📡 This Week in Colo

This week, we’re taking a break from our regularly scheduled GPU programming. Not in protest of Jensen’s extracurricular activities, we just happen to have a ton of colo space available right now.

🔥 Weekly Deals

Manhattan, NY | 500 kW, Available Now

The largest available colo space in Manhattan. Air-cooled. Built for latency-sensitive tenants: trading firms, fintech, inference, and interconnection-heavy networks. View listing (GM-285)

Mountain West | 1 MW, Available Now

Air-cooled at 16 kW per cabinet. Concurrently maintainable, diverse carriers, 24/7 remote hands. Good for Air-cooled B300s. View listing (GM-270)

Ohio | 600 kW, Available Now

Tier III-certified design, carrier- and cloud-neutral with diverse fiber entry, guarded campus. Offered as a single 600 kW block. View listing (GM-274)

Northeast US | 500 kW, Available Now

Freed up by a cancelled order. Air-cooled, ~17.5 kW per rack, racks already installed. Move-in ready. View listing

Chicago, IL | 1.2 MW, Ready in 120 Days

Air-cooled up to 17 kW per cabinet. N+1 1,200-ton chilled water plant, hot/cold aisle containment, no liquid retrofit required. View listing (GM-286)

Mountain West | 750 kW, Available Now

Single-tenant 4,000 SF private data hall with new UPS and generators. Up to 15 kW per rack. Immediate occupancy. View listing (GM-287)

San Francisco, CA | 500+ kW, Available Now

Downtown SoMa. Good for AI startups. 100 Gbps network in place, flexible term. Good fit for B300s. Ready to move in today. View listing (GM-284)

Michigan | 580 kW, RFS in ~2 Weeks

Enterprise data center with dual utility feeds, N+1 critical power, 24/7 NOC, and a private link to a second Michigan facility for replication. View listing (GM-314)

Michigan | 300 kW, RFS in ~2 Weeks

Private cages and cabinets in an enterprise-class facility. Redundant power, diverse fiber, 24/7 surveillance. Pair with the 580 kW above for 880 kW total. View listing (GM-313)

Looking for something bigger? We have quite a few large sites available in 2027-2028

Have a friend looking for colo? Submit a referral.

📊 Market Trend: landlords are getting picky 🧐

Data centers used to take anyone with a PO and a pulse. Not anymore.

After a rough few months for neoclouds (and Elon predicting "guppy-level death" for 200+ of them), landlords are underwriting tenants like banks. Investment grade credit, letters of credit, parent guarantees, months of rent prepaid. Some operators are capping how much neocloud exposure they'll take at all. Others just stop returning calls.

If you're a new inference company with real revenue and a Series A, that's a tough room. You need 500 kW to 2 MW, you need it in the next 60 days, and you don't have a balance sheet a landlord wants to lean on.

The good news: mid-sized blocks are still out there. Operators with freed-up or stranded capacity will work with smaller tenants when the deal comes in clean and someone vouches for it. That's the list above.

Meanwhile, everyone wants Vera Rubin in Q2.

A big chunk of the colo requests hitting our inbox right now are for Q2, liquid-cooled, built for Rubin.

Problem is, Rubin already slipped once this year (thermal lids, HBM4 qualification), and the first racks are going where NVIDIA allocation always goes first: hyperscalers and the big labs. If you're not on that list, our bet is Q2 delivery turns into Q3. Maybe Q4.

How we see it playing out:

  • Landlords won't sit on liquid-cooled halls for hardware that doesn't exist yet. Expect reservation fees, bigger deposits, and space going to whoever shows up with servers first.

  • Teams that sign Q2 leases and get Q4 hardware will pay rent on empty halls for six months. Ouch.

  • The winners deploy B200s and B300s now in space that's ready today, start earning, and roll into Rubin when it actually lands.

Need space this quarter, or want to lock in Rubin-ready capacity without paying for an empty room? Contact your favorite GPU Mart representative.

Until next time,

Jack Duhamel
President, GPU Mart