Real options. Real terms. In writing.
Not twelve introductory calls. Not a list of logos. A shortlist of facilities that can genuinely carry your load, with usable kW, the cooling architecture, the cross-connect model, the commercial terms, and the things the operator would rather you found out later.
How a megawatt actually gets placed.
Requirement
IT load, density per rack, cooling architecture, market, in-service date, term, growth. We will interrogate the power basis before anything else, because the gap between IT load and total facility power is roughly a quarter to a third of the cost, and it is where most colocation budgets quietly break.
NDA and non-circumvention
Executed before a single facility is named to you, or you are named to a single facility. This is not a formality and it is not negotiable. Brokered deals live or die on whether every link in the chain is documented.
Qualified RFI
We issue a structured RFI to a shortlist that has already been filtered on density tolerance, tier, power availability and delivery date. The RFI forces the operator to state usable kW, power basis, cooling capability, cross-connect and bandwidth pricing, remote-hands rates and escalators — before we discuss headline price.
Shortlist and negotiate
You get real options with real numbers and a plain reading of the trade-offs. Then we negotiate: rate, term, escalator, ramp schedule, SLA, power commitment, and contiguous expansion rights — which are the cheapest thing to secure at signature and the most expensive thing to fix afterwards.
Deployment
Freight, customs, delivery, fit-out, rack and stack, PDU and busbar, cross-connect, fabric bring-up, burn-in. If we supplied the hardware, this is one team. If you supplied your own, we coordinate the same way.
5.2 megawatts, 256 nodes, one date.
A representative AI deployment of the kind we handle: 256× 10U air-cooled 8-GPU HGX B200 nodes. 2,048 GPUs. Approximately 5.2 MW.
That is not a colocation requirement most facilities can answer honestly. A large part of the qualified estate cannot take the rack density in air. Of those that can, fewer still have contiguous, contracted power available on the date the client needs it, at a rate that lets a neocloud make money on tokens.
The work is in the filtering, the power-basis discipline, and the negotiation. And then in getting 256 nodes through customs, into a hall and into burn-in without losing a week.
Representative of the scale EOGSB works at. Client identities and facility names are never disclosed without an executed NDA.
A facility is not qualified because it says it is.
Power basis
IT load or total facility power, stated explicitly and in writing. Every time. No exceptions.
Usable kW
What the rack can actually draw sustained, not the breaker rating on the PDU spec sheet.
Density tolerance
Air, rear-door HX, or direct-to-chip liquid — with CDU capacity, coolant loop and water-side infrastructure evidenced.
Tier and redundancy
Tier III+ and N+1 minimum on power and cooling. Concurrently maintainable, or it does not go on the list.
Connectivity
Carrier density, cross-connect pricing model, IP transit, and whether the "great rate" survives adding A+B feeds and six cross-connects.
Delivery reality
Whether the power is contracted or aspirational, and whether the fit-out date has permitting risk sitting behind it.
Operator standing
Who owns the facility, who owns the power contract, and what happens to your tenancy if either changes hands.
Exit and growth
Contiguous expansion rights, ramp schedule, escalator caps, and what it costs to leave.
Brokerage, answered.
How long does a GPU colocation deal take to close?
From a clean requirement to a signed contract, typically four to ten weeks depending on market and scale. The long poles are almost never the negotiation — they are power confirmation, density validation and the client's own approval process. Sites move faster when the requirement arrives complete.
What are the usual traps in a colocation contract?
Usable versus contracted kW. Power basis stated as total facility power rather than IT load. Cross-connect and bandwidth pricing left out of the headline rate. Remote hands billed at rates that only appear at month three. Escalators. And no contiguous expansion right, which turns your growth into a second, worse deal in a different hall.
Can EOGSB act as principal rather than broker?
Yes. On integrated deals we can take the hardware and the hosting on our own paper and deliver the client a single contract with a single in-service date. This is common where the end client wants one throat to choke and no exposure to the underlying supply chain.
Do you work with clients who already own their hardware?
Yes. Plenty of our brokerage work is pure capacity placement for clients who bought their GPUs elsewhere. We do not require you to buy hardware from us to use the brokerage.
Send us the requirement. We come back with options.
Node count, GPU class, kW per rack, target market, in-service date. That is enough for us to open the conversation with the right suppliers and the right facilities. Under NDA from first contact.