Intel unveils Movidius Myriad X low-power vision processing unit (VPU) for AI acceleration in vision-based devices
Today, Intel subsidiary Movidius is announcing the Movidius Myriad X vision processing unit (VPU), a low-power system-on-chip (SoC) intended for deep learning and AI acceleration …
Context & Ripple Effects
This launch closes a loop Intel opened a year earlier: its stated AI chip strategy, built mostly around Nervana technology, was explicitly incomplete until the Movidius acquisition closed — and Myriad X is the first major silicon to come out of that subsidiary under Intel ownership. It lands just weeks after Movidius shipped the $79 USB Compute Stick built on the older 12-core Myriad 2, so Intel now has both a developer entry point and a next-generation chip.
The Myriad X then becomes the foundation of Intel's edge-AI product line rather than a one-off: it powers the $100 Neural Compute Stick 2 released the following year, and the low-power VPU line continues with the Keem Bay successor announced in 2019 — evidence Intel intends to compete at the edge, not only in data center accelerators.
First-order effects
- Device makers building cameras, drones, and robotics get a dedicated low-power SoC for on-device deep learning inference, removing the need to stream video to a host or cloud for AI processing.
- Developers already working with the Myriad 2-based Compute Stick gain an upgrade path within the same Movidius toolchain, keeping Intel's early edge-AI developer base inside its ecosystem.
Second-order effects
- The $79 Compute Stick and later the Myriad X-based Neural Compute Stick 2 turn edge inference into a shippable consumer-priced category, pressuring rival accelerator vendors to offer their own low-cost, USB-format development hardware.
- Cloud AI providers face incremental substitution at the margin: workloads that run locally on a VPU are workloads not billed as cloud inference, nudging Intel's strategy toward selling silicon into devices rather than services.
Third-order effects
- If the line holds — Myriad X in 2017, Keem Bay in 2020, and a Myriad 2 flying on the PhiSat-1 satellite doing onboard cloud-cover analysis — dedicated low-power VPUs become standard infrastructure wherever a camera exists, pushing AI compute toward the sensor itself.
- Intel ends up running two distinct AI silicon strategies in parallel: Nervana-oriented parts for the data center and Movidius parts for the edge, a split that forces clarity about whether one architecture can serve both or whether they remain separate businesses.
The trend: AI acceleration is migrating out of the data center and into low-power chips embedded beside cameras and sensors, with Intel's Movidius line as one of the earliest sustained attempts to own that edge tier.