Ceva Unveils NeuPro NPU Portfolio for AI Infrastructure

Ceva, a leading semiconductor intellectual property licensor, officially launched its next-generation NeuPro NPU portfolio on October 28, 2026, marking a significant expansion into AI infrastructure markets. The new neural processing unit architecture targets edge AI deployment and datacenter acceleration, positioning Ceva to compete in the rapidly growing AI chip sector alongside established players.
Next-Generation NPU Architecture
The NeuPro NPU portfolio introduces advanced silicon IP designed for efficient AI inference and training across multiple deployment scenarios. The architecture leverages optimized tensor processing units and memory hierarchies to deliver performance improvements for deep learning workloads. Ceva's design approach emphasizes power efficiency, critical for edge devices operating under thermal and battery constraints.
According to the company, the new NPU lineup supports a wide range of AI frameworks and model formats, enabling semiconductor manufacturers to integrate the IP into system-on-chip designs for mobile, automotive, IoT, and datacenter applications. The flexibility of the portfolio allows customers to select configurations matching their specific performance and power targets.
AI Infrastructure Expansion
Ceva's move into AI infrastructure represents a strategic pivot beyond its traditional edge computing focus. The company now offers NPU designs optimized for server-class AI acceleration, addressing demand from hyperscale cloud providers and enterprise datacenter operators. This expansion puts Ceva in direct competition with dedicated AI chip startups and established semiconductor vendors targeting the datacenter AI market.
The infrastructure-focused NPU variants prioritize throughput and multi-tenancy capabilities, essential for cloud AI services handling concurrent workloads. Ceva reports that early design wins include partnerships with unnamed chip manufacturers developing next-generation AI servers scheduled for deployment in 2027.
Release Date and Availability
Officially launched on October 28, 2026, the NeuPro NPU portfolio is immediately available for licensing to semiconductor companies and system integrators. Ceva provides comprehensive development tools, software stacks, and reference designs to accelerate customer time-to-market. The company expects first silicon implementations incorporating the new NPU IP to emerge in late 2027, with commercial products reaching market in early 2028.
Licensing terms follow Ceva's established royalty model, with upfront fees and ongoing per-unit royalties tied to chip shipment volumes. The company offers tiered portfolio access, allowing customers to license specific NPU configurations or the complete IP family.
Market Positioning
The NeuPro launch arrives amid intensifying competition in the AI chip sector, with dozens of startups and established vendors vying for market share. Ceva's IP licensing model differentiates it from vertically integrated chip companies, enabling broader ecosystem adoption through partnerships rather than direct silicon sales.
Industry analysts note that Ceva's edge AI heritage provides credibility in power-efficient design, a critical advantage as AI workloads proliferate across battery-powered devices. The infrastructure expansion, however, requires proving performance against specialized datacenter AI accelerators optimized for high-throughput training and inference.
What This Means
Ceva's NeuPro NPU portfolio launch and AI infrastructure expansion signal growing maturity in the semiconductor IP market for AI acceleration. As AI deployment scales across edge and cloud environments, demand for flexible, licensable NPU architectures will likely accelerate. For chip manufacturers, Ceva's offering provides an alternative to in-house AI IP development, potentially shortening product cycles. The success of this portfolio will depend on demonstrated performance advantages, ecosystem support, and ability to capture design wins in competitive datacenter and edge markets where specialized AI chips already dominate.
on Emergent today





