An Energy Layer for Edge Computing
VENDOR.Max is a 5 kW-class system: an additional energy layer for compute load on an existing distributed site or on a small standalone edge node.
After initiation, the VENDOR.Max architecture does not provide for continuous external power input to the load supplied through the unit’s normal customer power interface.
Compute is moved closer to the network, the machine or the source of data. It can only be moved to a site that is able to carry the additional electrical load. The next step is a check: whether your load falls within this scale, and whether the architecture of your site allows it.
Compute Moves. The Power Budget Does Not Move With It
The question is rarely whether servers are available. It is whether the site can carry them.
Compute is placed next to the function it serves: next to a production line, a network node, a pumping station, a camera, a sensor. The reason is usually latency, data volume or a requirement to process data locally.
Such a site already has a reason to exist: location, connectivity and an operational function. What its power architecture may not have been designed for is the added compute load.
What may be missing is spare electrical capacity for the new load. And the constraint does not necessarily sit at the service entrance. It may be at the transformer, the rectifier, the uninterruptible power supply, the distribution bus, the storage element, a protective device, a cable, or in the thermal envelope of the cabinet. Before power can be discussed, it has to be known where the margin of that particular site actually ends.
A distributed site also has a second practical question: where the energy for the added compute load has to come from every day. If its operation depends on a continuous external energy supply, then the dependency on that supply is moved onto the site together with the compute.
For a load carried by the VENDOR.Max energy layer, normal operation of the unit does not require continuous external power input after initiation. This changes the character of operation: instead of a continuous energy supply, what remains is start-up, scheduled service and replacement of wear components as needed.
Start-up requires initiation. In the standard configuration the built-in battery becomes an internal operating buffer after start-up; it has a finite service life, as do the other wear components of the unit. The equipment requires scheduled maintenance.
This concerns the energy layer of the added compute load. It is not a disconnection of the whole site from its existing energy supply: base loads, the grid connection and the backup architecture remain in their roles.
Not Every AI Power Figure Is the Same Kind of Number
The question of how many kilowatts edge AI needs has no answer until two things are named: the level of infrastructure and the kind of quantity.
| Level | What is usually published | What it means |
|---|---|---|
| Accelerator | Maximum thermal design power | A property of the component, not the power of the server |
| Embedded module | Configurable operating modes | Operating draw of the module, not the load of the node |
| Platform or server | Power supply rating | A capability class of the supply, not the actual load |
| Rack | Power shelf rating | A power architecture of a different scale class |
Quantities from different rows cannot be added together and cannot be compared as though they were the same load.
One error is worth separating out. A power supply rating does not establish the actual consumption of a server. Consumption depends on the specific configuration and operating mode, and it should be taken from the manufacturer data for that configuration or measured. Equipment manufacturers state this themselves, referring the reader to separate planning tools for real consumption.
The practical rule for your site follows from this: qualification starts from a measured or manufacturer-stated load profile, not from a rating plate on an enclosure.
One Compute Function, Two Integration Paths
The electrical interface is determined by the selected compute equipment and the site architecture. VENDOR.Max configures its output layer to the agreed customer interface.
Standard 230 V AC interface
The conventional path for existing server and edge equipment: power through the equipment’s standard factory power supply, using an interface the site already works with.
Direct DC interface
For equipment and sites where DC is the native input. The selected VENDOR.Max configuration supports low-voltage DC, and the output layer is configured to the agreed nominal voltage within that supported class.
This is not a theoretical fork. Some edge server platforms are offered with factory power supply options for −48 V DC. The choice of interface can therefore be a configuration item of the compute equipment itself.
Suitability of direct DC integration is determined by the permitted input range of the selected equipment, the site power architecture and the supported output class of the selected VENDOR.Max configuration. High-voltage DC interfaces, including the approximately 400 V class, are outside the current envelope.
The interface is determined by the site and the compute equipment, not by the energy layer. For the definition of the unit itself and its interfaces, see VENDOR.Max.
Where the Energy Layer Enters the Site Architecture
The layer is integrated at an agreed point of the site architecture and forms the customer interface for the compute load. The existing loads of the site stay where they are.
Existing protection systems and the backup architecture are retained unless the pilot specification explicitly provides otherwise. VENDOR.Max is assessed as an energy layer and is not introduced by removing safeguards that are already in place.
Placement and pilot packaging are adapted to the site. The unit may stand adjacent to the compute equipment or receive an enclosure engineered for the particular site. This is an engineering decision for a specific deployment, not an existing serial rack-mounted product.
VENDOR is not the compute, network or cooling infrastructure of the site. The subject of integration is the energy layer. Servers, accelerators, racks, cooling, networking, storage and the software stack remain with the operator, the equipment manufacturer or the integrator.
Which Sites Qualify
Suitability is determined by properties of the site, not by industry.
- the added demand is in the kilowatt class
- the load profile can be measured or obtained from the equipment manufacturer
- the electrical input of the equipment is defined
- the backup architecture of the site is known
- the site has an operational reason for local processing
- thermal conditions can be established
- installation space is available
- demand at hyperscale or multi-rack scale
- load neither defined nor measured
- no accessible electrical interface
- production deployment expected without a pilot
If the underlying site is a telecom site, the requirements of the site power architecture itself are addressed separately on Telecom Tower Power.
Ten Parameters for Assessing a Site
Ten items are needed to check the fit. These are technical parameters of equipment and site, not a request for the commercial terms of a project.
- Compute equipment: platform or server model, accelerator configuration.
- Continuous load of the compute equipment.
- Peak and transient behaviour.
- Electrical input of the equipment: AC or DC, nominal and permitted voltage, redundancy of the equipment inputs; for DC additionally the voltage class and range.
- Power and backup architecture of the site: service entrance, distribution, uninterruptible power supply, storage or other backup branch and its topology.
- Cooling load and auxiliary load.
- Required operating regime: hours, duty cycle, expectations regarding backup.
- Site conditions: temperature, humidity, dust and enclosure configuration.
- Available space and installation conditions: cabinet, equipment room or outdoor location, available footprint, mounting method, placement constraints.
- Measurement point and access to it.
Items four and five belong to different planes. Redundancy of the equipment inputs and the backup architecture of the site are not the same thing.
Transient behaviour. The transient profile of the compute load is part of the assessment and is verified during the pilot together with the existing backup architecture of the site.
Site conditions. These form part of engineering matching. The permitted environmental envelope of a specific configuration is established before the pilot begins.
Electromagnetic compatibility. Taken into account during engineering matching of the site and verified during the pilot together with the installed equipment.
From Load to a Measured Deployment Envelope
A preliminary fit established from site data remains an assumption until it is measured. The pilot turns it into a set of measured quantities for that particular site.
The following are recorded during the pilot:
- delivered power
- load transitions
- voltage behaviour at the agreed measurement plane
- interaction with the existing backup architecture
- compute availability during the test
- consumption of cooling and auxiliary loads
- operating time during the test
- interventions
- electromagnetic compatibility at the agreed boundary and the actual behaviour of the installed equipment
- environmental conditions throughout the test
- base load of the site
One of the key operational outcomes of a pilot is data on how a particular load and environment affect the need for intervention and the service regime. This becomes the basis for refining that regime; long-term service life is established through extended operation, not through a single run.
The result of a pilot is a measured deployment envelope for that site and a list of requirements for the next decision. Economic effect and payback are not outcomes of this stage and are not calculated on this page.
Current technology readiness level: TRL 4 — Prototype Rebuild After Relocation
Frequently Asked Questions
Is this intended for data centres?
No. This page addresses distributed edge nodes and power for small sites. Hyperscale and multi-rack power architecture is a different scale class and is not the subject of this page.
Is continuous external power required after start-up?
After initiation, the VENDOR.Max architecture does not provide for continuous external power input to the load supplied through the unit’s normal customer power interface. Start-up does require initiation; the built-in battery becomes an internal operating buffer after start-up and has a finite service life; the equipment requires scheduled maintenance. These are periodic service actions, not a continuous energy supply. Data on the need for intervention under a particular load and in a particular environment is collected during the pilot; long-term service life is established through extended operation.
Can VENDOR.Max affect compute or network equipment?
Electromagnetic compatibility was addressed in the engineering of VENDOR.Max during development and is treated as part of the engineering of a specific deployment. Equipment placement, shielding, filtering, earthing and cable interfaces are assessed for the site. Compatibility with the installed equipment is confirmed by measurement during the pilot rather than assumed in advance. For a description of how the unit works, see How it works.
Can edge servers be powered directly from DC?
Some edge server platforms are offered with factory power supply options for −48 V DC. Suitability of direct integration is determined by the permitted input range of the selected equipment, the site power architecture and the supported output class of the selected VENDOR.Max configuration.
Does this replace the uninterruptible power supply or the energy storage system?
No. Existing protection systems and the backup architecture of the site are retained unless the pilot specification explicitly provides otherwise.
How much data is needed before a pilot?
The ten items listed in the assessment section. These are technical parameters of equipment and site, not a request for the commercial terms of a project.
Can compute be added to a site that is already in operation?
That is the primary scenario of this page. The question comes down to whether the added load falls within the power envelope of the site, and what the standard electrical input of the selected equipment is.
Site Assessment
Send the ten parameters of your site and compute equipment. The initial assessment will show whether the site falls within the current VENDOR.Max envelope and which questions need to be closed before a pilot.
