Executive summary
1,200+
AETOS KoolScan units built and deployed, a Singapore Medical Device Register Class B device
Key conclusions
- AI software products are becoming harder to tell apart, as every competitor gains access to the same capability on the same terms.
- Embedded AI is constrained by physical engineering: enclosure, power, thermal design, certification and field service. A software team cannot acquire these skills quickly.
- Orion Five has already taken sensing hardware through design, firmware, manufacture and regulatory approval. This is the part of embedded AI that is hard to acquire.
AI software products are becoming harder to tell apart and margins are falling, because every competitor can access the same capability on the same terms. The model is no longer scarce.
Embedded AI is different because physical engineering constrains it: enclosure, power, thermal design, certification and field service. These constraints are barriers to entry. A firm with hardware skills has an advantage here that a software team cannot buy.
Three examples of embedded AI in practice
- Industrial equipment that diagnoses itself and orders its own replacement parts before failure.
- Instruments that run their own inference, with no dependency on a server that may be unreachable.
- Building and port systems that place the intelligence at the edge, because latency or connectivity rules out any other design.
We have already taken a sensing device through regulatory approval
Orion Five has taken sensing hardware through design, firmware, manufacture and regulatory approval. The AETOS KoolScan is listed on the Singapore Medical Device Register as a Class B device, registration number DE0506483. Over 1,200 units have been built and deployed.
In a physical product, everything around the model must also work reliably. This is the part of embedded AI that is hard to acquire quickly.
