Anaqua report: AI semiconductor patents grew 114% in five years

Anaqua, the IP management software company, has published its 2026 Semiconductor Industry Patent Report, analysing 713,755 semiconductor-related patent filings over five years using its AcclaimIP analytics platform. The headline finding: patents at the intersection of AI and semiconductors grew 114% over the period, compared with 78% growth across the broader semiconductor universe, a gap the report characterises as evidence of structural rather than cyclical change.

The data, drawn from queries run between May and June 2026 via AcclaimIP's Model Context Protocol connector, covers chip architecture, GPUs, inference processors and custom accelerators. Toni Njim, chief product officer at Anaqua, said the scale of AI's influence on semiconductor R&D was "astounding," adding that AI had moved from merely influencing chip innovation to "defining it."

Who leads where

The report surfaces some counterintuitive competitive positions. Samsung tops AI architecture filings at 1,194, leads high-bandwidth memory (HBM) patenting with 107 filings, and ranks second in both inference and analogue AI, spanning neural processing units, processing-in-memory technology and its Exynos system-on-chip line. The company is described in the report as the clearest example of a memory manufacturer converting its incumbent base into end-to-end AI-chip strength.

IBM leads AI-semiconductor crossover filings overall with 794 and tops analogue AI with 389. Its portfolio concentrates on frontier areas including neuromorphic chips and phase-change memory, the latter of which is reaching commercial deployment via its Spyre accelerator and reportedly delivers roughly 14 times better energy efficiency than conventional approaches, though that figure is drawn from IBM's own filings and has not been independently verified.

NVIDIA's position is more nuanced. Despite its commanding market share in GPU compute, it holds only 49 GPU-titled semiconductor patents, ranking seventh in that subcategory and fifteenth in inference chips. The report attributes this to a competitive moat built on the CUDA software ecosystem rather than hardware patents, a structural distinction that matters for anyone modelling the durability of NVIDIA's lead.

Qualcomm shows the steepest growth trajectory in the dataset, up 186% in AI architecture filings to 252, and now ranks ninth in inference with 655 filings, a category in which it recorded no granted patents five years ago. Its AI200 and AI250 data-centre inference accelerators signal a deliberate pivot away from mobile and edge into hyperscaler workloads, competing primarily on power efficiency and memory capacity.

Geopolitical and competitive read-across

Perhaps the most strategically significant finding concerns Chinese government-affiliated entities, which lead inference and accelerator filings with 5,387 over the five-year window, well ahead of Samsung's 1,897. The report frames this as a chip-sovereignty response to US export controls administered by the Bureau of Industry and Security. Startups including Moore Threads and Muxi are cited as emerging GPU developers within that ecosystem, though neither has yet disclosed commercially validated benchmark performance.

The data also documents hyperscalers' shift into silicon design. Alphabet and Microsoft appear across nearly every AI-semiconductor filing category, reflecting the trend toward custom accelerators (Google's TPU, Microsoft's Maia) designed for proprietary workloads rather than general-purpose off-the-shelf procurement. This vertical integration reshapes the traditional semiconductor supply chain and narrows the addressable market for merchant silicon vendors in the highest-volume AI training and inference segments.

The broader patent landscape reflects an industry reorganising around AI compute at every layer of the stack. For enterprise buyers, the practical implication is that competitive positioning in AI infrastructure will increasingly be shaped by software ecosystems, memory architecture and custom silicon rather than general-purpose GPU availability alone. Anaqua's report is available for download from the company's website.