Sector notes
The software AI market is crowded. The hardware one is not
Every enterprise platform has an AI layer and every SaaS product has a copilot. Intelligence embedded directly into equipment and physical infrastructure is a category that has not yet been reinvented.
Executive summary
1,200+
AETOS KoolScan units built and deployed — a Singapore Medical Device Register Class B device
Core conclusions
- Differentiation in AI software is narrowing as the same capability becomes available to every competitor on the same terms.
- Embedded AI is bounded by physical engineering — enclosure, power, thermal, certification, field service — barriers to entry a software team can't buy quickly.
- Orion Five has taken sensing hardware through design, firmware, manufacture, and regulatory approval before — the hard-to-acquire part of embedded AI.
Differentiation in AI software is narrowing and margins are compressing, because the same capability is available to every competitor on the same terms. The scarce thing is not the model.
Embedded AI is different, because it is bounded by physical engineering: enclosure, power, thermal, certification, field service. Those constraints are barriers to entry rather than inconveniences, and they are where a hardware-capable practice has an advantage a software team cannot buy.
What the category looks like in practice
- Industrial equipment that diagnoses itself and orders its own replacement parts before failure.
- Instruments that carry their own inference rather than shipping data to a server that may be unreachable.
- Building and port systems where the intelligence sits at the edge because latency or connectivity will not permit anything else.
Why this is our ground
Orion Five has taken sensing hardware through design, firmware, manufacture and regulatory approval before: the AETOS KoolScan is listed on the Singapore Medical Device Register as a Class B device, registration number DE0506483, with over 1,200 units built and deployed.
That is the part of embedded AI that is hard to acquire quickly — not the model, but everything around it that has to hold in a physical product.
