Risk stays elevated as financing expands and China’s cheap open models intensify rent pressure.

Recent evidence strengthens both sides of the cycle without changing any dashboard input: NVIDIA reported 117% year-on-year Data Center revenue growth, while infrastructure funding is moving further toward lease guarantees, private-capital platforms and project finance. China added evidence of open-model price-performance and domestic-chip deployment, but Alibaba’s reported cloud growth and margin expansion also show commercial demand rather than technology progress alone.

• Capex and cash-return risk remains the primary pressure: the authoritative dashboard data show +72.29% aggregate capex growth alongside -16.11% aggregate free-cash-flow growth.
• NVIDIA’s proposed $500 billion third-party compute-financing platforms and reported large lease/purchase commitments make the buildout more dependent on long-duration contractual and external-capital structures.
• The Ohio campus structure, including NVIDIA’s conditional lease and power-payment support, illustrates how power availability and counterparty commitments are becoming central to infrastructure delivery.
• China’s GLM-5.3-Flash and Qwen3.8-Flash-Next releases reinforce the open-model pressure signal; cheaper capable alternatives could reduce pricing power and returns for closed-model providers.

• NVIDIA reported $89.0 billion of quarterly Data Center revenue, up 117% year on year, and guided to $108.0 billion of total revenue for the next quarter; current accelerator demand has not broken.
• Chinese commercial evidence is not uniformly bearish: Alibaba reported 45% growth in AI Cloud and Compute Services revenue and a 12% segment adjusted-EBITA margin, indicating adoption can produce operating leverage.
• DeepSeek’s general-availability release introduced workload-sensitive reasoning and off-peak pricing, which can lower user cost and potentially expand inference volumes rather than merely displace incumbent revenue.
• IREN reported contracted AI-cloud revenue and customer prepayments alongside GPU financing, suggesting some specialist capacity is being supported by contracted demand rather than entirely speculative construction.

• Whether hyperscaler and AI-cloud contractual commitments convert into commissioned, utilised capacity and cash generation.
• Credit terms, guarantees, refinancing needs and any project delays or cancellations as external financing takes a larger role.
• Evidence on enterprise AI pricing, cloud revenue and free cash flow when the stale monetisation proxy refreshes.
• Whether Chinese open models sustain performance, availability and domestic-chip scale in production workloads, and whether this expands total inference demand or compresses supplier economics.