From DeepSeek to Kimi K3: 18 Months That Changed AI Investment Logic as Jevons Paradox Reaffirms Compute Is King

The market story has shifted: cheaper AI training is no longer seen as cutting hardware spending.

Instead, efficiency gains are boosting total compute usage, reinforcing demand for GPUs, HBM memory, and data centers.

New model releases like Kimi K3 and continued hyperscaler capex plans support the view that AI infrastructure demand will keep rising.

The broader message is that AI efficiency may expand, not shrink, demand for chips and cloud capacity.