Edition 20 asked us to look past the contracted rent and into the structure underneath it. Edition 21 takes the same underwriting instinct somewhere else: if the cheaper asset to build is not automatically the better asset to own, where does the saving actually go?
There is a market shorthand for this in APAC right now. JPMorgan's Malaysia research frames it in a line — Singapore optimises every MW; Malaysia captures the next MW. Read it as framing, not proof: a neat description of the split, not a measured claim that one market's owners out-earn the other's.
Because the split it describes captures a real market tension — and it hides the whole problem. Capturing the next MW is a capacity outcome. Capturing the economics on that MW — the pricing, the margin, the return — is an investment outcome. They are not the same thing, and this edition is about the gap between them. Not where capacity goes next. Who keeps the economics after the cost advantage has passed through the stack.
1 The deal
Start with the number that makes the cheaper market look obvious. Cushman & Wakefield's 2026 APAC construction guide shows how wide the dispersion runs. On a like-for-like mid-specification basis, Singapore is around US$14.4m/MW against roughly US$9.6m/MW in Malaysia. At the high end of each market's specification range, the comparison is about US$17.9m/MW versus US$12.0m/MW. Across APAC as a whole, costs range from roughly US$7.9m/MW in Taiwan to US$19.2m/MW in Japan — about a 2.4-times spread. JPMorgan later used the shorthand of roughly US$12m/MW for Singapore against US$7m/MW for Malaysia. These are reference points within broader cost ranges, not directly comparable project quotes. The dispersion is the finding, not the ratio.
Power costs show similarly wide dispersion. Wood Mackenzie put the annual electricity bill for a typical hyperscale facility at roughly US$123m a year in Singapore against about US$42m in Indonesia — a threefold difference on a major operating input.
Then Singapore did something that complicates the simple cost story. On 21 August 2026, EDB and IMDA provisionally allocated another 200 MW of data-centre capacity — 50 MW each to Digital Realty, Equinix, Keppel and STT GDC — with the supported facilities to be located on Jurong Island. The four proposals were selected from more than twenty submissions. EDB and IMDA have also said they will review the need for another data-centre call in eighteen to twenty-four months.
But this is not simply 200 MW of conventional capacity entering an expensive market. DC-CFA2 sets a programme-wide technical and sustainability baseline, including Green Mark for Data Centres 2024 Platinum and PUE of 1.25 or better, while the four winning proposals committed to power more than half their capacity through eligible green-energy pathways and to deploy advanced efficiency technologies such as liquid cooling. So comparing those megawatts with cheaper regional capacity purely on construction cost may already be asking the wrong question.
Which is the puzzle the rest of this edition addresses. If the cost gap between an expensive market and a cheaper one is this large, why doesn't the owner of the cheaper asset simply keep the difference? And why is capital still competing to build the expensive one? A cost advantage sitting at a location is not the same as a cost advantage sitting in an owner's cash flow. Almost everything between the two is contestable.
2 The engineering read
A cost saving is not free money waiting in the P&L. It is a number that has to survive a contract, a competitor and a customer before any of it reaches equity. But before tracing who captures the saving, there is an earlier question: are the megawatts being compared actually economically equivalent?
The MW itself is not a standard unit. A megawatt that can be energised reliably, cooled at higher rack densities and delivered to customer specification is economically different from one that exists only as nominal capacity. The owner is not simply delivering electrical capacity: power architecture, redundancy, cooling design, efficiency and the operator's commissioning capability help determine what workloads a facility can support, and how reliably. There is another layer: energy and carbon qualification. For customers carrying their own renewable-energy or decarbonisation requirements, a technically capable MW may still be commercially unsuitable if the operator cannot offer an acceptable energy pathway. So the transmission chain is broader than engineering alone:
Higher specification can therefore be economically rational even when it raises upfront capital — but different investments solve different problems, and none creates value merely by being installed. Liquid cooling can support higher-density compute. Power-system design and storage can affect resilience, grid interaction and operating flexibility. Green-energy procurement can affect the carbon characteristics and customer eligibility of the delivered capacity. A battery system matters to the owner only to the extent its incremental cost produces something economically useful. Greener power becomes an advantage only if it broadens the addressable customer pool, strengthens utilisation or contract durability, or protects pricing sufficiently to justify its cost. Engineering and sustainability capability are not automatically a return premium. They have to transmit into a cash-flow advantage. The relevant issue is not which technology or label sounds most advanced — it is what the capability allows the MW to do, which customers it allows the MW to serve, what it costs, and who captures the resulting economics.
Who the saving legally belongs to
The first commercial fork is contractual, and it decides much of the outcome. S&P Global describes wholesale and hyperscale data-centre contracts as typically triple-net or modified-gross structures in which tenants reimburse electricity and other operating costs, so electricity-price fluctuations are generally passed through to customers. Retail colocation more often leaves those fluctuations with the operator, subject to whatever escalation rights the contract provides. Same tariff advantage, different economic owner. The contract decides who captures it.
What competition does to what's left
Whatever the contract leaves with the owner, the market gets a second vote. A cheaper cost base in a market racing to fill capacity is an invitation to compete on price. The saving that survives the lease can still be handed to the customer through lower pricing to win the lease-up — not because the operator is generous, but because the next operator with a similar cost base can compete for the same customer. And the advantage itself can move: CBRE notes that after rapid expansion in Johor, power infrastructure has struggled to keep pace with investment, and expects completion delays over the next 24–36 months, with operators beginning to explore nearby markets such as Thailand. Capital does not simply exploit a location advantage; at sufficient scale, it can begin to compress it. A cost advantage everyone in a market shares is not an advantage; it is the new baseline.
Why the expensive market can still hold price
If low cost were enough to create an owner moat, cheaper markets should dominate not just capacity growth but pricing. The evidence is more complicated. Wood Mackenzie notes that despite carrying very high electricity costs, Singapore remains a favoured data-centre market, with grid reliability one reason operators and customers continue to value the location. CBRE has Singapore holding the highest colocation pricing in Asia-Pacific, roughly US$330–475 per kW per month, with vacancy around 2%, while Johor and Batam gain traction through scalable land and near-term power availability. The emerging pattern is that lower-cost expansion markets can gain capacity traction while constrained hubs such as Singapore retain stronger pricing — but whether that pricing premium ultimately produces the better owner return is precisely what the public evidence does not settle.
What Jurong Island is — and isn't
The new Singapore capacity makes the comparability problem concrete. DC-CFA2 sets a programme-wide baseline — Green Mark DC 2024 Platinum and PUE of 1.25 or better, with the four winners committing to more than half their capacity from eligible green-energy pathways and to advanced efficiency technologies. Keppel's 50 MW DC SGP 11 goes further on disclosed project targets, with PUE below 1.2 and WUE around 1.6 across liquid, district and seawater cooling. (Those PUE and WUE figures are Keppel's own project disclosures and should not be attributed to all four awards.) Jurong Island adds a broader energy-system option set — JTC, Keppel and EMA are studying a microgrid testbed integrating renewables, battery storage and smart-grid technologies, while JTC and NUS are separately studying a sustainable tropical data-centre testbed and microgrid, and the island hosts utility-scale battery storage and hydrogen-compatible generation.
None of these should be read as the disclosed dedicated supply architecture of any DC-CFA2 facility; they describe the surrounding ecosystem. "Not equivalent" is a comparability warning, not the answer: a more expensive, more differentiated megawatt is not automatically a better-returning one. The additional expenditure has to earn its way through the economics — denser workloads, a broader customer set, faster commissioning, better reliability, faster fill or sustained pricing. Differentiation can explain why two MW should not be compared on construction cost alone. It cannot, by itself, prove which MW is the better investment.
The seam worth marking. The evidence supports a narrower conclusion than "Singapore beats Johor." Lower-cost expansion markets are gaining capacity traction, while constrained hubs such as Singapore sustain stronger pricing. That this reliably produces higher owner IRRs in Singapore than in Johor is not something the public disclosures establish — and this edition does not claim it. The argument is non-equivalence at three levels: a MW of nominal capacity is not necessarily equivalent to a MW of technically usable, reliable capacity; a technically usable MW is not necessarily equally eligible for every customer's energy and carbon requirements; and a location's cost advantage is not automatically an owner's return advantage. Which one you are buying — and what you are actually paying for — is the underwriting question.
3 The capital allocation read
A location-level saving reaches equity only through a chain: cost base → contract → competition → utilisation and customer mix → operating margin → sustaining capital → equity cash flow. The lease structure can be decisive — wholesale/hyperscale pass-through may hand the power saving to the tenant; retail colocation can leave more of that exposure with the operator. A useful reference frame, not a formula: owner value capture depends on the size of the cost advantage, the share retained after contractual and competitive pass-through, and whether the resulting proposition drives enough utilisation and pricing to create cash flow. "Cheaper market" is a statement about cost; "better-owned asset" is a statement about retained cash flow. Do not read the first as the second.
Return profiles differ by operating model before location is even considered. Retail colocation tends to command higher unit economics in exchange for service complexity, smaller commitments and greater operating responsibility; wholesale/hyperscale accepts lower unit pricing for scale, duration and different risk characteristics. McKinsey-modelled ranges put retail-colocation equity IRRs around 20–25% against roughly 13–18% for wholesale — modelled, largely global ranges, not realised APAC returns, reflecting both weaker-bargaining fragmented demand and higher lease-up risk. Treat it as mechanism, not scoreboard. Two owners in the same cheap market, running different models, do not necessarily earn the same return.
Malaysia's July 2025 tariff redesign is a useful example. Wood Mackenzie estimates that qualified Peninsular Malaysia hyperscalers under the ultra-high-voltage time-of-use structure pay around 8% more than a typical industrial user. Reported estimates of the impact on large data-centre users have been around 10–14%, based on industry and government figures rather than a single regulator-stated figure. The sharper point is not "power became more expensive, therefore owner returns fell." It is who bore the change under the contract — where power was passed through, the tenant absorbs the exposure; where the operator carries it, the owner does. A cheap tariff is not automatically retained margin, and a tariff increase is not automatically an owner loss.
Here is the DC-CFA2 puzzle made concrete. Keppel and its co-investors expect to commit more than S$1 billion to the 50 MW DC SGP 11 development on Jurong Island — expensive capacity deliberately pursued while lower-cost regional capacity exists across the strait. Treat that >S$1 billion number as project investment, not a construction cost per MW: it carries project-specific energy, cooling, sustainability and development scope, and does not share a denominator with an ordinary benchmark. Then ask, in order: what kind of MW does the additional capital create; which customers can that MW serve, if power, cooling, density and carbon characteristics affect who can place workloads there; and which of those attributes actually changes pricing, utilisation, customer access, operating cost or revenue certainty enough to compensate for the capital. Capability must show up somewhere measurable — faster delivery, broader eligibility, higher utilisation, lower operating cost, stronger pricing or greater revenue reliability — before it deserves value in the underwriting. The same discipline applies to storage, microgrids and generation in the surrounding ecosystem: presence is not a premium. Competition for the expensive MW is evidence operators see value worth pursuing. It is not proof the eventual owner return will justify the capital. That is what has to be underwritten, attribute by attribute.
The Returns Inversion Test
- What kind of MW is actually being delivered — nominal, or reliably energised, cooled and commercially usable?
- Which customers is that MW eligible to serve — do its density, cooling, power and energy/carbon characteristics meet the target customer's requirements?
- Where is the cost advantage — construction, land, electricity, cooling, financing or another input?
- Who contractually captures it — owner, operator or tenant?
- How much does competition pass through — does the saving stay margin, or become lower customer pricing?
- What does it do to pricing and utilisation — does cheaper capacity fill faster, or does differentiated capacity sustain stronger economics?
- How much survives into owner cash flow — the question that turns any advantage into an investment argument.
Investment Lens
4 What this means for the broader market
Cheaper to build is a fact about a location. Better to own is a fact about a cash flow. The second does not automatically follow from the first. A cost advantage becomes an ownership advantage only for the part of it the owner can technically convert into usable capacity, qualify for the target customer, contractually hold, competitively defend and actually turn into filled, priced capacity.
The winner is not necessarily the operator with the cheapest concrete, land or electricity. It is the owner who can retain an advantage others cannot easily take back — engineering capability customers value, an energy pathway customers can use, a contract that preserves the economics, a business model that monetises them, or a position competitors cannot readily replicate. The loser is the one who underwrites the build-cost line as though it were the return line — or assumes a sophisticated engineering or sustainability specification must automatically earn a premium. Both shortcuts make the same mistake: they skip the transmission mechanism. The broken assumption is the reflex that the cheapest market is automatically the best market to own in. Sometimes it may be. But that has to be underwritten at the level of usable, customer-eligible capacity and retained cash flow — not assumed from the cost of concrete and power, and not assumed away by a more impressive technical or sustainability specification either.
The returns inversion in one line: the cheapest megawatt to build and the most valuable megawatt to own need not be the same megawatt. And it raises the next question — if a cost advantage only matters where the owner can capture it, what happens when the cheapest place to build is no longer the same place the workload has to run, when AI starts to change what "location" even means?
That is the question I'm taking forward next.