Many facilities won’t, and they won’t tell you upfront. Some have AUPs that exclude consumer cards; others have infrastructure designed only for server-grade hardware. We know which facilities accept consumer and prosumer cards so you don’t spend a week getting rejected after the first intake form. If you’re running RTX 4090/5090s or RTX 6000 Pro, say so in your specs and we route accordingly.

Air cooling with containment or rear-door heat exchangers handles up to ~25-30 kW per rack reliably. Above that, you need liquid. B200 and GB300 clusters typically require DLC (direct-to-chip) at 40-80+ kW/rack. If you’re not sure what your hardware peaks at, give us the GPU model and quantity and we’ll tell you what cooling you actually need.

Committed kW is the default billing model. You reserve a kW amount and pay for it whether you use it or not. For inference workloads that idle most of the month and spike occasionally, this is a significant overpay. We specifically look for facilities that offer true metered per-kWh billing for variable workloads. They exist, they’re just not the majority. Tell us your utilization pattern and we’ll filter accordingly.

Yes. Most Tier 1 operators now gate small deployments behind 50-100kW+ minimums. We route to mid-tier and regional facilities that accept single-node and 1-2 cabinet deployments without those gates. The regional providers with available capacity for small GPU deployments often don’t publish this publicly — they fill capacity through referrals. That’s the inventory we access.

We match to facilities based on what your hardware actually requires. B200 clusters need 40-80 kW/rack with DLC. GB200 NVL72 is a single-rack system at ~100+ kW with integrated liquid cooling — very few facilities outside purpose-built AI campuses can host it. B300/GB300 push 80-120+ kW/rack with DLC or immersion required. When you give us your SKU, we check usable kW at the rack level (not breaker nameplate), CDU manifold installation status, and 800G networking availability before shortlisting. ‘Liquid-ready’ on a sales deck doesn’t clear the filter.

Most Tier 1 operators (Equinix, Digital Realty) now gate single racks behind 100kW+ minimums, especially in Ashburn and Dallas. We specialize in mid-tier and regional providers that accept 5-50kW GPU clusters without MW-scale minimums. We check live availability when you submit specs.

Committed kW at $130-$250/kW/month is standard. For a 300kW deployment in a secondary market, that’s a ~$23,000-$39,000 base MRC before cross-connects and bandwidth. A recent Colorado secondary market quote came in at $78/kW base, with the all-in cost running 25-30% higher after cross-connects. Metered billing exists and is preferable for variable inference workloads — we know which facilities offer it.

No. Air with containment handles up to 25-30kW reliably. Above that, liquid (DLC or immersion) is required. Adding liquid cooling typically adds $500-$2,000/month (CDU rental, leak sensors, maintenance) on top of the base rack rate. We verify whether a facility’s liquid cooling is actually installed and operational, not just ‘available on request.’

2-6 weeks for air-cooled single-rack inference in a secondary market with available power. 3-6 months for primary-market liquid-cooled deployments in power-constrained metros. Colorado, Atlanta, Phoenix, and Ohio typically run 4-8 weeks. When we send a quote, available power is confirmed, not estimated.

GPU colocation is typically 40-60% cheaper than equivalent cloud for steady-state workloads. An 8x H100 node on AWS P5 runs ~$3.90/hr, or ~$7,000/month at 80% utilization. A comparable colo rack runs $2,500-$5,000/month all-in, with break-even usually at 6-9 months. For inference at scale, the math is even more favorable. Hybrid is common: train in cloud, infer in colo.

10Gbps baseline is usually included. 40-100Gbps dedicated ports add $500-$2,000/month. Cross-connects to cloud on-ramps (AWS Direct Connect, Azure ExpressRoute) run an additional $1,000-$3,000 depending on the market. For InfiniBand or RoCEv2 clusters at 800G, verify MMR compatibility before committing.

Dallas and Phoenix lead on power availability and cost ($130-$160/kW). Colorado offers the best pricing at equivalent density for secondary-market deployments (~$78-$140/kW base). Ashburn and Northern Virginia have the best carrier ecosystem but constrained power and high pricing. If latency to a specific carrier hotel isn’t the driver, secondary markets typically run 20-35% less.