HomeHost ServersLiquid-Cooled Colocation
Bespoke · By reservation · Texas
Reserve rack space 6–9 months before a GR1D site comes online. Tell us your GPUs and what they need in cooling and UPS; our partner builds the cooling and power modules to match, and GR1D stays responsible for your racks' power-on date.
Bespoke liquid-cooled colocation from $180/kW/month for direct-to-chip racks such as NVIDIA GB300 NVL72 and Vera Rubin NVL72. Binding contracts, not LOIs.
Starting at
$180
/kW/month
Bespoke liquid-cooled colocation · Not the cheapest. Available.
How It Works
Everything we build is bespoke. Your liquid-cooled capacity is built after you reserve it, to your rack spec, and ready at a powered shell before your racks arrive.
The whole process, with 2026 timelines and sources: How liquid-cooled colocation works, step by step →
The Contract
A reservation that names no site and no date is a waiting list. Ours names both, and puts the date on GR1D.
GR1D doesn't sign letters of intent. A reservation is a binding contract, because the cooling systems for your racks are ordered once it is signed.
Demand is high, so liquid-cooled colocation delivered in 12 months or less requires a significant deposit. Your contract sets it out with the rest of the terms.
Your reservation is bound to a specific site. You get the address to ship your racks to and the date that site can accept the shipment.
GR1D stays responsible for your racks coming online on time, whichever data center shell they go into, including a partner's site not wholly owned by GR1D, so your racks have space to power on if construction is delayed.
Your Spec
The cooling and power modules are built to these numbers, so they are set when you reserve, not after.
Your platform and rack count: NVIDIA GB300 NVL72, Vera Rubin NVL72 or your own direct-to-chip configuration.
50–100+ kW per rack, on a named basis. A GB300 NVL72 is rated at 132 kW nominal and about 155 kW at peak (HPE, Sep 2026).
Direct-to-chip liquid cooling: coolant supply temperature, flow and CDUs. A GB300 NVL72 rack needs 59–177 L/min at 25–45 °C (Lenovo, Aug 2026).
Redundancy and battery minutes for your workload. AI training swings power in milliseconds, so the UPS is specified with the rack (Schneider Electric, 2025).
Your OEM ship window, matched to the date your site can accept the racks, with the route from dock to position checked.
Positions planned for the next rack: Vera Rubin NVL72 is specified up to 250 kW on 45 °C coolant (NVIDIA, Oct 2026).
The full list for a GB300 hall is in GB300 NVL72 colocation requirements, and the cooling itself in liquid cooling for GPU data centers. Need the GPUs too? GR1D sources servers on request.
Pipeline
Every GR1D site is in Texas, and your reservation is bound to a specific one when you sign.
2027
Our pipeline for 2027 includes several 20 MW sites, where the first liquid-cooled colocation could land.
2028
Larger 75 MW data centers come online in 2028.
Our South Houston 20 MW site has live power and is under contract, so it is not offered for new reservations. See all GR1D sites →
Availability
In 2026 the constraint is capacity, not price. Liquid-cooled space is spoken for before it is built, so the date your racks can power on is set by when you reserve.
1.4%
Primary-market vacancy, H1 2026, a record low (CBRE)
80.4%
Of primary-market space under construction already preleased (CBRE)
2028
Deliveries most tenants signing today are contracting for (JLL)
Sources: CBRE, North America Data Center Trends H1 2026 (Aug 2026); JLL, North America Data Center Report, Midyear 2026 (Aug 2026). How far ahead to reserve for GPUs on order →
Questions
You reserve rack space 6–9 months before a GR1D site comes online and specify your GPUs, cooling and UPS. A partner builds cooling and power modules to that spec, GR1D binds your reservation to a site with a ship-to address and an acceptance date, and the modules are installed at the powered shell before your racks power on. The six steps, in detail →
Your GPU platform and rack count, such as NVIDIA GB300 NVL72 or Vera Rubin NVL72; power per rack, from 50 to 100+ kW; coolant supply temperature and flow; the CDU arrangement; and your UPS requirement. Everything GR1D builds is bespoke, so the cooling and power modules are built to those numbers. Need GPUs too? GR1D sources servers on request. What a GB300 NVL72 rack needs →
Bespoke liquid-cooled colocation starts at $180/kW/month. The price depends on density, term and the cooling and power you specify. GR1D sells on availability, not on being the cheapest: in the first half of 2026, primary-market vacancy hit a record-low 1.4% (CBRE).
No. GR1D signs binding contracts, not LOIs. Once a reservation is signed, GR1D binds it to a site and orders the cooling systems for your racks, so the commitment is binding on both sides from the start.
Yes. Because demand is high, GR1D requires a significant deposit for liquid-cooled colocation delivered in 12 months or less. Your contract sets out the deposit with the rest of the terms.
When your reservation is bound to a site, GR1D gives you the address to ship your racks to and the date that site can accept the shipment. Book your OEM delivery window and freight against that date.
GR1D stays responsible for bringing your racks online on time, whichever data center shell they go into, including a partner's site that GR1D doesn't wholly own. That way your racks have space to power on even if a construction schedule slips. How far ahead to reserve for GPUs on order →
In Texas. GR1D's 2027 pipeline includes several 20 MW sites, where the first liquid-cooled colocation could land, and larger 75 MW data centers come online in 2028. Your reservation is bound to a specific site when you sign. Our South Houston 20 MW site is under contract and not offered for new reservations. All GR1D sites →
GPU colocation pricing and lead times on Host Servers; more answers on the FAQ page.
Go deeper
Get Started
Placing GPU orders? Reserve your rack space today for liquid-cooled GB300 and Vera Rubin. Tell us your GPUs, delivery date and power per rack.
Walk through your GPU delivery dates, rack density, cooling and UPS with the GR1D team.
Or email us directly
AI data center infrastructure — from raw land to GPU-ready racks. Build, host, or rent compute at scale.
South Houston, TX
© 2026 GR1D. All rights reserved.
AI Data Center Infrastructure — Build, Host, Compute