Colocation

Colocation Pricing Models Compared: Per-kWh vs Per-kW vs Flat Rate

Choosing a data center colocation provider is not just about location, uptime, and connectivity. The pricing model determines how much you actually pay -- and the wrong model for your workload can cost 30-50% more than the right one. Whether you are hosting ASIC miners, GPU clusters for AI training, or traditional enterprise servers, understanding the mechanics of per-kWh, per-kW, and flat rate pricing is essential for making an informed decision.

This guide breaks down each pricing model, calculates real-world costs, and identifies which model works best for specific workload types.

The Three Colocation Pricing Models

Every colocation provider prices their service using one of three models (or a hybrid). Each model allocates costs differently between power consumption, cooling overhead, physical space, and managed services.

Model 1: Per-kWh Metered Pricing

Per-kWh pricing charges the customer for actual electricity consumption, measured by a power meter at the rack or circuit level. The rate includes the raw electricity cost plus a markup that covers cooling, facility overhead, and the provider's margin.

How it works: Your equipment draws power. A meter tracks total kilowatt-hours consumed per billing cycle. You pay that number multiplied by the per-kWh rate.

Typical rates (mid-2026):

• Enterprise colocation (Tier III/IV): $0.10 - $0.18/kWh
• High-density hosting (AI/GPU): $0.08 - $0.14/kWh
• Bitcoin mining / ASIC hosting: $0.045 - $0.085/kWh (US markets)
• Bitcoin mining / ASIC hosting: $0.035 - $0.060/kWh (UAE/Middle East)

What is included: Electricity, cooling (PUE overhead is embedded in the rate), basic physical security, and environmental controls. Space rental, network connectivity, remote hands, and cross-connects are typically billed separately.

Best for: Workloads with consistent, high power draw -- particularly ASIC mining and GPU hosting -- where power is the dominant cost and you want to pay only for what you use. Also ideal for operators who can shift workloads or curtail during peak pricing periods.

Model 2: Per-kW Committed Power Pricing

Per-kW pricing charges a fixed monthly rate per kilowatt of committed power capacity. The customer reserves a specific power allocation (e.g., 50 kW per cabinet) and pays for that capacity regardless of actual consumption.

How it works: You commit to a power capacity (say, 100 kW). You pay 100 kW multiplied by the monthly $/kW rate, every month, whether your equipment draws 40 kW or 100 kW. If you exceed your committed capacity, overage charges apply at a premium rate.

Typical rates (mid-2026):

• Enterprise colocation: $120 - $200/kW/month
• High-density AI hosting: $80 - $150/kW/month
• Wholesale colocation (1 MW+): $60 - $100/kW/month

What is included: Everything. Electricity, cooling, space, basic connectivity, security, and facility management are bundled into the per-kW rate. This simplicity is the model's main appeal -- one number, one invoice, no surprises.

Best for: Enterprise workloads with variable power draw, organizations that value budget predictability, and deployments where power consumption fluctuates significantly (e.g., batch AI training jobs that spike and idle). Also preferred by CFOs who want a fixed infrastructure line item.

Model 3: Flat Rate (Per-Rack or Per-Unit)

Flat rate pricing charges a fixed monthly fee per rack, cabinet, cage, or suite, regardless of power consumption up to a defined limit.

How it works: You lease a full rack (typically 42U) with a power allocation (e.g., 5 kW per rack). You pay the same monthly fee whether you fill the rack with 2U of equipment or 42U, as long as you stay within the power cap. Exceeding the power cap triggers overage billing, usually at per-kWh metered rates.

Typical rates (mid-2026):

• Standard rack (5-8 kW): $800 - $1,500/month
• High-power rack (10-15 kW): $1,500 - $3,000/month
• High-density rack (20-30 kW): $3,000 - $6,000/month
• Private cage (4-10 racks): $5,000 - $25,000/month

What is included: Space, a fixed power allocation, cooling, physical security, and usually 1 Gbps of included bandwidth. Cross-connects, additional bandwidth, and remote hands are extra.

Best for: Small to mid-size deployments (1-10 racks) where simplicity and predictability matter most. Common for enterprise IT that needs a few racks of servers, switches, and storage with minimal complexity. Not optimal for power-intensive workloads like mining or AI, because the per-rack power cap limits density.

Side-by-Side Cost Comparison

To illustrate the real cost difference, here is the same workload priced under all three models. Assume a deployment of 100 kW of ASIC miners running 24/7 at 95% average utilization.

Cost Factor	Per-kWh ($0.065)	Per-kW ($90/kW/mo)	Flat Rate ($2,500/rack x 10)
Monthly power cost	$4,504	(included)	(included up to cap)
Monthly facility/space	$500	(included)	(included)
Monthly total	$5,004	$9,000	$25,000
Annual total	$60,048	$108,000	$300,000
Effective $/kWh	$0.072	$0.130	$0.432

Calculation: 100 kW x 0.95 utilization x 730 hours/month = 69,350 kWh/month. Per-kWh model: 69,350 x $0.065 = $4,508 + $500 space = ~$5,004. Per-kW model: 100 kW x $90 = $9,000. Flat rate: 100 kW across 10 high-density racks at $2,500/rack = $25,000.

The takeaway is dramatic: for a power-intensive, constant-draw workload like Bitcoin mining, per-kWh metered pricing costs roughly 55% less than per-kW and 80% less than flat rate. The cost advantage reverses for low-utilization workloads, where per-kW pricing's predictability and included services justify the premium.

Choosing the Right Model for Your Workload

Bitcoin Mining and ASIC Hosting

Recommended: Per-kWh metered. ASIC miners draw constant, predictable power. There is no variable utilization that benefits from per-kW bundling. Every cent saved on the per-kWh rate drops directly to mining margin. At scale (1 MW+), negotiate a power purchase agreement for even lower rates, or seek providers offering demand-response discounts for load curtailment during grid peaks.

AI Training (GPU Clusters)

Recommended: Per-kW committed. GPU training clusters have bursty utilization -- intense during training runs, idle between jobs. Per-kW pricing absorbs the peaks without overage charges, and the bundled cooling is critical for high-density GPU racks that can draw 30-70 kW per cabinet. The budget predictability also simplifies cost allocation across ML teams sharing infrastructure.

AI Inference (Production Serving)

Recommended: Per-kWh metered. Production inference workloads run continuously at predictable power levels (similar to mining), making per-kWh pricing more efficient. Inference servers also tend to be more power-efficient per operation than training hardware, so metered pricing rewards that efficiency directly.

Enterprise IT (Mixed Workloads)

Recommended: Per-kW committed or flat rate. Traditional enterprise deployments -- web servers, databases, storage arrays, networking -- have moderate and variable power consumption. The all-inclusive nature of per-kW pricing simplifies budgeting. For small deployments under 10 racks, flat rate per-rack pricing is the simplest option with no usage tracking required.

Hidden Costs to Watch For

The headline rate is never the full picture. These additional fees can add 10-25% to your colocation bill.

• PUE overhead (per-kWh models): Some providers quote the raw electricity rate but add a PUE multiplier (e.g., 1.3x) that inflates the effective rate by 30%. Always confirm whether the quoted rate is all-in (power + cooling) or power-only.
• Cross-connect fees: Physical connections to network carriers typically cost $150-$300/month per connection. A deployment requiring 4-6 cross-connects adds $600-$1,800/month.
• IP transit and bandwidth: Included bandwidth is often basic (1 Gbps burstable). Dedicated 10 Gbps ports or premium IX peering can cost $1,000-$5,000/month.
• Remote hands: Most providers include 1-2 hours of free remote hands per month. Beyond that, rates range from $75-$150/hour. For mining operations requiring frequent hardware swaps, this adds up quickly.
• Power redundancy tier: The base rate may cover N power (single feed). N+1 or 2N redundancy (required for enterprise SLAs) adds 15-30% to the power cost component.
• Contract escalators: Annual price increases of 2-5% are common. Over a 3-year contract, a 3% annual escalator turns a $0.065/kWh rate into $0.071/kWh by year 3 -- a 9% increase that compounds.

Negotiation Strategies by Model

Negotiating Per-kWh Rates

Volume is the primary lever. Providers offer tiered pricing at thresholds of 500 kW, 1 MW, 5 MW, and 10 MW+. Committing to a minimum monthly consumption floor (even below your actual draw) can unlock lower rates. Ask about demand-response programs where you agree to curtail during grid emergencies in exchange for 5-15% rate reductions. In markets like the UAE, where grid operators value flexible load, this discount can be substantial.

Negotiating Per-kW Rates

Contract duration is the primary lever. A 12-month commitment might yield $150/kW/month, while a 36-month commitment drops to $120/kW/month. Negotiate for a right-to-grow clause that locks your rate for additional capacity added during the contract term. Also negotiate the overage rate -- some providers charge 150-200% of the base rate for power exceeding your commitment, which can be brutal during unexpected spikes.

Negotiating Flat Rates

The main negotiation point is the included power cap. Standard racks come with 5-8 kW allocations; pushing to 10-12 kW at the same price effectively reduces your per-kW cost. Bundling multiple racks (5+ rack deal) typically unlocks 10-20% discounts. Also negotiate included bandwidth -- upgrading from 1 Gbps to 10 Gbps included eliminates a costly line item.

Hybrid Pricing: The Best of Multiple Models

Increasingly, operators are combining models to optimize costs for mixed environments. A common hybrid approach:

• Base capacity on per-kW: Reserve 80% of your expected power consumption at a committed per-kW rate for budget stability
• Burst capacity on per-kWh: Pay metered per-kWh rates for the top 20% that varies with workload intensity
• Excess/curtailment on spot: Access interruptible power at deeply discounted per-kWh rates ($0.02-$0.04/kWh) for non-critical workloads that can be shed during peak periods

This approach is particularly effective for AI compute operators who run steady inference loads (per-kW base) with periodic training bursts (per-kWh metered) and can absorb cheap excess renewable generation (spot) for batch preprocessing.

Frequently Asked Questions

What is the cheapest colocation pricing model for Bitcoin mining?

Per-kWh metered pricing is typically the most cost-effective model for Bitcoin mining colocation. ASIC miners run at near-constant power draw 24/7, so per-kWh pricing charges only for actual consumption without the overhead markup embedded in per-kW or flat rate models. Current per-kWh rates for mining colocation range from $0.045 to $0.085/kWh in competitive US markets and $0.035 to $0.060/kWh in the UAE. However, per-kWh pricing carries more exposure to utility rate fluctuations, so miners with long time horizons may prefer a fixed per-kW rate for budget certainty.

What does per-kW pricing include at a data center?

Per-kW (kilowatt) pricing bundles all data center costs -- electricity, cooling, physical security, network connectivity, and space -- into a single monthly rate per kilowatt of committed power capacity. A typical per-kW rate ranges from $100 to $200 per kW per month for enterprise colocation, and $60 to $120 per kW per month for high-density mining and AI workloads. The rate covers your allocated power capacity regardless of actual consumption, meaning you pay the same amount whether you use 50% or 100% of your committed power.

How do I calculate the true cost of colocation per server?

To calculate the true monthly cost per server, convert all pricing components to the same unit. For per-kWh pricing: multiply your server's average power draw (in kW) by hours per month (730) by the $/kWh rate, then add any rack space, cross-connect, and support fees. For per-kW pricing: multiply your committed kW by the monthly $/kW rate. For flat rate: divide the total monthly fee by the number of servers. Always add indirect costs like IP transit, remote hands, and insurance. A 3 kW GPU server at $0.08/kWh costs roughly $175/month in power alone (3 x 730 x $0.08), while the same server at $150/kW/month costs $450/month all-in.