From Space to Power: How AI Demand Is Reshaping Colocation Pricing, Design Points, and Unit Costs
What AI-driven power density is doing to colocation economics
|
TL;DR
|
Why Colocation Pricing Is Changing
Colocation pricing has traditionally centered on space-based pricing — billing by rack or cabinet, commonly known as per-rack or per-cabinet pricing. Power was treated as a secondary spec bundled with that space.
AI demand is shifting the center of gravity from space to power. Increasingly, the first question in a colocation negotiation isn't “how many racks do you need,” but “how many megawatts, when, and at what price per kW.” The billing categories themselves haven't changed — what's changed is which one, space or power, is the actual bottleneck and the primary driver of price.
|
Definition What Is Power-Based Pricing? Power-based pricing sets colocation fees according to actual power supplied and consumed (kW), rather than floor space. The most common form is per-kilowatt pricing, which is increasingly preferred for high-density workloads. As the scarce resource shifts from space to power, a market can show a low vacancy rate while still facing an effective power shortage — this dynamic is the core driver behind the shift to power-based pricing. |
Why the Design Point Is Shifting
Data centers are typically built to a standardized design point that matches space and power capacity — for example, 5–7.5kW per rack across roughly 30 square feet. Traditional enterprise colocation has generally been built to a design point of around 5kW per rack.
AI workloads blow past that design point entirely. AI inference workloads typically require 50kW or more per rack, while AI training clusters can require 30–100kW or more per rack — roughly 10x the traditional design point. This is driving several detailed shifts:
• Design point escalation: from roughly 5kW to 50kW+ per rack, a roughly 10x increase in power density
• Retrofit costs passed through: the cost of retrofitting lower-density facilities for high-density use (e.g., liquid cooling) is increasingly folded into per-kW pricing, regardless of actual power consumption
• Design point mismatch: space built to older, lower-density standards sits vacant while space capable of delivering high-density power remains scarce, creating a two-track market
This mismatch shows up clearly in vacancy data. In H1 2025, colocation vacancy in primary North American markets fell to 1.6%[1], and vacancy in Northern Virginia — the world's largest colocation market — stood at just 0.72%[1]. A separate analysis puts vacancy in major markets at 1.4% by year-end 2025[3]. Vacancy alone suggests an extreme supply shortage, but industry sources note the real constraint isn't a lack of empty racks — it's a lack of power to plug into them.
This trend is widely discussed as a qualitative pattern across global data center markets. However, most of the verifiable quantitative data comes from reports focused specifically on the North American market — these figures represent North America, not a global average. In Korea, as power procurement itself becomes a core business variable amid ongoing discussion of subordinate legislation under the AI Data Center Special Act, this shift in design standards is likely to gradually affect the domestic market as well.
How AI-Driven Demand Is Showing Up in Unit Pricing
As the pricing model and design point shift, actual unit pricing — price per kW — is moving in step. The following North American figures are drawn from CBRE's North America Data Center Trends Report and Lightyear's 2026 State of Connectivity Report.
|
Metric |
Data |
|
Average pricing for 250–500kW deployments (H1 2025) |
$184/kW/month |
|
Pricing increase for 10MW+ requirements (same period) |
Up to 19% |
|
Median all-in $/kW pricing increase (H1 2025 → H2 2025) |
$324 → $380 (+17.3%) |
|
10th / 90th percentile pricing increase (same period) |
$199→$236 (+18.6%) · $550→$663 (+20.6%) |
|
Share of North American capacity under construction that is pre-leased |
~75% (some commitments extend into deliveries after 2027) |
※ All figures above are drawn from reports focused on the North American market and do not represent a global average. See References below for full sourcing.
Two related trends sit behind this pricing increase:
• Shrinking volume discounts: as contiguous blocks of power capacity become harder to find, the volume discounts once standard for 1MW+ deployments are disappearing
• Pre-leasing: large-scale customers are increasingly securing capacity before construction even begins, meaning new entrants effectively face a market with no fixed price
How DAOU Datacenter Is Responding
As power becomes the new pricing basis for the colocation market, the way pricing and unit economics are understood and designed is being redefined right along with it.
DAOU Technology is preparing for this shift by building an AI data center campus. Starting with its first facility opening in 2027, DAOU plans to expand with additional AI data centers in phases, ultimately building out a full AI data center campus. Stay tuned for what's next.
Discover More.
Join us as we take you through every step of the development of our state-of-the-art hyperscale data center in this blog. We loves to talk about what we do. Feel free to chat with our sales experts about data center, managed colocation, and connectivity.
Content is protected by copyright law and is owned by Daou Technology Inc
It is prohibited to modify or commercially use this content without prior consent.
Featured images via gettyimages.
References
1. "Power, Not Space: The Colocation Battleground in 2026," Data Center Knowledge, 2026 — figures cited from CBRE's North America Data Center Trends Report (2025)
2. "The Ultimate Guide to Colocation Pricing," Lightyear, 2026 State of Connectivity Report