feature rather than a bug. A neutral jurisdiction that manages compliance predictably is often more commercially attractive than a louder “independent” jurisdiction that creates ambiguity about whether deployed hardware will trigger regulatory exposure back home. The question operators should now be asking is not whether Malaysia’s neutrality is real. It largely is. The question is whether the chip permit regime will be administered with enough consistency and speed to support operational timelines. That is where the practical risk lives, not in the headline geopolitics.
Industrial Base: Why Malaysia Starts From Strength
If Johor were building an AI corridor inside a country with no hardware history, the execution risk would be far higher. But Johor is not rising in isolation. MIDA figures indicate Malaysia accounts for roughly 13% of global semiconductor assembly, testing and packaging volume, placing it among the world’s major semiconductor exporters.
The bulk of the deepest back-end semiconductor base remains elsewhere in the country, particularly in Penang. But what this industrial history gives Johor’s compute build-out is something that matters more than it sounds: a surrounding ecosystem that already understands the physical economy of chips, advanced electronics, and mission-critical reliability. Hyperscalers are not just buying parcels of land and utility connections. They are buying into a contractor base that can build to specification, an engineering talent pool that understands what these facilities actually require, and a regulatory environment that knows the difference between an AI data centre and a generic industrial shed.
Countries with no hardware muscle can still attract flashy announcements. Sustaining a high-reliability ecosystem is a different problem entirely. Malaysia starts from a stronger position than most of its regional competitors on this, and that advantage compounds as build volumes increase.
Second-Order Effects: What Happens Next
The first-order story is now reasonably well understood in infrastructure circles. Johor has land, power, water, adjacency, and a government that has achieved information parity with the operators it is hosting and is acting on it. The more useful question, for anyone using this analysis operationally, is what follows from that.
The window for competitive entry is narrowing. Malaysia’s shift to a scarcity model means the first-mover advantage in securing large, power-allocated land parcels is real and compressing. Operators still in “watch and evaluate” mode should understand that the evaluation period has a cost, and not just in price appreciation. Power allocation queues are already forming. The sites that can support 200MW+ campuses at reasonable total cost are a finite list. Some are already spoken for.
Competitor jurisdictions are being structurally disadvantaged. Jakarta has land but faces grid reliability constraints and water stress that limit genuine hyperscale ambition. Vietnam is politically more complex for Western operators following the tightening of chip controls. Thailand is building, but starts from a weaker industrial base and lower network density. Johor’s rise is not happening in a vacuum. It is partly happening because the alternatives have identifiable ceilings, and those ceilings matter more as AI infrastructure requirements become more demanding, not less.
The talent and supply chain question will surface sooner than most projections assume. The build phase attracts announcements. The operate phase exposes gaps. As Johor moves from greenfield construction toward live operations at scale, the pressure will shift to whether the local talent pool in AI infrastructure operations, cooling engineering, and high-density power management can grow fast enough to support what is being built. Malaysia’s semiconductor heritage helps here, but it is not a complete answer. Operators with long investment horizons should be modelling talent localisation strategies now, not after the facilities are live.
For Singapore-based operators and investors, Johor’s emergence is not a competitive threat. It is an extension. The corridor effect means Singapore’s position as the regional commercial, legal, and financial hub becomes more valuable as the physical compute base next door grows. The pairing strengthens both sides. The risk for Singapore-based players is not displacement. It is misjudging how quickly the JS-SEZ is transitioning from a cost play to a strategic infrastructure layer, and moving too late as a result.
What To Watch
Three signals worth tracking closely.
Power allocation policy. How Malaysia manages grid access requests over the next 18 months will reveal whether the scarcity model is being applied consistently or selectively. Inconsistency here would be the first serious crack in the thesis.
The chip permit regime in practice. The July 2025 MITI requirements are the policy. The operational question is turnaround times, interpretive consistency, and whether the regime creates predictable compliance pathways or bureaucratic friction. Watch for operator commentary on this through 2026.
Submarine cable progress. The ALC landing at Kuala Sedili and the CANDLE timeline are the milestones that will determine whether Johor’s connectivity story moves from optionality to genuine redundancy. Slippage here matters.
The old story was that Johor won because it was cheaper than Singapore.
The new story is more formidable. Johor may win because Malaysia figured out what the hyperscalers already knew, and stopped negotiating like a country that needed the deal more than the other side did.
Once that happens, a place stops being a spillover market. It starts becoming a reserve.
In the AI era, reserves matter more than excess. The operators who understand that earliest will find the best positions are already being taken.
Nasser Ismail
Founder, JS-SEZ Monitor
Former IRDA (Founding Team) · Former PTP Free Zone Leadership
Clear-eyed commentary on the Johor–Singapore SEZ without the corporate spin, ministry optimism, or developer gloss. Just the structural incentives, the policy logic, and what they mean for capital and outcomes.
Subscribe if you prefer uncomfortable clarity over comforting narratives.