ไฮเปอร์สเกล

Hyperscale is where the data center industry’s entire supply-demand imbalance is concentrated: global capacity demand is projected to reach 219 GW by 2030 — roughly triple 2023 levels — while the top four US markets run at sub-1% vacancy with most future supply already preleased. This hub covers campus economics, the preleasing arms race, and the Southeast Asian build-out, with live benchmarks from our price index and multi-country pipeline data on stats.

Market context: a market that is sold out years in advance

The defining statistic of the hyperscale era is not price — it is preleasing. Per CBRE’s North America Data Center Trends, roughly 74-80% of all capacity under construction in the primary US markets is already committed before delivery, overwhelmingly to cloud and AI tenants. In Northern Virginia — the world’s largest market at well over 3 GW of inventory — vacancy sits near 0.5%, roughly 2 GW is under construction, and effectively all 2026 scheduled supply is committed, pushing preleasing conversations into 2027 and beyond.

Scarcity has done what scarcity does. Average North American wholesale asking rates reached a record $196.25/kW/month (+6.6% YoY) in 2025, and the premium concentrates precisely where hyperscale buyers live: 3-10 MW blocks rose 12.5% year-over-year, with larger contiguous requirements in constrained markets clearing higher still. Ashburn wholesale asking rates now start around $215/kW. Meanwhile the construction pipeline in primary US markets contracted for the first time since 2020 — 5,994 MW under way at end-2025, down from 6,350 MW — not for lack of demand, but for lack of power and equipment.

For anyone buying, funding or competing with hyperscale capacity, the operational conclusion is the same: the market clears 18-36 months before buildings exist. Waiting for delivered space means bidding on remainders.

Anatomy of a hyperscale campus

The unit of hyperscale development is no longer the building — it is the campus. A representative modern masterplan assembles 100-500+ acres with utility-scale power (dedicated substations of 300 MW to over 1 GW), then delivers buildings of 32-100 MW in phases over 5-10 years. The economics reward this structure three ways: land and substation costs amortize across phases; tenants gain contractual expansion rights (the single most valuable clause in a hyperscale lease); and developers convert one entitlement fight into a decade of delivery.

Scale benchmarks worth knowing — the figures below reflect current CBRE reporting, our own facility catalog, and disclosed campus masterplans across the US and Southeast Asian markets we track:

Metric	Typical 2026 hyperscale campus
Campus masterplan	100 MW - 1+ GW IT load
Single building	32-100 MW
Build cost (US)	~$11.3M per MW average; $12-14M in tier-1 metros
Build cost (SEA)	$8-12M per MW; Vietnam $5.7-8.7M
Lease term	10-15 years, 2-4% annual escalators
Preleasing (top-4 US markets)	~74-80% of under-construction capacity
Rack density (AI halls)	60-130+ kW/rack, liquid-cooled

Two shifts differentiate the 2026 campus from its 2020 predecessor. Density: AI training halls at 60-130 kW/rack with direct-to-chip liquid cooling now anchor new builds — covered in depth in our AI data centers hub. Power strategy: campuses increasingly arrive with their own generation story — on-site gas turbines as bridge power, solar-plus-storage PPAs, and nuclear contracts (over 10 GW committed by big tech to date) — detailed in our power hub.

The Southeast Asia wave: where the next gigawatts land

Hyperscale growth follows a simple gradient: demand flows from constrained, expensive markets toward available power and land. Nothing illustrates it better than Singapore’s spillover into Southeast Asia.

Singapore — the region’s connectivity hub — rations new data center capacity through a government allocation regime, keeping colocation rates at a world-leading $330-475/kW/month. The demand didn’t disappear; it moved 20 km north. Johor, Malaysia built a pipeline of roughly 4-5 GW across committed and planned projects in under five years (DCD’s Johor analysis), attracting Microsoft, AWS, Google, ByteDance, and campus builders like STACK (220 MW) and multiple 500 MW entrants — before the government began prioritizing AI-grade projects and screening for power and water intensity.

Thailand is the next leg. The Board of Investment approved roughly THB 746 billion (~$22 billion) of data center investment in 2025 — the largest industry in its pipeline — behind an 8-year tax holiday for high-efficiency facilities. TikTok committed $8.8 billion to Thai hosting infrastructure; Google, AWS and Microsoft have all announced Thai regions or campuses; and Bangkok’s announced pipeline is approaching 2.5 GW. Our full breakdown: Thailand’s data center boom.

The arbitrage for capacity buyers is quantifiable: SEA hyperscale capacity builds at $8-12M/MW (versus $12-14M in tier-1 US metros), runs on industrial power at $0.08-0.12/kWh, and leases at rates anchored far below Singapore — while sitting on the same submarine cable systems. For latency-tolerant workloads, including most AI training, the case for placing the next block in Johor or Bangkok rather than a waitlisted US market is increasingly just arithmetic.

Buying wholesale blocks: the 10 MW+ playbook

Securing large blocks in this market is a different discipline from colocation procurement. The buyers winning capacity share three behaviors:

1. They buy on the development curve, not the delivery date. Committing at powered-shell stage — or funding a build-to-suit from land stage — earns pricing 10-20% inside delivered-space rates and locks expansion options. The 12.5-19% year-over-year inflation on large blocks is, equivalently, the annual cost of waiting.
2. They underwrite power before real estate. A block is only as good as its substation timeline. Diligence now leads with interconnection status, utility contract terms and backup generation — see our power & energy hub for grid queue dynamics by market.
3. They diversify geographies deliberately. With top-tier US markets rationed, sophisticated buyers split requirements: latency-critical capacity in core markets at premium pricing, training and batch workloads in SEA or secondary US markets at a 30-50% discount. Market-by-market pricing on our index makes the spread explicit.

Contract structure matters as much as price: expansion rights with defined pricing, escalator caps (3% caps against a market inflating 6-12% are worth more than day-one discounts), and termination/assignment flexibility for workloads that may migrate as AI hardware generations turn over every 18-24 months.

Who is buying: the demand stack behind the numbers

Understanding who is absorbing capacity explains where the market is headed. The classic hyperscale tenants — AWS, Microsoft, Google, Meta, Oracle — remain the volume anchor, leasing wholesale blocks even as they build owned campuses, because leasing converts capex to opex and compresses time-to-capacity. But two newer cohorts have transformed demand since 2023:

• AI-native companies — model labs, GPU cloud providers (“neoclouds”) and AI application companies — now compete directly for the same multi-MW blocks. Their requirements skew toward extreme density (60-130+ kW/rack, liquid cooling — see the AI hub) and their credit profiles vary wildly, which is reshaping how operators underwrite leases: the market increasingly prices tenant credit into the rate, with investment-grade covenants earning discounts of 5-15% against venture-backed paper.
• Sovereign and regional cloud programs, particularly across Asia and the Middle East, are contracting national-scale capacity for data-residency reasons — a structural, policy-driven demand layer that did not exist at scale five years ago and that lands disproportionately in markets like Thailand, Malaysia, Indonesia and Vietnam.

The composition shift matters for strategy: when demand was purely big-cloud, capacity followed a predictable regional-expansion logic. AI-native and sovereign demand is lumpier, faster-moving, and more price- and power-sensitive — which is exactly why markets with energized land and incentives are capturing an outsized share of the new wave.

Delivery risk: why announced gigawatts are not real gigawatts

A pipeline headline is not capacity. Between announcement and energization sit four attrition filters, and sophisticated buyers discount for each: grid interconnection (3-7 year queues in major markets; the number one killer of announced projects), equipment lead times (switchgear, transformers and generators quote 12-24 months from Western suppliers — a constraint that contracted the US construction pipeline in 2025 for the first time since 2020), water and permitting (Malaysia now screens projects for power and water intensity; several jurisdictions restrict evaporative cooling), and financing reality (announced SEA projects without anchor tenants or committed equity routinely slip years or vanish).

Applied to the regional numbers: Johor’s ~4-5 GW pipeline includes roughly 700 MW actually under construction; Bangkok’s ~2.5 GW announced pipeline will similarly deliver on a curve, not a cliff. The practical discipline for capacity buyers is to classify every option by stage — operating, under construction, permitted-with-power, or announced — and to treat only the first three as procurable. This is precisely how our facility catalog and stats pages structure the data, and why our quote matching prioritizes operators with verified, energized or near-energized capacity over press-release megawatts.

How Coloprice helps

Hyperscale capacity intelligence is traditionally locked inside brokerage relationships and $20k+ analyst subscriptions — workable for Microsoft, less so for the AI startup that needs 5 MW or the fund underwriting its first data center deal. We publish the data openly and make our money on introductions:

• Capacity and pricing intel, free. The price index tracks wholesale and retail rates across markets; stats aggregates supply and pipeline by country — including the Thai, Malaysian, Vietnamese and Indonesian markets where our coverage is deepest and Western trackers are thinnest.
• A 202-facility catalog across 19 countries. Filter facilities by market, power scale and GPU-readiness to map who actually has — or is building — the block you need.
• Early access to Southeast Asian blocks. Because we work directly with operators and developers across SEA, we frequently know about uncommitted phases and upcoming powered shells before they are marketed. If you are placing 1-50 MW in the region, tell us the requirement at /quote/ — we match you with operators holding real availability within one business day, free, with no listing-fee bias in what we recommend.
• The AI capacity bridge. If your hyperscale requirement is GPU-driven, our GPU tracker and AI hub let you price the alternative — renting accelerators at market rates (H100s around $2.35/hour on annual commitments) versus deploying your own hardware in wholesale space.
• Deal-stage guides. The data center investment guide and build guide cover underwriting and build-to-suit structures in detail.

In a market where 219 GW of demand is chasing a shrinking construction pipeline, information advantage compounds into capacity advantage. The buyers who secured 2026 capacity did so in 2024; the buyers who will have 2028 capacity are committing now, in markets most of their competitors have not yet priced. The tools to be in that group are on this site and cost nothing: check the index, map the catalog, and put your requirement in front of the right operators via /quote/ — before this year’s supply becomes next year’s waitlist.