数据中心转用退役喷气发动机供电

Data centers turn to retired jet engines amid power grid delays

As grid interconnection queues stretch to 5+ years and new industrial gas turbines face 36-month lead times, hyperscalers and colocation providers are retrofitting retired commercial jet engines (e.g., GE CF6) into on-site generators. While offering rapid deployment and high power density, these units operate at ~25% lower efficiency than modern turbines and emit nearly twice the CO2 per kWh—raising both operational costs ($0.18–$0.22/kWh vs. grid average $0.07–$0.10) and ESG risk. For AI application startups, this signals accelerating infrastructure cost pressure. Cloud providers are already passing through rising power expenses via higher GPU instance pricing and stricter quota limits. Teams without granular inference cost controls—such as dynamic precision scaling (FP16 → INT4), request batching, or CPU-offload fallbacks—risk 100%+ monthly cloud bill spikes at modest user growth. Embedding ‘kWh per 1,000 API calls’ as a core metric—and validating CPU+quantized inference paths during MVP—can prevent severe unit economics erosion before product-market fit is achieved.