High Density Colocation: What It Is & How It Works
High density colocation is a data center service that provides significantly more power and cooling per rack than standard colocation, typically supporting 10 kW to 50+ kW per rack, compared to the 2-5 kW of traditional setups. You rent space, power, and cooling for your servers, but with the infrastructure designed to handle the intense heat and electrical load of modern AI, GPU, and high-performance computing (HPC) workloads. This article explains how high density colocation works, what to look for in a provider, and the common challenges you’ll face.
Background: Why Standard Colocation Isn’t Enough
Traditional colocation was built for the era of 1U or 2U servers with modest power draws. A standard rack might have a 5 kW power allocation, which is fine for a few dozen web servers or storage arrays. But the rise of GPU-accelerated computing (for AI training, scientific simulations, and rendering) changed the game. A single server with 8 GPUs can draw 10-14 kW by itself. When you put several of these in one rack, you quickly exceed the power and cooling capacity of a standard cage or cabinet.
High density colocation emerged as a response. Providers install specialized power distribution (e.g., 415V AC or 48V DC), high-capacity cooling (liquid cooling, rear-door heat exchangers, or in-row cooling units), and heavier floor loading to support the weight of dense equipment. The goal is to let you pack more compute into a smaller footprint without risking overheating or tripping breakers.
How High Density Colocation Works
High density colocation is not just about a bigger power feed. It involves a coordinated set of infrastructure choices:
- Power delivery: Instead of a standard 30A or 60A single-phase circuit, high density racks often use 3-phase power with capacities from 10 kW to over 100 kW. The provider must have the upstream utility capacity and backup generators to match.
- Cooling systems: Air cooling struggles above about 15-20 kW per rack. High density facilities use liquid cooling (direct-to-chip or rear-door heat exchangers), in-row cooling units, or a combination. Some offer a choice between air and liquid, depending on your hardware.
- Floor and rack design: Dense racks weigh more. The raised floor or slab must support the load, and the racks themselves are often wider or deeper to accommodate cable management and cooling connections.
- Monitoring and management: You get granular power and temperature monitoring per rack, often with API access, so you can track utilization and avoid hotspots.
When you sign up, you specify your power requirement (e.g., 20 kW per rack) and cooling preference. The provider then provisions a dedicated or shared cooling loop and ensures the power path is redundant (e.g., A/B feeds).
How to Choose a High Density Provider
Not all colocation providers are equal when it comes to high density. Here are the key factors to evaluate:
- Power capacity and redundancy: Check the total facility capacity and whether your rack will have N+1 or 2N redundancy for power and cooling. Ask about the upstream utility feed and generator runtime.
- Cooling technology: Does the provider support liquid cooling? If your hardware uses direct-to-chip cooling, you need a provider with a liquid loop. If you’re air-cooling, ask about the maximum kW per rack they can handle without hot spots.
- Floor weight limit: A rack with 8 GPUs can weigh over 1,000 kg. Confirm the floor can support that without reinforcement.
- Contract flexibility: Some providers charge per kW, others per rack. Make sure the pricing model matches your growth plans. Also, check if you can scale up power without moving racks.
- Connectivity and latency: For AI workloads, you often need low latency to other compute or storage. Check the provider’s network options and whether they have direct connections to cloud providers or research networks.
Common Problems and How to Avoid Them
High density colocation comes with its own set of challenges. Here are the most common and how to mitigate them:
- Hot spots: Even with high capacity cooling, poor airflow management can cause localized overheating. Use blanking panels, manage cable openings, and ensure your servers are arranged for front-to-back airflow.
- Power spikes: GPU workloads can draw sudden power surges. Ensure your rack’s power allocation includes headroom, and consider using power capping software to avoid tripping breakers.
- Cooling failure: If a liquid cooling loop fails, you have minutes before shutdown. Ask about the provider’s maintenance procedures and whether they have a fallback air cooling system.
- Cost overruns: High density power is expensive. You may be billed for the power capacity, not just the actual usage. Understand the billing model and monitor your consumption to avoid surprises.
High Density vs. Standard Colocation: Comparison
The main differences boil down to power, cooling, and cost:
- Power per rack: Standard colocation offers 2-5 kW per rack; high density offers 10-50+ kW.
- Cooling method: Standard uses room-level or row-level air cooling; high density often uses liquid cooling or in-row units.
- Cost per rack: High density racks cost significantly more per month, but the cost per kW may be lower than standard because you’re using the space more efficiently.
- Hardware compatibility: Standard colocation works for most enterprise servers; high density is necessary for GPU servers, HPC clusters, and high-frequency trading systems.
Costs: What to Expect
High density colocation pricing is typically quoted per kW per month, rather than per rack. For example, a provider might charge $150 per kW per month for a 20 kW rack, which comes to $3,000 per month. This often includes power, cooling, and network connectivity, but not the hardware. Some providers also charge a one-time setup fee for the power and cooling installation. You should always ask about:
- Whether the price includes power usage or just capacity.
- Overage charges if you exceed your power allocation.
- Cost of additional cooling (e.g., liquid cooling loop).
- Contract length and penalties for early termination.
Safety and Reliability Considerations
High density environments pose unique safety risks. The high electrical currents require proper circuit protection and grounding. The cooling systems, especially liquid cooling, must have leak detection and shutoff valves to prevent water damage. Providers should have fire suppression systems (e.g., inert gas or water mist) that are safe for electronics. When choosing a provider, ask about their safety certifications (e.g., Uptime Institute Tier level) and their maintenance track record. For your own equipment, ensure your servers are certified for the operating temperature range and that you have remote monitoring to alert you to thermal or power issues.
Alternatives to High Density Colocation
If high density colocation isn’t the right fit, consider these alternatives:
- Managed hosting: The provider owns the hardware and manages it for you. This can be simpler but less flexible.
- Cloud GPU instances: For AI workloads, cloud providers offer GPU instances without the hassle of physical infrastructure. This can be more cost-effective for bursty workloads.
- Edge data centers: Smaller facilities that offer higher density in a smaller footprint, sometimes closer to your users.
- On-premises data centers: If you have the capital and expertise, building your own high density facility gives you full control, but requires significant investment.
Frequently Asked Questions
What is considered high density in colocation?
Generally, high density refers to racks with power allocations of 10 kW or more. Some providers consider 20 kW and above as high density, while standard racks are typically 2-5 kW.
Can I use air cooling for a 20 kW rack?
It’s possible but challenging. With proper hot/cold aisle containment and in-row cooling, some facilities can handle up to 20 kW per rack with air, but above that, liquid cooling is recommended. Always check the provider’s specific capabilities.
How much does high density colocation cost?
Pricing varies widely by region and provider. You might pay $100-$300 per kW per month, meaning a 20 kW rack could cost $2,000-$6,000 per month. This usually includes power, cooling, and space, but not the servers.
What is the difference between power capacity and power usage?
Power capacity is the maximum amount your rack can draw without tripping breakers. Power usage is what your servers actually consume. Providers may bill based on capacity, usage, or a combination. Make sure you know which.
Is liquid cooling safe in a colocation facility?
Yes, when properly installed. Providers use leak detection, drip trays, and shutoff valves. The cooling fluid is typically non-conductive, but it can still damage equipment if leaked. Ask about the provider’s leak prevention and response procedures.
Can I upgrade from standard to high density later?
It depends on the facility. Some have the infrastructure in place, while others may require you to move to a different area or even a different data center. Check with your provider about upgrade paths before signing a contract.
Related
Method: the standard formula for this calculation, rounded as each result states. See how the calculators work or report an error.