AI demand remains firm, while capex, power and financing strain persist; China adds adoption momentum but no new cycle break.

The underlying risk picture is broadly unchanged. Demand remains supported by NVIDIA's outlook and reported Chinese demand for offshore advanced compute, while high capital intensity, weaker free-cash-flow conversion, power constraints and more cautious project-finance terms continue to test returns. No material China-driven change was verified today: recent Chinese policy actions support industrial and commercial AI adoption, but do not yet establish a new change in global model pricing, chip self-sufficiency or realised AI profits.

• Hyperscaler investment remains materially ahead of demonstrated cash conversion: the authoritative readings show rapid capex growth alongside declining aggregate free cash flow and cloud-revenue growth below capex growth.
• Power and permitting remain binding constraints. The Oracle/Blue Owl Project Jupiter delay illustrates how grid delays can defer revenue and alter contractual-risk allocation even where tenant commitment remains intact.
• AI infrastructure finance remains sensitive to funding costs and project specificity; recent reporting highlights heightened lender attention to power, permits, construction milestones and protections.
• China's low-cost model ecosystem remains a potential rent-compression channel. Faster adoption could increase inference demand but reduce the pricing power assumed for proprietary model providers.

• NVIDIA's latest reported data-center revenue grew 117% year over year to $89 billion, and its forward outlook pointed to continued broad demand across hyperscalers, AI labs, enterprises and sovereign customers.
• NVIDIA reported supply constraints, including memory, rather than a broad demand retrenchment; this supports the view that current infrastructure utilisation and procurement demand remain robust.
• Contracted-capacity structures still exist: AIB Data Centers disclosed a 12-year, 50 MW Nebius agreement with expected customer prepayments funding a meaningful part of initial development.
• China's reported offshore-chip lease and official adoption policies point to incremental compute use and industrial deployment, which can expand the global inference market rather than only displace incumbent revenues.

• Alphabet, Microsoft and Amazon results for evidence that cloud and AI revenue, margins and operating cash flow are catching up with infrastructure spending.
• Whether Project Jupiter's power delay changes tenant payments, lender terms, construction milestones or wider data-center financing conditions.
• Whether grid interconnection, permitting and construction capacity defer contracted capacity in constrained US markets.
• Whether reported Chinese demand for export-compliant or offshore compute converts to delivered capacity, and whether Chinese model price-performance changes commercial terms outside China.