Qualcomm Technologies has awarded five European PhD students a total of $200,000 in research funding to advance cutting-edge work in artificial intelligence reasoning, hardware verification, and generative systems. The 17th edition of the Qualcomm Innovation Fellowship (QIF) Europe marks a turning point: submissions climbed to an all-time high this year, nearly 50 percent above last year's total, reflecting how rapidly machine learning and AI are evolving.

The five fellows selected from a shortlist of 18 finalists represent some of Europe's top research institutions. Each recipient receives $40,000 in funding plus one-on-one mentorship from Qualcomm's research team. The winners are Mar González I Catalá from the University of Cambridge, Jiajun He also from Cambridge, Abhinandan Pal from the University of Birmingham, Naila Sebastián Esandi from INRIA and ENSAE Paris, and Christopher Wewer from the Max Planck Institute for Informatics.

"Submissions climbed to an all-time high this year, close to 50 percent above last year's total, which says a great deal about how fast machine learning and AI is moving, and how vital it is to keep our rapidly evolving hardware and software secure," said Michael Hofmann, Senior Director of Engineering at Qualcomm Technologies Netherlands B.V.

Michael Hofmann, Senior Director of Engineering at Qualcomm Technologies Netherlands B.V.

What Makes This Year's Research Proposals Stand Out?

The range of research topics selected this year reveals where the AI field is focusing its energy. Rather than concentrating on a single narrow problem, the 2026 fellows are tackling interconnected challenges that span from reasoning transparency to hardware efficiency to scientific discovery.

González I Catalá's work, titled "A Geometric Theory of Autoregressive Reasoning," addresses a fundamental puzzle: why do AI reasoning models sometimes get stuck or produce incorrect answers? His research treats chain-of-thought traces, the step-by-step reasoning that models show their work, as trajectories through a geometric space. By understanding the underlying structure of these reasoning paths, he aims to develop new ways to intervene at inference time, redirecting models away from incorrect conclusions before they finalize an answer.

He's not alone in focusing on test-time intelligence. Jiajun He's proposal tackles diffusion models, which are generative systems used across images, molecules, and proteins. His work develops a replica-exchange framework that allows these models to handle multiple constraints that evolve during inference, a critical capability for real-world applications where requirements shift or harden as the system runs.

The fellowship also recognizes the growing importance of AI-assisted hardware design. Abhinandan Pal's Neural Model Checking project aims to automate hardware verification, a task that currently requires extensive manual proof effort. By training small neural networks as candidate correctness proofs and then checking them against every possible behavior of a circuit, his approach could dramatically accelerate the verification process as AI tools begin generating hardware code more rapidly.

How Are These Research Areas Reshaping AI Development?

The fellowship selections reflect a broader industry shift toward solving practical bottlenecks in AI deployment. Rather than chasing raw model size, researchers are increasingly focused on efficiency, interpretability, and real-world constraints.

• Reasoning Transparency: González I Catalá's geometric framework could help developers understand when and why models fail, enabling better intervention strategies at inference time rather than relying solely on training-time fixes.
• Constraint Handling: He's diffusion model control work addresses the gap between lab conditions and production environments, where systems must adapt to multiple evolving requirements without sacrificing accuracy or diversity.
• Hardware Acceleration: Pal's Neural Model Checking could unlock faster hardware design cycles, allowing chip manufacturers to keep pace with the rapid evolution of AI accelerators and edge devices.
• Zero-Shot Reinforcement Learning: Sebastián Esandi's theoretical framework aims to reduce computational redundancy in robotics and mobile applications, critical for autonomous vehicles and surgical robots that cannot afford to solve planning problems independently for each task.
• World Model Efficiency: Wewer's scene state tokenization work tackles a fundamental limitation of current video-based world models, which waste compute on rendering complex scenes and cannot adapt inference cost to scene complexity.

Hofmann emphasized the breadth of the research landscape: "What stood out was the sheer range of the proposals, stretching from multimodal generation, trustworthy agents, world models, robotics, and generative AI safety to hardware verification, secure systems, privacy, communications, edge AI, and machine learning for scientific discovery".

Hofmann

Why Does a 50 Percent Surge in Applications Matter?

The dramatic increase in fellowship submissions signals that European research institutions are treating AI as a priority area. The jump from last year's baseline to nearly 50 percent higher this year suggests that PhD programs across the continent are ramping up their AI research capacity, and that talented early-career researchers see the field as offering both intellectual challenge and career opportunity.

The finalists came from 13 leading European institutions, including ETH Zurich, INRIA, KU Leuven, the Max Planck Institutes, Technical University of Munich, and the Universities of Amsterdam, Birmingham, Cambridge, Oxford, and Tübingen. This geographic and institutional diversity indicates that AI research excellence is not concentrated in a single hub but distributed across Europe's top technical universities.

For Qualcomm, the fellowship serves a dual purpose: it identifies emerging talent early and builds relationships with researchers whose work aligns with the company's hardware and software roadmap. By providing mentorship alongside funding, Qualcomm gains insight into where the field is heading while supporting the next generation of AI researchers who may eventually shape the industry.

The 2026 cohort's focus on inference-time efficiency, reasoning transparency, and hardware-software co-design suggests that the AI field is maturing beyond the era of simply scaling up model size. Instead, researchers are tackling the harder problems of making AI systems more interpretable, more efficient, and more adaptable to real-world constraints. For developers and companies building AI systems, these advances could translate into models that are cheaper to run, easier to debug, and more reliable in production environments.