Edge-Rechenzentren

Edge data centers solve the one problem hyperscale cannot: physics. When an application needs sub-10-20 ms round trips — real-time AI inference, industrial control, content delivery — capacity must sit near users, in facilities of 0.1-5 MW rather than 100 MW. This hub covers micro and modular economics, edge inference, telco-site conversion and emerging-market deployment, with siting support through our facility catalog and quote service.

Market context: the quiet counterweight to hyperscale centralization

The industry’s headline story is centralization — 219 GW of projected global demand by 2030, concentrated in ever-larger campuses (see our hyperscale hub). But a structural counter-current is building. Sub-1% vacancy and record pricing in core markets — $196.25/kW/month average North American wholesale rates, per CBRE — push latency-tolerant workloads toward cheap-power mega-campuses while pushing latency-sensitive workloads out toward users. Industry analysts consistently project edge infrastructure spending growing at 15-25% annually this decade, driven by three demand engines:

• AI inference distribution. Training centralizes; serving distributes. As AI features embed in consumer and industrial applications, inference moves toward users for latency and bandwidth economics — a single camera-analytics deployment can generate more raw video than it is economic to backhaul.
• 5G and network densification. Telco network functions (RAN processing, packet core) are becoming software on distributed compute, pulling capacity to thousands of network sites.
• Data sovereignty and locality. Regulations increasingly require data processing in-country — and in emerging markets, “in-country” often means building where no hyperscale facility exists yet.

Edge is not a substitute for core capacity; it is the distribution tier of the same architecture. The practical question is never “edge or core?” but “which workloads, how far out, at what unit cost?” — and answering it requires local numbers, not global averages.

Micro and modular economics: what 0.1-0.5 MW actually costs

Edge economics are unforgiving of the fixed costs that hyperscale amortizes away. A security fence, a generator, a fiber build and an operations contract cost roughly the same whether they serve 300 kW or 30 MW — which is why cost per MW rises as facilities shrink, and why form factor selection dominates edge financials:

Form factor	Typical size	All-in cost	Deploy time	Best use
Micro data center (in-building)	1-10 racks (5-100 kW)	$50-300k per unit	4-12 weeks	Retail, factory, branch, on-prem inference
Containerized (ISO-based)	100-500 kW	$7-12M per MW ($2-3.5M typical site)	3-6 months	Telco sites, secondary cities, temporary capacity
Prefab modular building	0.5-5 MW	$8-12M per MW	6-12 months	Regional edge hubs, sovereignty builds
Conventional small build	1-5 MW	$9-13M per MW	18-24 months	Rarely optimal at edge scale anymore

Three rules of thumb from deployments we track. Prefabrication wins on time, not just cost: factory-built, factory-tested modules cut on-site work to weeks and sidestep local construction-quality risk — decisive in emerging markets. Power is the siting gate: even 300 kW can strain a rural feeder, so utility capacity confirmation belongs at the start of site selection, not the end (the same discipline as hyperscale, scaled down — see our power hub). Unmanned operation is an economic requirement: a site too small to justify staff needs remote monitoring (DCIM), high-MTBF design, and a local hands-and-eyes contract priced in from day one — typically $2-5k/month in most markets.

Full modular and container cost breakdowns are in our edge data centers guide and the general build cost guide.

Edge inference: AI’s second wave lands close to users

The AI build-out’s first wave was training capacity — 60-130 kW racks in giant campuses (covered in our AI hub). The second wave is inference, and its economics point outward. Serving a model requires far less power per node than training — a capable inference node runs 5-30 kW, comfortably within air-cooled or lightly liquid-cooled edge form factors — but it multiplies across every metro where users live, and it is latency-bound: interactive AI applications budget 50-200 ms for the whole round trip, leaving little room for a 100+ ms backhaul.

The emerging reference architecture is hub-and-spoke: a regional colocation hub (where our price index and GPU tracker give you the cost baseline — H100-class rentals around $2.35/hour on committed terms) hosts model storage and heavier inference, while 0.1-0.5 MW edge nodes serve real-time traffic within a 20-50 km radius. For operators and investors, this architecture creates a new asset class — small, numerous, standardized sites — and a new tenant profile: AI application companies that need ten 200 kW deployments across a region rather than one 2 MW hall.

Telco sites and emerging markets: where edge deployment is cheapest

Telco real estate is the edge’s unfair advantage. Carriers own thousands of central offices and tower compounds that already solve the three hardest edge siting problems: grid power with backup, diverse fiber, and physical proximity to population. Central-office conversions into 100-500 kW edge colocation are among the most capital-efficient edge plays available — the shell, power entrance and fiber vault are sunk costs — and carriers across Southeast Asia are actively opening these assets to partners as 5G capex pressure meets underused switching floors.

Emerging markets flip the edge business case from optimization to necessity. In Thailand, Vietnam, Indonesia and the Philippines, digital infrastructure concentrates intensely in capital metros: Bangkok’s pipeline approaches 2.5 GW while secondary cities hundreds of kilometers away sit 20-40 ms distant with limited local capacity. For a bank, retailer, manufacturer or game publisher operating nationally, a modular 100-500 kW site in Chiang Mai, Da Nang, Surabaya or Cebu routinely beats both alternatives — the latency and bandwidth cost of backhauling to the capital, and the reliability of self-managed server rooms. Government cloud and data-localization programs add a policy tailwind, and Thailand’s BOI incentives (which approved ~THB 746 billion of data center investment in 2025 — see our Thailand analysis) extend to smaller facilities meeting efficiency criteria.

The regional dynamics — costs 20-50% below Western builds, incentive regimes, and where the fiber actually runs — are covered in our development hub.

Business models: own, lease, or partner

Edge capacity reaches users through four distinct commercial models, and choosing the wrong one is the most common way edge projects destroy value:

1. Owned and operated. The enterprise builds and runs its own micro or modular sites. Maximum control, full capex, and an operations burden that only makes sense with dozens of sites or hard sovereignty requirements. Typical all-in: $2-3.5M per 300 kW site plus $50-100k/year to operate unmanned.
2. Edge colocation. Lease racks in someone else’s regional or edge facility. Zero capex, fastest deployment, and in most secondary cities the realistic default — where such facilities exist. Pricing typically runs 10-30% above metro-hub rates for equivalent space, a premium worth paying against the alternative of building.
3. Modular-as-a-service. Vendors deploy and operate containerized capacity on your site or a partner site under a multi-year service contract — converting capex to a monthly fee, usually with a 15-25% financing premium embedded. Attractive when capital is constrained or site tenure is uncertain.
4. Telco partnership. Deploy into carrier central offices or tower sites under revenue-share or lease terms. The best physical footprint in most countries, with the caveat that carrier decision cycles and standards vary enormously — local knowledge decides whether this takes three months or eighteen.

The portfolio answer is usually mixed: colocation where inventory exists, modular where it doesn’t, telco partnerships where the carrier relationship is already strong. What all four models share is dependence on accurate local cost baselines — which is what our index and catalog provide.

A worked example: the 300 kW secondary-city node

To make the economics concrete, consider a retailer or platform serving northern Thailand from Bangkok, evaluating a 300 kW node in Chiang Mai:

Line item	Modular build (own)	Edge colocation (lease)
Capex	~$2.5M (container, gear, install)	~$0
Space and power (annual)	~$260-320k (power at ~$0.10-0.12/kWh + site costs)	~$400-550k all-in
Operations (annual)	$50-100k (remote monitoring + hands contract)	Included
Backhaul savings vs Bangkok	$100-300k/year (reduced long-haul bandwidth)	Same
Latency improvement	~15-30 ms round trip for local users	Same
Break-even vs leasing	Years 3-5 of sustained use	Immediate, lower commitment

The pattern generalizes: leasing wins on speed and flexibility for the first sites; building wins at scale or where no leasable inventory exists — which in emerging-market secondary cities is often the deciding factor. Either way, the decision turns on numbers most teams currently guess: local power tariffs, realistic colocation rates, backhaul pricing, and modular vendor quotes. Getting those four numbers verified is precisely the feasibility work we do through /quote/.

How Coloprice helps: feasibility and siting, not guesswork

Edge projects die from two causes: sites that can’t be powered or connected, and business cases built on guessed costs. Both are information problems, and both are what we solve:

• Cost benchmarks before you commit. Our price index tells you what regional colocation actually costs — the number your edge site must beat or complement — and stats shows capacity by country so you can see exactly where the coverage gaps (your opportunity) are. If your edge case involves inference hardware, the GPU tracker prices the compute layer.
• A 202-facility catalog for hub selection. Hub-and-spoke architectures need the hub: filter facilities across 19 countries — with Southeast Asia coverage deeper than any Western directory — to anchor your regional design.
• Feasibility and siting help, free, within one business day. Tell us the requirement at /quote/ — “300 kW modular in northern Thailand,” “inference nodes in three Vietnamese cities,” “convert our telco floor into edge colo” — and we match you with modular vendors, local operators and equipment suppliers we’ve vetted, within one business day, at no cost. Because we also work the equipment supply chain (racks, UPS, cooling, prefab modules sourced directly from qualified Asian factories at 20-40% below Western list), we can support the build, not just the plan.
• Guides that go deeper. Start with edge data centers, then the build guide for the development process end to end.

The edge opportunity rewards whoever can answer three questions fastest: where the demand is, what the site really costs, and who can build and run it locally. Enterprises that answer them with verified numbers deploy in months; those that guess spend a year discovering their site can’t be powered. We built this portal so you can answer all three before your competitors finish their first RFI — start with the catalog and index, then bring the requirement to /quote/.