What Governments Miss When They Win the Data Center Race

The announcement follows a familiar script. A hyperscaler or data center developer names a country. A figure — in the billions — gets attached. Government ministers appear in photographs. The words jobs, sovereignty, and digital future appear within the first three sentences of the press release.

Then the campus gets built. The servers go in. The cooling systems hum. And the conversation about what the host country actually received begins — usually much later, and usually more quietly than the announcement.

This is not, in 2026, a story about who gets to be the next Singapore. Singapore has already made peace with its own limits — capping new development under a strict sustainability-linked approvals scheme, pushing hyperscale demand deliberately toward its neighbours, and accepting the role of high-value hub over high-volume one. Singapore is not the finish line. It is the starting gun.

APAC’s data center development pipeline hit a record 19.4GW in 2025, with Southeast Asia accounting for the largest share of under-construction capacity at 31%. The race now is a multi-front competition across a dozen markets, each offering variations of the same package — land, power, incentives, regulatory stability — and each trying to become the anchor hub for their subregion before capital flows somewhere else.

Johor, Mumbai, and Bangkok are now central to hyperscaler deployment strategies — not as overflow contingencies, but as primary targets. Malaysia and India together accounted for 58% of new operational capacity added across APAC in 2025.

The pitch decks are converging. The reality underneath them is not.

The most instructive case in the region sits twenty kilometres from Singapore’s central business district.

Johor more than doubled its operational data center capacity in 2025, rising from 401MW to 897MW — a 124% increase in a single year, with 315MW under construction and 2,099MW in planning. The investment figures are real. The construction cranes are real. The hyperscaler logos on campus signage are real.

Johor’s strategic location adjacent to Singapore, coupled with favourable government incentives and lower operational costs, establishes it as a viable alternative for hyperscale expansion. But the majority of workloads running in Johor are Singapore overflow. The latency-sensitive applications, the financial services platforms, the customer-facing systems — those stay where the regulatory environment, submarine cable density, and enterprise relationships already exist.

Johor is winning the land grab in the only way the land grab can currently be won — by being useful to Singapore. That is a genuine prize. But it is not sovereignty. It is adjacency.

The decisions about what runs where, and for whom, are still made elsewhere.

Here is the number no pitch deck leads with: a hyperscale data center facility — the kind that anchors a government’s digital infrastructure announcement — typically employs between 30 and 50 permanent workers once operational.

The construction phase is different. Thousands of workers cycle through a major campus build over two to three years. Those jobs are real, and for communities around a new campus, they matter. But they are temporary by design. Once a facility is commissioned, the on-site headcount settles into a small team of engineers, technicians, and security staff.

Investigations across multiple markets found that many data centers employ fewer than 125 permanent workers — and some as few as 25. The permanent roles that do exist are heavily weighted toward technical operations, which require skills that often need to be imported in markets where the training ecosystem is still being built.

This is not a failure of execution. It is the nature of the asset class. Data centers are capital-intensive and labour-light by design. Research is clear that fiscal incentives for data centers cannot be justified on the grounds of job creation alone. The case for hosting them rests elsewhere.

The honest accounting is more complex than the jobs figure alone — and more honest than either the press release or its critics tend to be.

Communities hosting data centers receive several things that rarely appear in the headline number. Grid infrastructure built to serve a campus frequently benefits surrounding areas — in markets like Johor and Mumbai, where reliable power access was previously constrained, this is a genuine and lasting upgrade. Local contractor spend during construction — catering, security, civil works, facilities management — circulates in the local economy in ways that persist beyond the build phase.

Over time, the picture expands. Hyperscalers across APAC are running certification programs and graduate hiring pipelines in partnership with local polytechnics and universities. Submarine cable landing stations, anchored by data center demand, lift connectivity for entire countries — cheaper, faster internet for businesses and households downstream. Lower latency, higher reliability cloud access enables local startups, fintechs, and health platforms to operate at costs that were not previously available to them.

The honest caveat is not that these benefits are absent. It is that they arrive on a timeline governments rarely acknowledge when signing the initial agreement.

Loudoun County, Virginia — the strip running through Ashburn known as Data Center Alley — is the most mature case study in the world for what a community can actually collect from this industry. Roughly 200 data centers. More concentrated capacity than the next six US markets combined.

In the proposed FY2027 budget, data center revenue makes up nearly half of Loudoun County’s total tax revenue at $1.3 billion. For every dollar of county services provided to data centers, the county receives $26 back in tax revenue. Property tax rates are approximately 25% lower than neighbouring Virginia counties as a result. Schools, roads, libraries, parks — funded in meaningful part by server racks.

This is what winning looks like. It took twenty-five years to get here.

And even now, winning is complicated. The Board of Supervisors adopted new zoning controls on data center development in March 2025, explicitly because the county cannot become overly dependent on a single revenue source that may evolve in the future. Virginia’s state legislature is simultaneously debating whether to remove the sales tax exemptions that helped attract the industry — a change local officials have described as ranging from bad to disastrous for the county’s fiscal position.

Twenty-five years of concentration, at a scale no APAC market is currently approaching, produced a genuine community dividend — and a new kind of vulnerability. Both are part of the answer.

The more sophisticated argument for the land grab is not jobs. It is sovereignty. Data residency requirements, cloud-first government mandates, and concerns about cross-border data flows have made domestic data center infrastructure a genuine policy priority across the region.

The paradox is structural, not accidental.

A data localisation law keeps data inside a border. It does not determine who controls access to it, under whose legal framework, or what happens when the jurisdiction of the hyperscaler’s parent company and the jurisdiction of the host government come into conflict. Several APAC governments have found that owning the regulatory requirement is not the same as owning the outcome.

The building is onshore. The terms of service, the encryption keys, and the parent company’s registered address are not.

The governments most likely to navigate this well are those building domestic regulatory capability in parallel with foreign investment — training local data governance expertise, developing meaningful audit rights, and insisting on contractual clarity about access conditions before ground is broken. Assuming that the infrastructure produces the sovereignty, rather than treating it as something that must be actively constructed alongside the build, is how governments end up with the building but not the keys.

Several parties win. Not always the ones in the photographs.

Hyperscalers win clearly. De-risked expansion into high-growth markets, subsidised power, streamlined permitting, and the optionality to scale or exit based on demand signals. The incentive competition between governments — each trying to outbid the last — works in their structural favour.

Global institutional capital wins. Sovereign wealth funds backing platforms across APAC with commitments in the billions are not speculating — they are buying cash-flow infrastructure in a region where AI-driven compute demand is structurally growing for the next decade. They understand exactly what they are purchasing.

The local professional class wins: engineers, data center operators, lawyers structuring the deals, consultants writing the feasibility reports. Real, well-paying jobs — just not the ones that feature in the headline employment figure.

Host communities get something real — grid infrastructure, construction activity, downstream connectivity gains, skills investment, and eventually, at sufficient scale and after incentive periods expire, meaningful tax revenue. The gap between the announcement and the arrival of that revenue is measured in years. Sometimes decades.

The deal is genuine. The question is whether governments are negotiating with clear eyes about what they are providing — and what they will actually receive, and when.

Asia Pacific is expected to nearly double its data center capacity to 57GW by 2030. The capital is real. The infrastructure is being built. The demand driving it — AI compute, cloud migration, digital services expansion across a four-billion-person region — is not a mirage.

The question is not whether to participate. It is on what terms, and with what honest accounting of what arrives when.

Loudoun County shows what the long game can produce. It also shows what it costs — in dependency, in land, in power, in a county now actively limiting the industry it built itself around. APAC markets are still in the early chapters of a story that takes decades to fully read.