Imagination Technologies, under its seventh CEO in a decade, is doubling down on GPU IP. The new strategy: kill the RISC-V CPU launched in 2021 and shelve the pre-transformer NPU from 2024. The British IP firm took out a $100M loan two years ago to fund a next-generation F-series GPU architecture that will scale from automotive to data center and from graphics to AI inference—the first major architectural refresh in years.
The pivot reflects brutal market realities. The RISC-V CPU required years of R&D before payoff and faced consolidation headwinds; NPUs designed for pre-transformer workloads are obsolete post-ChatGPT. By focusing all engineering on graphics and AI on a single unified GPU architecture, Imagination avoids programming-model fragmentation and can serve customers who want cloud gaming (GPU-heavy) by day and AI inference by night on the same silicon. Two major customers have already signed for F-series: one Chinese OEM building AI PCs, one global consumer-electronics giant.
For architects licensing GPU IP, Imagination's bet is significant. The F-series will scale beyond E-series 300mm2 in TSMC N5 but won't hit 1 PFLOPS; the generation after could. Most customers won't sell chips—they'll verticalize: building custom PCIe cards, servers, or full data center stacks. This positions Imagination as an alternative to NVIDIA's licensing model, betting that custom-silicon builders want a stable graphics-first foundation with AI as a co-equal tenant, not a pivot.