Licensing
We license the method. You keep the hardware, the workload and the control.
ATHLAZ does not build hardware and does not lay cable. This page states the shape of a licence, the order it happens in and what is settled in writing before anything reaches production.
AI-Driven Adaptive Power Technology
ADAPT
A software licence covering the managed accelerators in your fleet, with a support path, a defined removal route and no access to your workloads, models or data.
Intelligent Power Management Module
iPMM
A design licence for silicon, power module and system vendors who cast the method into the power rail itself, delivered with the engineering support that goes with it.
ATHLAZ Power Network
APN
A multi site agreement that treats verified headroom across your estate as one resource inside a national power envelope.
The order it happens in
Four stages, each one earning the next.
No stage asks you to take a figure on trust that the previous stage did not already prove on your own fleet.
Baseline on your own fleet
A measurement on your hardware, your workload and your power envelope. It produces a number you own, before any commitment.
Paired evaluation
A managed arm and a baseline arm run side by side under an equal cap, in a scope you approve, for a window long enough to be believed.
Limited production
A bounded slice of the fleet, under your change control, with reporting your own teams can read and a removal route already tested.
Fleet licence
The licence widens to the fleet once the evaluation holds and your platform, security and change control reviewers have signed it off.
What each side supplies
Clear on day one, so nothing stalls in week three.
Most evaluations slip on access and scheduling rather than on physics. These points are agreed before the first measurement.
- You supply
- The accelerators in scope, the workload, a maintenance window and the reviewers who can approve a production change.
- We supply
- The runtime, the measurement method, the paired arm design, the reporting and the engineers who sit with your team through the window.
- Access taken
- None to workloads, models or data. The runtime sits adjacent to the driver, and removal restores driver defaults.
- Agreed in writing
- The measurement window, the exit and removal route, the support path and how a change reaches production.
- Reported to you
- Power, throughput and latency from the same window, per accelerator, in records your own auditor can check.
- Commercial terms
- Set per engagement against the capacity released on your own fleet. Nothing is quoted from a slide.
Start where it is cheapest to be wrong
The first conversation is about your fleet, not about a contract.
Measured on production NVIDIA H100 NVL under an equal power cap on managed and baseline arms. Your fleet baseline confirms the figure for your site.