Compute, concentrated.
US adjacent reserved GPU clusters.
Reference architecture sized, reserved by the node or the full block. Built for AI labs, startups and neoclouds that need dedicated GPUs with low latency to their US users.
We buy the GPUs, rack them in border metros, and hand you a reserved cluster.
We handle the space, power, cooling, network and software around NVIDIA B300. You get dedicated hardware on bare metal, on a contract a US team already knows how to sign.
Hardware
We buy and own the NVIDIA HGX B300 nodes, deployed in 256 GPU blocks.
Colo space
We place every block in colocation data centers in border metros.
Power and cooling
Power and cooling sized for B300 density, with the colocation operator.
Networking
An NVIDIA Spectrum X fabric in the POP and low latency paths to US hubs.
Software
Managed Kubernetes on bare metal, GPU ready, with monitoring and support.
No quota request. Your nodes, ready.
What a reserved block looks like at handover: every node burned in, the fabric validated, your clusters ready. See the bar every cluster clears on the platform page.
Five ways to slice one block.
From a single reserved node to the full dedicated block. Pick a size here and it carries into the request form.
| SKU | GPUs | Nodes | Typical use | |
|---|---|---|---|---|
| 8 | 1 | Single node reserved | select | |
| 16 | 2 | Small reserved | select | |
| 64 | 8 | Mid training / serving | select | |
| 128 | 16 | Serious reserved pod | selected | |
| 256 | 32 | Full block, dedicated or wholesale | select |
Serious reserved pod
Enterprise reserved from 12 months. Wholesale and capacity partner blocks run 24 to 36 months take or pay.
Nearshore power, US latency, reserved for you.
We own the hardware and the path to the US. You get the result.
- 01 Compute
A block in a border metro
Dedicated B300 nodes in colocation next to the border, on one fabric.
- 02 Network
Low latency to US hubs
Typically sub 20 ms round trip to the Central US.
- 03 Contract
Bare metal, or your own endpoint
One contract, one set of terms, one support team.
In the first 12 months, built in three steps on one standard.
- Q2 2027
Block 1256 GPUs
First block live. 256 GPUs in service.
- Q4 2027
Block 2+256 GPUs
Second block. 512 GPUs in service.
- Q2 2028
Phase 3+512 GPUs
Two more blocks. 1,024 GPUs in service.
The first block goes live in Q2 2027.
Colocation in border metros, with power and cooling matched to B300.
NVIDIA HGX B300, bought and owned by dense².
Texas corridor. Sub 20 ms to Dallas and Houston.
West coast corridor toward San Diego and Los Angeles.
Central Mexico data center cluster.
Southeast edge toward Florida and the Caribbean.
For workloads that require US data residency.
Nearshore capacity for the isthmus and the Caribbean basin.
Long term capacity on low cost power markets.
Who it is for, and who it is not.
We would rather lose a deal on the first call than on the compliance review.
Strong fit
- ✓US startups, AI labs and growth stage tech running training and inference
- ✓US neoclouds and aggregators needing wholesale or white label overflow capacity
- ✓Cross border AI, from industrial and logistics to enterprise and consumer apps
Weaker fit, we will say so upfront
- ✕Deals that require US data residency, FedRAMP, or strict healthcare / government US only hosting
- →Those belong on a future US POP, or a split design: 256 in Monterrey + 256 in the US
Honest framing. This is Nearshore Americas, not a US data residency region. Residency is Mexico and we disclose it clearly. Ideal when your policy allows Mexico hosted compute and you want Texas class latency with reserved B300 capacity.
Secure reserved capacity on the first block.
Tell us the size, the term and the start window. We come back with a rate, a term sheet and a delivery window. The first block goes live in Q2 2027.
Tell us your target. We return a delivery window and a term sheet.
Size, buyer type and where your users are. That is all we need for a first answer.
GPU hours and reserved clusters only. We do not sell tokens.