What colocation actually costs in 2026 — by rack size, by region, and what's typically included vs. billed separately. Real numbers, no marketing math.
Colocation pricing works two different ways depending on your hardware. Knowing which model applies to you will save you from comparing apples to oranges when you're shopping providers.
Use rack-space pricing ($/U) if you have conventional 1U–4U servers with modest power draw — typically under 500W per U. Web servers, databases, NAS, game servers, dev machines. You're paying for physical space in a rack, and power is bundled to a stated wattage.
Use power-based pricing ($/kW) if you're running high-density GPU servers, AI inference or training clusters, or crypto mining ASICs. A single 8-GPU server can draw 3–7kW — more power than a full rack of conventional servers. At those densities, pricing by rack unit stops making sense. You reserve a power allocation and deploy whatever hardware draws that load.
A standard colocation quote should include all of these. If any of these are absent, ask why — or expect a separate line item.
The sticker price on a colocation plan can be misleading. These are the most common sources of bill surprise:
This is the single biggest hidden cost in colocation. Some providers advertise a low monthly rate but charge $0.01–$0.10 per GB of outbound data. If you're running a file server, game server, media server, or any application moving significant data, this adds up fast.
Some commercial facilities charge $50–$500 for initial rack installation. This covers the technician time to receive, rack, and cable your hardware. Columbia Colocation: $0 setup fee.
Many enterprise facilities require 12–36 month contracts, with early termination fees. A month-to-month arrangement costs more per month but gives you flexibility. Know what you're signing before you ship hardware.
At commercial facilities, a technician visiting your server for a reboot or cable check is often billed at $75–$150/hr with a 1-hour minimum. If you have hardware that needs occasional physical attention, this is a real ongoing cost. At small, ISP-operated facilities, this is often included in the monthly rate because the staff managing the facility handles these requests directly.
Location is the biggest driver of colocation pricing variation within a region. The same 1U of rack space with the same network connectivity can cost 2–4x more depending on where the facility is located.
Commercial real estate in Seattle and Bellevue is expensive. Data centers occupy large, climate-controlled industrial buildings — and in the Seattle metro, those buildings are expensive to build, lease, and operate. Additionally, commercial power rates in King County are higher than in Grant County. These costs flow directly into the rack space pricing.
Grant County PUD provides hydroelectric power at some of the lowest commercial rates in the United States. This is why Microsoft, Sabey Data Centers, and multiple hyperscale operators have built major facilities in Quincy. The power cost advantage translates directly into lower operating costs, which translate into lower prices for customers.
The latency tradeoff is real but small: 10–20ms from Seattle to Quincy vs. under 5ms for a Seattle-area facility. For the overwhelming majority of workloads — web servers, databases, backup targets, game servers, storage — this difference is imperceptible. The cost difference (often 2–4x) is very much perceptible.
For stable, 24/7 workloads, colocation consistently beats cloud pricing. The comparison gets even more favorable when egress fees are included in the cloud side.
| Workload | AWS Monthly Estimate | Colocation Monthly (Quincy) | Annual Savings |
|---|---|---|---|
| 8-core, 64GB dev server + 5TB egress | ~$841/mo | $99/mo bare metal | ~$8,900/yr |
| Game server (64-player) + 10TB egress | ~$1,586/mo | $99/mo bare metal | ~$17,800/yr |
| 1U colo, own hardware (no egress) | ~$180–$400/mo equiv. | $85/mo | ~$1,100–$3,780/yr |
| 20TB S3-compatible storage | ~$460+/mo (before egress) | $200/mo | ~$3,100+/yr |
The economic breakeven point for buying hardware and colocating vs. renting cloud VMs is typically 12–18 months. After that point, cloud is almost always more expensive for predictable, stable workloads.
Drop your email and what you're running — we'll send back a real number, not a brochure. No sales calls unless you want one.
For reference — here's our current rack-space pricing as of 2026. Founder pricing is available for the first 24U deployed and is locked for 24 months from install. All plans include power, unmetered bandwidth on Grant County PUD fiber, static IP, VLAN isolation, and $0 egress fees. For kW-based GPU or high-density compute pricing, see our AI Compute page or contact us directly.
Bare metal server rental (dedicated Dell PowerEdge, 8c/16t, 64GB RAM, 1TB SSD, iDRAC): $99/mo flat, $0 setup, month-to-month.
Object storage (S3-compatible, no egress): $7/mo (500GB) to $200/mo (20TB).
First 24U. Rate locked 24 months. $0 setup. No contract. Quincy, WA on Grant County PUD fiber.
GPU servers and mining ASICs are priced by the kilowatt, not by the rack unit. You reserve a power allocation — power, cooling, space, and connectivity bundled in — and deploy whatever hardware draws that load. This is simpler to budget and typically cheaper for dense workloads because you're not paying for rack space you don't fill.
Most startup AI teams begin with a single multi-GPU server before scaling. Here's what realistic entry-level deployments look like at our Quincy, WA rates:
| Setup | Approx. Power Draw | Monthly at Columbia Colo | Seattle-Area Equivalent |
|---|---|---|---|
| 4× RTX 4090 inference node | ~1.5–2kW | ~$330–$450/mo | $800–$1,500/mo |
| 8× RTX 4090 training server | ~3–4kW | ~$600–$800/mo | $1,200–$2,500/mo |
| Single 8× H100 / A100 server | ~5–7kW | ~$1,000–$1,400/mo | $2,000–$4,000/mo |
| Small cluster (2–3 GPU servers) | ~10–15kW | ~$1,750–$2,625/mo | $4,000–$8,000+/mo |
Pricing scales with commitment size. Larger power allocations get a lower per-kW rate because you're buying in bulk and we're not splitting a circuit across multiple customers.
| Power Commitment | Rate | Example Monthly Cost | Typical Use Case |
|---|---|---|---|
| Up to 5kW | $200–$225/kW | $1,000–$1,125/mo at 5kW | Single GPU server, small inference node |
| 5–15kW | $185–$200/kW | $1,850–$3,000/mo at 10kW | Multi-server AI cluster, mid-scale mining |
| 15kW+ | $175–$185/kW | ~$2,625–$2,775/mo at 15kW | Dedicated deployment, custom arrangement |
Four campuses. Grant County PUD hydro power. Sub-millisecond inter-campus VLAN. Owner-operated, no corporate overhead. Rate depends on power commitment size — contact us for a quote.
If you're running conventional 1U–4U servers — web servers, databases, NAS, game servers, dev boxes — rack-unit pricing is the right model. Power draw is typically well under 500W per U, and you pay for the physical rack space you use. Here's what the market looks like in 2026.
| Rack Size | Seattle / Major Metro DC | Regional / Secondary Market | Columbia Colocation (Quincy, WA) |
|---|---|---|---|
| 1U | $150–$350/mo | $100–$175/mo | $85/mo (founder) · $110 standard |
| 2U | $250–$600/mo | $175–$300/mo | $150/mo (founder) · $175 standard |
| 4U | $400–$900/mo | $300–$500/mo | $260/mo (founder) · $280 standard |
| 10U | $800–$2,000/mo | $600–$1,000/mo | $500/mo (founder) |
| Half Rack (20–22U) | $1,500–$3,500/mo | $800–$1,500/mo | $900–$1,100/mo |
| Full Rack (42U) | $2,500–$6,000/mo | $1,500–$2,500/mo | $1,600–$2,000/mo |
These ranges represent typical all-in monthly rates excluding egress. Seattle-area pricing is elevated due to commercial real estate costs and labor. Quincy, WA pricing is structurally lower because of cheap Grant County PUD hydroelectric power and no metro real estate overhead.