AI spending and demand remain strong, while China’s advancing domestic stack adds competitive and monetisation pressure.

The underlying risk picture is broadly unchanged. Recent evidence still shows rapid data-centre spending and additional capacity financing, while power availability, equipment costs and long payback periods remain constraints. China added evidence of faster domestic AI-stack development and enterprise deployment, which could accelerate lower-cost adoption and intensify competition, but no material China-driven change to global AI investment demand was verified today.

• Infrastructure spending remains exceptionally high relative to disclosed AI-specific returns; hyperscalers still do not separately report the profitability of AI infrastructure.
• AI financing is broadening into equipment facilities, project structures and credit support for smaller infrastructure operators, increasing execution and refinancing sensitivity beyond the largest platforms.
• Power, permitting, memory and supply-chain constraints can raise project costs or delay utilisation, weakening returns on committed capacity.
• China’s domestic chips, cloud services and open-model ecosystem could make capable AI cheaper, shifting value from frontier-model providers toward users, infrastructure and application layers.

• Data-centre capex growth remains supported by accelerator, storage and networking demand rather than a verified broad cancellation cycle.
• DigitalOcean’s new equipment facility and lender participation indicate continued availability of capacity financing for a smaller cloud provider.
• Chinese AI progress can be demand-positive: lower-cost models and domestic infrastructure may expand enterprise use, token consumption and global compute needs.
• China still faces export-control constraints on the most advanced chips, and US models retain a lead in raw computing capability according to recent reporting.

• Whether cloud revenue, margins and disclosed AI workloads improve sufficiently to support the installed-capacity buildout.
• Debt, lease and project-finance terms for neoclouds and data-centre operators, especially any cancellations, covenant pressure or refinancing difficulty.
• Evidence that grid connection delays, power pricing or memory inflation postpone data-centre delivery or impair economics.
• Whether Huawei’s claimed Ascend and cloud deployment progress converts into independently evidenced performance, volume adoption and commercially competitive pricing.