Data center power & energy

Power has replaced capital, land and even chips as the binding constraint on data center growth: global capacity demand is heading toward 219 GW by 2030 while grid interconnection queues in the top markets stretch 3-7 years. This hub covers the grid bottleneck, the nuclear pivot, Southeast Asian PPAs and electricity cost benchmarks — the inputs for power-aware site selection, which our facility catalog and quote service are built to support.

Market context: the grid is the bottleneck

The clearest evidence that power now governs the industry: in 2025, for the first time since 2020, the construction pipeline in primary US markets shrank — to 5,994 MW from 6,350 MW, per CBRE — while vacancy sat at record lows and rents at record highs ($196.25/kW/month average, +6.6%). Developers are not short of demand or money; they are short of megawatts. Utility interconnection requests in hot markets now face 3-7 year timelines, large connection deposits, and take-or-pay terms; Dublin and Amsterdam have imposed outright connection moratoria in the past cycle, Singapore rations capacity by government allocation, and Northern Virginia’s utility has publicly flagged multi-year delays for large new loads.

The market’s response has been to reprice power access itself. “Powered land” — sites with secured utility capacity — trades at multiples of comparable unpowered land. Tenants preleasing 74-80% of under-construction capacity in top US markets (see our hyperscale hub) are, in substance, queuing for electricity. And the largest buyers have gone around the queue entirely: building on-site generation, contracting nuclear plants, and moving workloads to power-rich geographies.

For anyone siting capacity, the decision hierarchy has inverted. The old order was: market, connectivity, land, power. The 2026 order is: power first — cost, timeline, reliability, and expansion headroom — with everything else negotiated afterward.

The nuclear pivot: 10+ GW and counting

Nothing signals the industry’s power desperation — and time horizon — like the nuclear renaissance hyperscalers are personally financing. The committed pipeline now exceeds 10 GW:

Buyer	Deal	Capacity	Timeline
Microsoft	20-year PPA; Three Mile Island Unit 1 restart (Constellation)	835 MW	Target 2028
Google	SMR fleet agreement with Kairos Power — first US corporate SMR fleet deal	~500 MW	First units ~2030
Amazon	$700M+ into X-energy; target up to 5 GW of SMRs by 2039; Susquehanna nuclear-adjacent campus	5 GW pathway	Late 2020s-2030s
Meta	RFP for new nuclear generation	1-4 GW	Early 2030s

The table understates the totals: beyond contracted capacity, Microsoft, Google and Amazon have each signed framework agreements and made equity investments across the SMR supply chain, effectively pre-buying manufacturing slots for the 2030s.

Three lessons for everyone who is not a hyperscaler. First, the majors are underwriting 20-year power positions — they expect the constraint to persist, not clear. Second, SMR timelines mean nuclear solves the 2030s, not the 2020s — bridge strategies (on-site gas turbines, grid-plus-storage, geographic arbitrage) decide who gets capacity this decade. Third, restarts and uprates of existing reactors are the only near-term nuclear — which is why Three Mile Island commanded a 20-year commitment and why nuclear-adjacent sites command premiums.

The strategic conclusion: with the biggest buyers locking multi-decade generation, mid-market operators and enterprises compete for what remains of grid headroom — making markets with available, reliable, reasonably priced grid power disproportionately valuable. Increasingly, those markets are in Southeast Asia.

PPAs and Direct PPAs in Southeast Asia: buying power like a hyperscaler

Southeast Asia’s regulatory landscape for corporate power purchasing has moved faster in three years than in the prior two decades, and it is a core reason the region’s data center pipeline (Johor ~4-5 GW; Bangkok approaching 2.5 GW behind ~THB 746B of BOI-approved investment — see our Thailand analysis) is bankable:

• Vietnam legalized Direct PPAs in 2024, allowing large consumers to contract directly with renewable generators — through either private wire or the grid with wheeling charges. Combined with the region’s lowest industrial tariffs (~$0.07-0.08/kWh) and lowest build costs ($5.7-8.7M/MW), it makes Vietnam the region’s cost floor.
• Thailand launched a Direct PPA pilot of roughly 2 GW explicitly aimed at data centers, layered on top of BOI incentives (8-year tax holiday for high-efficiency facilities). Industrial grid power runs ~$0.10-0.12/kWh with strong reliability in the Bangkok metro.
• Malaysia offers the Corporate Green Power Programme and CRESS framework for green supply to large consumers, with industrial tariffs around $0.08-0.11/kWh — while now screening data center applications for power and water intensity.
• Singapore cannot expand domestic generation meaningfully and instead imports green electricity via interconnectors — one driver of its $330-475/kW colocation pricing and its demand spillover into Johor and Batam.

The practical caveat: every scheme carries eligibility thresholds, wheeling charges, and curtailment terms that move delivered cost by 20-40% versus headline generation prices. A “cheap” Direct PPA with punitive wheeling can lose to plain grid tariff — modeling delivered $/kWh per site is exactly the analysis our team supports through /quote/.

Electricity benchmarks: why $0.01/kWh beats $1M of capex

Electricity is the dominant lifetime cost of any data center. The arithmetic: a 30 MW IT-load facility at 80% utilization and PUE 1.3 draws roughly 273 GWh/year. Each $0.01/kWh is therefore worth ~$2.7M per year — over a 15-year life, roughly $41M, comparable to the entire construction cost of several MW. Benchmark industrial rates across key DC markets:

Market	Industrial power (approx.)	Annual cost, 30 MW IT @ PUE 1.3
Vietnam	$0.07-0.08/kWh	$19-22M
Indonesia	$0.07-0.10/kWh	$19-27M
Malaysia	$0.08-0.11/kWh	$22-30M
Thailand	$0.10-0.12/kWh	$27-33M
US (major DC markets)	$0.09-0.15/kWh	$25-41M
Singapore	$0.20+/kWh	$55M+

These figures are directional industrial-tariff ranges; actual delivered cost depends on voltage level, demand charges, time-of-use structure and any green-supply premiums, all of which vary by utility and contract.

Two corrections to naive tariff-shopping. Reliability is part of the price: a market with frequent grid events forces more generator runtime and stricter UPS architecture — diesel at $0.25-0.40/kWh effective — narrowing headline gaps. Time-to-power is part of the price: a $0.08/kWh site energized in five years loses to a $0.10/kWh site energized in eighteen months for any business with current demand; at hyperscale rents of $196.25/kW/month, a year of delay on 30 MW is roughly $70M of foregone revenue.

Water is the second utility. Large campuses using evaporative cooling draw 1-5 million liters daily, and water stress has become a formal siting criterion — Malaysia screens applications for water intensity, and several jurisdictions have restricted evaporative designs. Closed-loop and air-cooled architectures cut water use 70-90% but raise PUE and power cost; liquid-cooled AI halls (see the AI hub) change the calculus again. Power and water must be underwritten together, per site.

Bridge power: buying time while the grid catches up

With nuclear a 2030s answer and interconnection queues measured in years, the decisive competition this decade is over bridge strategies — ways to energize capacity before the grid can:

• On-site gas turbines and reciprocating engines are the workhorse bridge: 10-100+ MW installable in 12-24 months, producing power at roughly $0.08-0.15/kWh depending on gas prices and utilization. Several flagship AI campuses in the US have deployed hundreds of megawatts of on-site gas as primary power pending grid connection. The trade-offs are emissions permitting, fuel logistics, and community opposition — all faster to resolve in some jurisdictions than others.
• Fuel cells offer cleaner on-site generation at smaller scale with easier permitting, at a cost premium over turbines; they suit urban and constrained sites where combustion permits are slow.
• Battery storage (BESS) does not create energy but converts interruptible or time-limited grid offers into firm capacity — utilities increasingly offer flexible-connection deals (curtailable during peaks) years earlier than firm connections, and 2-4 hour batteries make those offers usable for data centers. Flexible-connection-plus-storage is quietly becoming the fastest path to megawatts in queue-bound markets.
• Geographic arbitrage remains the cleanest bridge of all: moving latency-tolerant workloads to markets where the grid has headroom now. This is Southeast Asia’s structural advantage — Thailand, Malaysia, Vietnam and Indonesia can deliver firm industrial connections in 12-24 months at $0.07-0.12/kWh, timelines that no queue-bound Western market can match. It is the power story behind the region’s multi-GW pipeline, and the reason our coverage concentrates there.

The underwriting rule: every bridge has a cost per kWh, an emissions profile, and an exit plan back to grid or contracted clean power. Deals that price the bridge honestly — rather than assuming the queue will move — are the ones that close on schedule.

How Coloprice helps: power-aware site selection

Most site selection still starts with real estate and discovers power problems in month six. We built our data to run the analysis in the right order:

• Power-aware facility data. Our 202-facility catalog covers capacity and expansion headroom across 19 countries, with the deepest coverage in Southeast Asia — the region where the power arbitrage is largest and public data is thinnest. Stats aggregates supply by market so you can see where capacity is actually being energized, not just announced.
• Cost benchmarks on both sides of the meter. The price index shows what powered capacity rents for by market; the GPU tracker shows what the compute layer earns — together they let you model whether a cheap-power market’s colocation discount actually reflects its tariff advantage (when it doesn’t, that spread is your opportunity).
• Site selection support, free, within one business day. Bring us the requirement at /quote/ — “10 MW in a market with sub-$0.10 power and Direct PPA access,” “GPU-ready capacity where the grid can support expansion to 30 MW,” “compare delivered power cost Bangkok vs Johor vs Da Nang” — and we match you with facilities and development partners that clear your power criteria, within one business day, at no cost. We work with operators and developers across the region daily; we know which “available” megawatts are real.
• Build-side depth. If you are developing rather than leasing, the development hub, build guide and cost guide cover substations, generators and the equipment supply chain — including direct factory sourcing of electrical equipment at 20-40% below Western list prices.

In a 219-GW decade, every data center decision is a power decision wearing a real estate costume. The operators and investors outperforming this cycle share one habit: they underwrite the electrons first — tariff, timeline, reliability, headroom — and let the real estate follow. That analysis takes weeks with brokers and consultants, or a day with the right data and the right introductions. Start with the electrons: check the index, screen the catalog, compare markets on stats, and put your power criteria in front of us at /quote/ — matched options within one business day, free.