AI Capex Boom Faces Profitability Scrutiny; Margins Compressed

Alibaba and Tencent, China's twin titans of cloud infrastructure and AI, delivered a cold splash of reality to market narratives around unbounded AI capex growth. Both companies missed revenue estimates, disappointing investors who had priced in accelerating AI model and cloud service monetization. Alibaba's AI cloud business is surging, but overall growth is anemic, signaling that infrastructure investment is not yet translating into proportional revenue growth. Tencent, similarly, is investing heavily in generative AI but struggling to extract near-term margin expansion, a pattern that echoes Western hyperscalers.

The profitability squeeze is multifaceted. First, energy costs are rising due to geopolitical shocks (Iran war pushing global power prices higher), making data-center operations more expensive. Second, competition for AI model talent and compute resources is driving capex intensity higher just as demand growth slows below consensus expectations. Third, regulatory headwinds in China (restrictions on chip exports, data sovereignty rules) are forcing redundancy in infrastructure buildouts, raising cost-per-unit-of-deployed-capacity. US and European cloud providers face their own margin pressure from tariff uncertainty and potential trade friction.

NVIDIA's record high and the broader chip rally have masked a reality: the installed base of high-end GPUs is growing slower than the pace of capex outlays. This suggests either (a) demand forecasts are conservative and a ramp-up is coming, or (b) many deployed accelerators are running below utilization and firms are over-investing in optionality. If the latter, capital-efficient alternatives (custom silicon, AMD competition, quantized models) could gain traction, compressing NVIDIA's gross margins from current highs of 65%+.

The consensus AI capex narrative assumes a multi-year power-law compute growth trajectory ending in $1T+ per year by 2030. But if ROI metrics deteriorate (customer CAC/LTV ratios compress, or AI model outputs fail to justify premium pricing), capex discipline will snap back hard. Companies with fortress balance sheets (Alphabet, Microsoft, Meta) can survive a long capex-to-revenue ratio compression; younger or highly leveraged players cannot. The valuation repricing risk is asymmetric to the downside if profit margins disappoint.