Ceva, Inc. filed a press release on October 28, 2026, disclosing the launch of its next-generation NeuPro NPU portfolio, which the company says delivers up to 2X higher AI throughput per mm² and 50% better energy efficiency versus the first generation.

Key Highlights

  • Performance claims: The next-generation NeuPro portfolio is described as delivering up to 2X higher AI throughput per mm² and a 50% improvement in energy efficiency compared to the first generation.
  • Four-product lineup: The portfolio comprises NeuPro Nano, NeuPro Prime, NeuPro Max, and NeuPro Ultra, each targeting different inference workload segments.
  • AI infrastructure entry: NeuPro Ultra extends the portfolio into cloud and data center inference, as well as intelligent memory, connectivity, and data-processing silicon.
  • Unified software stack: All four NPUs share a common core architecture and a single unified SDK, according to the filing.
  • Availability timeline: The next-generation architecture is available now to select early-access customers, with general availability planned for the second quarter of 2027.
  • Customer base cited: Ceva stated it worked closely with more than 15 customers who licensed the first generation of NeuPro in developing the new portfolio.
  • Market forecast referenced: The filing cites ABI Research forecasting cloud AI inference capacity to grow at a 42% CAGR through 2035.

Ceva's Next-Generation NeuPro Portfolio: Architecture, Products, and Market Reach

According to the filing, Ceva, Inc. (NASDAQ:CEVA) introduced its next-generation NeuPro NPU portfolio on October 28, 2026, describing it as built on a common architecture and unified software development kit spanning applications from always-on sensors to AI infrastructure. The company stated that advanced AI numerics and mixed precision, including FP4 and FP8, are supported, alongside dynamic shapes, paged KV cache, and support for network architectures such as Mixture-of-Experts and linear attention.

The portfolio is organized into four distinct products. NeuPro Nano is described as targeting always-on sensing and entry-level perception in highly constrained devices. NeuPro Prime addresses mainstream CNN and transformer workloads. NeuPro Max is positioned for advanced edge AI, including high-resolution perception, multimodal models, on-device assistants, robotics, and agentic AI. NeuPro Ultra, the company disclosed, targets AI infrastructure, supporting large language model inference in cloud and data centers as well as integration into memory and networking subsystems.

Ceva CEO Amir Panush was quoted in the filing as stating: "A common NPU architecture and software stack across every class of AI workload lets our customers build once and deploy across their roadmap." The filing noted that leading semiconductor and system companies are already evaluating and integrating next-generation NeuPro technology, with early engagements spanning multiple market segments. Ceva also announced technology symposium events in Hsinchu, Taiwan, on November 2 and in Shanghai, China, on November 5, 2026, at which the company said it would discuss the portfolio and its Physical AI strategy in more detail.

What Was Disclosed

Ceva disclosed the launch of its next-generation NeuPro NPU portfolio, including four new products, claimed performance improvements over the first generation, a unified SDK, and a general availability timeline of the second quarter of 2027, with early-access availability beginning immediately.