Open-model pressure eased, but China’s full-stack buildout and heavy U.S. capex keep cycle risk elevated.

The public risk reading declined to 3.125 from 3.250 as the price-performance gap between the leading proprietary reference and qualifying open-weight alternatives widened, easing immediate pressure on proprietary-model economics. The broader cycle remains unchanged: U.S. investment and free-cash-flow pressure remain heavy, while China’s latest Alibaba and Huawei announcements reinforce both competitive pressure on model pricing and additional global demand for chips, cloud capacity and power.

• Aggregate hyperscaler capital expenditure remains up 72.29% while aggregate free cash flow is down 16.11%, leaving returns the central vulnerability.
• AWS and Google Cloud revenue growth of 24.30% still trails the 64.73% capex growth comparison, leaving a 40.43-point monetisation gap.
• Financing is broadening from corporate cash flow into bonds, project finance, securitisations and equipment-backed structures; marginal projects are more exposed to execution and customer-credit risk.
• Data-centre construction is macro-significant at 0.23% of U.S. GDP and up 57.22% year over year, raising sensitivity to power availability, build timing and utilisation outcomes in 2027-28. The top ten S&P 500 holdings represent 38.98% of SPY weight, amplifying valuation and concentration exposure.

• NVIDIA’s latest reported Data Center revenue grew 117% year over year to $89.00 billion, indicating that accelerator demand remains exceptionally strong.
• Recent infrastructure reporting still shows tight vacancy, expanding development pipelines and strong demand; the immediate issue is financing discipline and power delivery rather than an evident demand collapse.
• The broad U.S. investment-grade spread proxy is only 2 basis points wider over 90 days, not evidence of a general credit shock.
• China’s lower-cost models can expand workloads and adoption; China remains constrained by access to the most advanced chips, domestic supply limits and lower sector revenues than U.S. peers.

• Whether third-quarter results show cloud and AI revenue accelerating enough to close the capex-to-revenue gap.
• New debt, lease guarantees, project-finance terms and any cancellations or refinancing stress among neoclouds and data-centre developers.
• Evidence on power interconnection delays, memory and networking supply, and delivered rather than announced data-centre capacity.
• Whether Chinese model price-performance gains translate into international enterprise adoption, or remain limited by frontier-task quality, security and export-control constraints.