The neocloud foralternative chips
We are the deployment arm for heterogeneous compute.
Backed by
Speed
Speed stopped being UX. It became the product.
A chat turn is one pass through a model. An agent run is hundreds or thousands, in sequence, each waiting on the one before it. Latency stops averaging out and starts compounding into wall-clock time — and wall-clock time is the entire product experience.
100–1,000×
model passes per task, as work shifts from chat to agents and reasoning. Cost and latency per pass now compound instead of cancelling out.
Up to 16.1x faster
OpenCode running the identical task on both sides. A real coding session at measured operating points.
GPU cloud today
Prefill vs decode
GPUs are prefill machines. Decode is where they stall.
Prefill is compute-bound and parallel — exactly what a GPU is built for. Decode is memory-bound and autoregressive: one token at a time, limited by bandwidth, not FLOPs. Adding compute does not move it, and the ceiling drops as the model gets larger.
The B300 ceiling sits under
~320tokens / s / user
and materially lower on frontier-scale models. Above it, the GPU simply cannot operate.
Interactivity frontier — MiniMax M2.7
Output throughput per rack against the per-user token rate it sustains.
Everything to the right of the line is a market only purpose-built decode silicon can serve. It is also where agents live.
GC internal benchmark set, 34kW / 16-RDU decode rack, B300 approximated from SemiAnalysis InferenceX.
Disaggregation
We split the workload. Prefill on GPUs, decode on ASICs.
Two different workloads have shared one chip because there was nothing else to put them on. Disaggregate them, run each on silicon built for it, and the cost curve moves immediately.
Prefill
NVIDIA B300 tier
compute-bound, parallel
Decode
Alternative ASICs
memory-bound, autoregressive
Your surface
Bare metal, or one endpoint if you want it
one contract, one set of SLAs
The offer
We buy the chips. You get the metal.
Not a shared endpoint and not a queue. We own the racks, we do the bring-up, and you get dedicated silicon under contract — sited with your fleet, with root on the machines you're paying for.
You pick a latency target and a price. We put the hardware behind it.
Prefill
NVIDIA B300
Compute-bound. We bring the allocation, so you keep the stack you know.
Fast
Alternative ASICs
1–2k tok/s/user on multi-trillion-parameter models. In production today, with more of the ladder opening as each vendor reaches volume.
Fastest
Wafer-scale SRAM
SRAM wafer-scale, at the top of the ladder for latency-critical work.
One contract. One set of SLAs. However many vendors it takes.
What it takes
Four things a vendor needs. We run all four.
Hardware
We own the racks and carry them on our balance sheet, not yours.
Allocation
Price-protected quota across three vendors, placed ahead of merchant availability.
Data centers
~20kW per rack, air-cooled. Drops into existing US colocation. No liquid retrofit, no new build.
Software
Model bring-up on closed stacks, and orchestration of prefill and decode as one service.
Switching
We don't take GPUs away. We bring the allocation for prefill.
Every switching objection assumes we are asking you to abandon a stack you have already paid to build. We are not. The prefill half stays on the GPUs you want, we bring the allocation, and the decode half moves to silicon that can actually serve it.
“A new chip means a new software stack.”
Model bring-up is our obligation, not yours. One OpenAI-compatible API spans prefill and decode; bare metal if you want it.
“We need day-zero model support.”
Written into the SLA as a contracted bring-up commitment with a defined window — not best efforts.
“Our workloads are lumpy. Utilisation risk is ours.”
Reserved decode capacity with burst onto the paired B300 fleet. You are never decode-only, so a quiet week is not a stranded rack.
“We'd be locked to one vendor's chip.”
We hold quota across all ASICs. The lock-in you are avoiding is the one you already have.
Deployment
We handle deployment. The pipeline is ours.
Siting, bring-up and operations are our obligation. You are buying capacity under contract, not standing up a build.
In our colocation pipeline
30MW+
Site
We secure the space, the power and the cooling. Capacity sits alongside your existing fleet where possible, with full location choice where not.
Bring up
Racks, interconnect and model bring-up on closed vendor stacks. The part that makes a new chip expensive is ours, not yours.
Operate
We run the fleet for the life of the contract — one operator across prefill and decode, one set of SLAs.
The business
Every chip that is not a GPU needs a route to market. We're building it.
A more plural hardware base is not a side effect of this business. It is the business.
Tell us the model, the traffic shape and the interactivity target. We'll come back with a rate card and a deployment slot — not a waitlist.