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 roughly 180-person maker of vision chips for drones and VR, Intel is now sketching the full picture: a soup-to-nuts AI offering built largely on Nervana technology, with the strategy explicitly incomplete until the Movidius deal closes. The pairing is deliberate — Nervana covers the datacenter side while Movidius brings low-power vision processing for devices.
The subsequent record shows the blueprint filling in: Intel shipped the Myriad X low-power vision processing unit under the Movidius brand in 2017, disclosed that Facebook is providing technical input on the Nervana Neural Network Processor, and later folded deep-learning startup Vertex.AI into the Movidius unit.
First-order effects
- Once the Movidius deal closes, Intel can sell one AI portfolio spanning datacenter training on Nervana silicon down to low-power vision inference in drones and VR headsets, instead of two disconnected product lines.
- Customers evaluating AI hardware get a single vendor claiming every tier of the stack, raising the bar for point-product rivals in both server accelerators and device-side vision chips.
Second-order effects
- Facebook's technical input on the Nervana Neural Network Processor shows hyperscalers co-designing silicon with their suppliers — a template other cloud operators will feel pressure to replicate to keep their workloads' requirements baked into future chips.
- Rivals in edge vision silicon face a competitor bundling Movidius VPUs with Intel's broader compute and software stack, shifting competition from individual chips toward integrated offerings.
Third-order effects
- Intel's follow-on moves — shipping Myriad X, then acquiring Vertex.AI into the Movidius unit while open-sourcing its PlaidML engine — suggest the company intends to build out edge AI by serial acquisition rather than purely organic development, consolidating a fragmented startup field under one roof.
- If the Nervana-plus-Movidius structure holds, AI hardware markets split along a datacenter-versus-edge axis where vendors compete on owning the whole path from cloud training to device inference, making full-stack breadth a prerequisite rather than a differentiator.
The trend: Chipmakers are assembling end-to-end AI portfolios that span datacenter training and edge inference through acquisition, with Intel's Nervana-and-Movidius combination an early instance of the pattern.