After AI explosion ... what’s next?

We are at the inflection point where compute power is no longer the most important indicator for AI systems. Over the last decade, AI has been going through breakthroughs, year after year, with a focus on scaling from training to test-time compute. On the software side, models compute demand and sophistication has been growing with a factor 10 every year. On the hardware side, silicon designs have been required to grow from tens of billions of transistors to hundreds of billions. Similarly, power consumption has grown with the same ratio.

At this inflection point, we need to ask ourselves whether this path is sustainable, from an economical and societal perspective. In this keynote, we’ll present a vision for the future of AI semiconductors, rooted in concepts such as trusted technology, sustainability and scalability. We’ll talk how a systems perspective is needed to build sustainable AI and AI for citizens is. The current state of the art will be dissected. This will be followed by a critical, but constructive, analysis of what fundamental areas we are missing from technology and usage model perspective. Finally, a proposal on what are the main vectors that should be driven during the next decade in order to make AI and technology in general citizen centric.

We are at an inflection point where raw compute power is no longer the sole driver of AI progress. Over the last decade, breakthroughs have come year after year, driven by scaling compute—from training over post-training to test-time compute. On the software side, the demand for compute and model complexity has grown tenfold annually. On the hardware side, silicon designs have been required to follow from tens of billions of transistors to hundreds of billions. Similarly, power consumption has grown with the same ratio.

But now, we must ask: is this trajectory sustainable—economically, environmentally, and societally?

In this keynote, we’ll explore a new vision for AI semiconductors, grounded in trust, sustainability, and scalability. We’ll argue that building AI for citizens —not just for performance—requires a shift to a systems-level mindset. We’ll challenge today’s status quo, uncover the gaps in both technology and deployment models, and propose a set of bold yet actionable directions for the next decade. Because if we want AI to serve humanity, we must start designing technology with humanity at its core.