IMEC Says Today’s AI Will Look Ancient in 10 Years

Semiconductor innovation is shifting away from traditional Moore’s Law scaling as transistor sizes approach the physical limit of one nanometer. Current AI infrastructure faces critical bottlenecks in power, memory, cooling, and data transfer, prompting a move toward more specialized hardware architectures. Photonics emerges as a transformative solution, offering superior energy efficiency and speed compared to copper by enabling faster data movement and reducing thermal constraints. As the industry moves beyond the "scaling hypothesis"—the reliance on increasingly massive models—future development will prioritize hardware-software co-design to achieve greater efficiency. Stephen from IMEC and investor Adam Chambers highlight that while current AI relies on brute-force computation, the next decade will likely see a shift toward more sophisticated, efficient systems that mirror the complex, high-speed data processing capabilities of the human brain.