Big Tech Is Spending More Than Ever on AI and It's Still Not Enough (www.wsj.com)

Big tech companies are pouring unprecedented sums into AI — buying vast GPU fleets, building hyperscale data centers, and designing custom accelerators — but even this surge in spending isn’t keeping pace with demand. The result is a frenzied market for AI hardware, tight supply of high-end GPUs and AI chips, prioritized capacity for large incumbents, and significant capital outlays for power, cooling and interconnects. Companies are redirecting talent and budget toward foundational-model training and inference infrastructure, squeezing other initiatives and compressing margins despite the promise of new AI-driven revenue streams. For the AI/ML community this means two practical shifts: first, compute and energy remain the primary bottlenecks driving who can train the biggest models — favoring hyperscalers and well-funded labs — and second, technical emphasis is moving toward hardware/software co-design, efficiency research, and alternative architectures (custom ASICs, model compression, sparsity, quantization, and smarter parallelism). Geopolitics and supply-chain constraints complicate access to chips, while the economics of ever-larger models push organizations to balance in-house build vs. cloud services. The bottom line: money accelerates capability but doesn’t eliminate fundamental resource limits, so algorithmic efficiency and novel system designs will be as strategic as raw capital.