Intel details its vision for AI chips, mostly based on Nervana technology, but the strategy remains incomplete until the Movidius deal closes
AI strategy incomplete without Movidius — SAN FRANCISCO - Intel rolled out its intentions for a soup-to-nuts offering in artificial intelligence …
Context & Ripple Effects
Two months after announcing it would buy Movidius, the startup making low-power vision chips for drones and virtual reality, Intel has laid out a soup-to-nuts artificial intelligence roadmap built mostly on Nervana technology — but the picture is deliberately unfinished until that deal closes. The gap matters because Nervana gives Intel the datacenter side while Movidius is the edge piece.
The subsequent coverage shows how the two halves were meant to meet: Facebook signed on to give technical input on the Nervana Neural Network Processor, and Movidius' line was pushed forward with the Myriad X vision processing unit, with deep-learning software later folded in when Vertex.AI joined the Movidius unit.
First-order effects
- Intel's AI pitch to customers now spans cloud training (Nervana) through device-level vision inference (Movidius), but buyers evaluating the stack must wait for deal close before the edge half is contractually Intel's.
- Movidius' roughly 180 employees and its drone/VR vision-chip business shift from independent vendor to the edge arm of Intel's AI roadmap the moment the acquisition completes.
Second-order effects
- Hyperscale customers gain leverage as co-designers rather than just buyers — Facebook's technical input on the Nervana Neural Network Processor sets a template where large cloud players shape Intel silicon before it ships.
- Edge-device makers building drones and VR hardware face a consolidated supplier: the vision chips they bought from a startup now sit inside a company that also owns their potential datacenter training path.
Third-order effects
- If the pattern holds, AI silicon competition moves from single chips to end-to-end stacks — training silicon plus edge inference plus software layers like PlaidML — favoring companies that can assemble all three through acquisition.
- The structure points toward a split market where a few vertically integrated vendors supply both cloud and edge AI compute, raising the bar for standalone AI chip startups to stay independent.
The trend: AI chipmakers are assembling end-to-end cloud-to-edge stacks through acquisition, turning point products like vision processors into components of integrated platforms.