The output
Produced.
Inference, produced continuously by the Factory.
Practice 02 / Edge AI Factories
The work is already producing something.
Movement. Readings. Images. Events. Exceptions.
An Edge AI Factory turns what the operation produces into intelligence it can act on, where that intelligence can participate in the work.
Below · the operation, at work
Above · what the work already produces
Representative depiction. Mark spacing carries no rate, volume or yield, and no operation, customer or deployment is depicted.
Factory · operating context
CMS field · media.factory.context · full bleed
Operational intelligence, produced inside the operation.
FACTORY production states
What the work makes
Signal
Movement, readings, images, events, exceptions. Produced at the work, in conditions nobody else sees as clearly.
What the Factory produces
Inference
Produced continuously from the work the capability was established to perform.
What it may do
Authority
Resolved before anything runs, for people and machine actors on the same terms.
How the operation meets it
Interaction
Usable by the people running the work, under the authority the operation sets.
What can be shown after
Evidence
What happened, what did not, and the authority each was decided under.
A production system, not a delivery sequence. Every state is present continuously while the Factory is established, and none is a stage the operation passes through and leaves.
FACTORY production
Give the operation something to produce.
Infrastructure gives an operation somewhere to run things.
An EdgeQub Factory gives it something to produce.
The Factory continuously produces inference from the work it has been established to perform.
Above the line · what the Factory produces
Below the line · the operation, doing what it does
Representative depiction. Mark spacing carries no rate, volume or yield. No result, performance, deployment or customer is asserted or implied.
Above the line is what the Factory produces. Below it is your operation.
The value sits in the second half, which is why the first half has to be where the work is.
Factory · production
CMS field · media.factory.production
Production capability. Not an AI deployment.
FACTORY output / outcome
Production is not the same as outcome.
The Factory produces inference.
The operation decides what to do with it.
A model running successfully is not evidence that the operation improved.
The outcome
Realised.
What the operation does differently because it has it.
No quantified result, improvement, benchmark or performance value is asserted.
Output is produced. Outcome is realised in the operation.
FACTORY placement
Put the work where it belongs. Put the computation where it belongs.
Intelligence can be produced where the work requires it while heavier computational work is placed somewhere better suited.
That can include Edge AI Campus capacity where the requirement calls for it.
Left · produced where the work requires it
Right · engineered where that work belongs, where a requirement calls for it
The link is drawn as conditional. Not every Factory, and not all of it. No requirement that computation be local, and no requirement that a Campus exist, is asserted.
Placement follows the requirement.
FACTORY engineering relationships
Engineer before establishing.
What the Factory will do, what it may do, what it runs on and what evidence it must produce are made explicit before the capability is established.
Declare · Compile · Execute · Evidence
Above · settled once
Below · compiled, executed, and evidenced every time
Representative depiction. Not a software delivery lifecycle, not a project methodology, and no execution count, rate or duration is asserted.
FACTORY one capability
One capability to establish. One to govern. One to answer for.
Physical engineering, intelligence architecture, operating context and control are resolved as one operational capability.
The configuration can change. The accountability cannot disappear between the parts.
Power · Thermal · Communications · Compute · Operating context · Production architecture · Governance · Operational interaction
One boundary, cut as one part. The division is analytical, not physical, and none of these is separately purchasable.
FACTORY production bounds
Production needs boundaries.
Activity is not achievement.
Scope
Specify the work.
Control
Bound execution.
Outcome
Measure what changes.
Assurance
Evidence what occurred.
Four bounds of one production field. Not four features, not four benefits, and not four separately purchasable things.
FACTORY experiential entry
See a Factory work.
An Edge AI Factory makes more sense when intelligence production can be seen rather than simply described.
See how production, control, authority and evidence operate together.
FACTORY operating relationships
The Factory follows the operation.
GO · INTRA · WEAVE · AUGMENT
Four operating relationships. Not four Factory products.
The relationship can change how the Factory exists physically without changing what the customer is buying.
The work moves, and the requirement goes with it.
Go
The Factory moves with the operation. For campaigns, seasons, relocations and temporary sites, without establishing the capability again at each one.
The decision has to be made where the data is made.
Intra
The Factory sits at the point of work. For large sites and precincts where the point of work keeps moving but the operating environment does not.
The capability is meant to stay, and to be built on.
Weave
The Factory becomes part of how the operation runs. Persistence that is operational rather than physical.
The surface that touches the work is spread out.
Augment
The Factory reaches through the operation's own interfaces, where people, assets and systems meet the operation.
Four operating relationships belonging to one Factory capability. Presented without number, arrow or order between them. Not products, not tiers, not stages, and not a sequence. Selecting a view explores a relationship. It does not assign one. An estate can use more than one.
Four operating relationships, presented without number, arrow or order between them. Not products, not tiers, not stages, and not a sequence. An estate can use more than one.
FACTORY configuration
Physical configuration is an engineered consequence.
Power, thermal conditions, compute, connectivity, installation and access may change the configuration.
They do not define the Factory.
Dashed · the configuration, which changes
Solid · the capability, which does not
Representative depiction. Not three products, not three sizes, and no enclosure, footprint or physical dimension is asserted.
Configuration comes after engineering.
FACTORY end state
What the operation ends up holding.
When established, the operation holds a capability it did not have before, and keeps producing from it as the work changes.
The Factory is that capability. What it looks like physically follows from the conditions it was engineered to meet.
Left · the operation, producing what it always produced
Right · the same operation, holding a capability it did not have
Representative depiction. No enclosure, footprint, hardware or deployment form is depicted, and none of them is the commercial object.
The capability is what is established. The physical form is an engineered consequence of it.
FACTORY sovereignty
Control the production. Control the value.
Running AI nearby does not make it sovereign.
Control means retaining authority over the capability as dependencies change.
The dashed line divides by location
The control boundary crosses it without breaking
Representative depiction. Dependencies shown are illustrative. No hosting arrangement, jurisdiction, supplier or third party is depicted or implied.
Sovereignty becomes real when a dependency is tested. Not when a location is claimed.
FACTORY authority
Ability to act is not authority to act.
Human and machine actors are both subject to authority.
Permission is resolved before execution.
Evidence is produced as part of the work.
Left · equally able to act Centre · resolved before anything runs
Right · permitted or refused, recorded either way
Representative depiction. Actors and actions shown are illustrative and no control, policy or permission set is asserted.
Every action has an authority behind it, and a record after it.
FACTORY reader recognition
Should the operation produce intelligence here?
Not a scorecard.
A way to establish whether the problem is worth engineering.
Does the decision have a shelf life?
Does capability need to keep working when something else does not?
Do you need accountability for the intelligence being produced?
Can the work realistically move elsewhere?
Representative questions. Not a checklist, not scored, not a qualification instrument, and not all of them need an answer.
FACTORY category questions
Category questions
A Factory is a new kind of object. These are the questions it creates.
Most of these come from assuming a Factory is a thing rather than a capability.
01What is an Edge AI Factory?
A persistent intelligence-production capability established inside an operation.
When established, the operation holds a capability it did not have before, and keeps producing from it as the work changes.
02Is a Factory hardware, an appliance or a modular GPU system?
No. The Factory is the capability. Its physical manifestation is an engineered consequence of the conditions it was engineered to meet.
Power, thermal, compute, connectivity, installation and access may change the configuration. They do not define the Factory, and none of them is the commercial object.
03What does a Factory actually produce?
Inference, produced continuously from the work it has been established to perform.
What the operation does differently because it has it is the outcome. The two are not the same, and a model running successfully is not evidence that the operation improved.
04Does a Factory require an Edge AI Campus?
No. Intelligence can be produced where the work requires it while heavier computational work is placed somewhere better suited.
That can include Edge AI Campus capacity where a requirement calls for it. Not every Factory, and not all of it.
05Is the Demo the Factory?
No. A demonstration is an experience, and the demonstration paths are not Factory products.
A demonstration does not commit you to a Factory configuration. Where a requirement follows from it, what gets established is the capability itself.
06Are GO, INTRA, WEAVE and AUGMENT four products?
No. They are four operating relationships between one Factory capability and an operation.
They are not products, not tiers, not stages and not a sequence. The condition decides which applies, and an estate can use more than one.
07What determines the physical configuration?
The conditions the capability has to meet. A Factory is standardised where it must be, configurable where it can be, and engineered where the operation demands it.
What you can choose is what has already been engineered.
08What does sovereignty mean for a Factory?
Control, not location. Running AI nearby is a deployment decision.
Being able to control the capability that produces the intelligence, and the dependencies it rests on, is a different thing, and it decides how much of that value stays yours.
09What does authority to act mean?
Ability to act is not authority to act. Human and machine actors are both subject to authority on the same terms.
Permission is resolved before execution, and evidence is produced as part of the work rather than reconstructed after somebody asks.
10What happens when operating conditions change?
What the Factory does, what it may do and what it runs on are explicit, so altering any of them is a decision somebody takes rather than something that drifts.
Change should not mean starting again.
FACTORY commercial resolution
Bring us the operation.
Tell us where the work happens, what needs to be decided and what is preventing that today.
Or see the capability operate first.