AI's Driving Force: Computing Power Unleashed, Suleyman Declares 'Just the Beginning'

Microsoft AI division CEO Mustafa Suleyman has declared that the era of explosive growth in computing power, the bedrock of artificial intelligence (AI), is only just beginning, predicting that AI development will continue its relentless advance. Since Suleyman commenced his AI research in 2010, the computational power dedicated to training cutting-edge AI models has surged by a trillionfold. Initial systems operated at roughly 10¹⁴ FLOPS (floating-point operations), whereas today's largest-scale models surpass 10²⁶ FLOPS.

This increase in computational capacity is underpinned by concrete hardware advancements. NVIDIA chips have seen an eightfold performance improvement over six years, scaling from 312 teraFLOPS in 2020 to the current 2,500 teraFLOPS. The Maia 200 chip has delivered a 30% improvement in performance per dollar compared to its launch. Technologies like NVLink and InfiniBand connect hundreds of thousands of GPUs to function as a single colossal supercomputer.

Separate from hardware evolution, improvements in software efficiency are also driving down costs. According to an analysis by AI research organization Epoch AI, the computation required to reach the same performance level is halving approximately every eight months. This pace significantly outstrips Moore's Law, which predicts a doubling of performance every 18 to 24 months.