Ambarella Pushes More AI Processing Onto Camera Silicon at Embedded World
Chipmakers demonstrated edge AI hardware moving inference directly onto camera chips, with Ambarella extending from vision into the broader industrial edge market.
At Embedded World 2026, chipmakers including Ambarella demonstrated the next generation of edge AI hardware, pushing more inference directly onto camera silicon and extending into the wider industrial edge computing market.
Moving inference onto the camera chip changes the economics of a deployment more than it changes its capability. A camera that streams continuously to a server consumes bandwidth whether or not anything is happening, and in most installations almost nothing is happening almost all of the time. Detection on the sensor means only events leave the device, which on a metered or wireless link is frequently the difference between a viable rollout and one whose connectivity cost exceeds its value.
It also narrows the privacy surface, since raw video never crosses the network boundary. That is easier to explain to a customer or a regulator than a data processing agreement covering footage held centrally, and it removes an entire category of obligation rather than managing it.
Source: Counterpoint Research