Imec CEO Patrick Vandenameele on scaling and Europe in the age of AI. AI is taking the world by storm, and the semiconductor industry can’t keep up. Memory prices have skyrocketed because of unprecedented demand, fabless chip designers push for more foundry capacity and lead times on semiconductor manufacturing equipment have doubled. While partly a capacity issue, technology will have to do most of the heavy lifting: The race to deliver more computational power at lower cost and lower energy consumption has intensified dramatically. Bits&Chips’ Dutch state of semicon 2026 This article is part of the Bits&Chips special edition on the Dutch state of semicon, published on 23 October 2026. Order your hard copy of the special at info@techwatch.nl. “AI has made it unmistakably clear that compute matters more than ever,” confirms Imec CEO Patrick Vandenameele, who took the helm from Luc Van den hove last April. Although CMOS scaling is and always will be an Imec core business, Vandenameele admits to being caught off guard by the introduction of ChatGPT in 2022. “In those days, I was having preliminary discussions about succeeding Luc. The vision I put forward for Imec emphasized building applications on top of the technologies our organization develops. But after ChatGPT, our emphasis flipped back to compute within months.” The difference between today and the pre-AI era, when smartphones were still the primary driving force for scaling semiconductor technology, is that the appetite for computational power used to have a ceiling. These days, it’s insatiable. Squeezing, say, 30 percent better power performance going from one node to the next would satisfy the needs of smartphone manufacturers. Today, hyperscalers take everything they can get and still want more – much more. Credit: Imec That difference is palpable at Imec, Vandenameele says. Traditionally, the Leuven-based institute derives most revenue from precompetitive research programs, which it runs together with dozens of partners at once. “These days, we’re spending a growing share of our time on one-on-one collaborations, looking to push specific pieces of shared research to market faster.” And so, as AI raises the stakes, it raises the pace of scaling, Vandenameele affirms. “The industry is once again willing to entertain higher-risk concepts than it used to.” Orchestrator The massive uptick in CMOS development coincides with the definitive transformation of scaling, which is evolving from shrinking two-dimensional transistors to a – literally – multidimensional effort that includes 3D integration, photonics, new memory types that can be integrated directly with logic and even the algorithms themselves. “Each new generation of algorithms can reshape hardware requirements once it’s well understood. For example, the CPU-to-GPU ratio in data centers being built now has shifted compared to even just a quarter ago, reflecting emerging AI agent technology over the training of large language models.” This eclectic approach to scaling, in which every aspect of the system’s performance is tightly intertwined and in need of co-optimization, has had a profound impact on Imec. Historically, its core clientele consisted of foundries, IDMs, tool makers and materials suppliers. Today, it also works with chip designers, system companies and hyperscalers. Tellingly, attendance at Imec’s Partner Technical Week, a biannual closed-door conference to review research results, today exceeds a thousand people, and includes computer architects, software developers and AI researchers. Their input feeds back into decisions about future process technology just as much as the ideas of electrical engineers and materials scientists. Credit: Imec “The complexity has become enormous. No single company can solve these challenges alone,” Vandenameele says. That goes for Imec, too, he adds: “Our budget isn’t infinite. We have to be deliberate about where we can add the most value.” Increasingly, Imec position