AI demand remains exceptionally strong, but financing and capex risks persist as China adds both competition and monetisation.

NVIDIA’s August 26 results provided the clearest new counterweight to the risk case, with Data Center revenue up 117% year on year and a $108 billion quarterly-revenue outlook that excludes China compute revenue. Offsetting this, disclosed NVIDIA lease obligations, reported Broadcom-linked debt discussions and Alibaba’s funding/capex expansion show that the buildout is increasingly dependent on large external financing commitments. China-specific evidence verified today shows accelerating cloud commercialisation and domestic inference capability, but no single China development materially changes the dashboard’s rules-based signals.

• Capex and cash-return pressure remains the primary risk: the authoritative dashboard data show 72.29% aggregate hyperscaler capex growth alongside 16.11% aggregate free-cash-flow contraction.
• Financing structures are becoming larger and more contingent: NVIDIA disclosed obligations connected to a 4.25-GW Ohio AI campus, while Broadcom was reported to be discussing a potentially very large debt-backed AI-chip financing structure.
• Alibaba reported 75% quarterly capex growth and said it had already deployed roughly half of its planned 2026-29 AI investment; its stated payback horizon is three years, underscoring that returns remain prospective.
• Open-model pressure remains high under the authoritative input. Chinese open-model and domestic-cloud investment can widen access and adoption but may further reduce scarcity rents for frontier proprietary models and hosted inference.

• NVIDIA reported $89.0 billion of Data Center revenue, up 117% year on year and 18% sequentially, with Q3 revenue guided to $108.0 billion plus or minus 2%; this is direct evidence that infrastructure demand has not broken.
• NVIDIA’s reported demand is broader than a single lab, including hyperscale and AI-cloud, industrial and enterprise channels; that breadth is a counterpoint to an immediate demand-collapse thesis.
• Alibaba reported 45% AI Cloud and Compute Services growth, triple-digit AI-related product growth and improved cloud profitability; Baidu reported AI Cloud Infrastructure growth of 50% and GPU Cloud growth of 283%. These results test the proposition that AI demand cannot be monetised.
• Chinese efficiency gains may lower unit costs and accelerate application adoption. That can expand total token and compute demand even if it pressures model pricing.

• Whether newly disclosed guarantees, SPVs and debt raises close on the proposed terms, and whether project sponsors retain risk rather than distribute it.
• Hyperscaler cash flow, cloud growth and disclosed AI revenue in the next reporting cycle, particularly evidence that revenue conversion catches capex.
• NVIDIA customer concentration, receivables, supply commitments and China revenue disclosures following its exclusion of China Data Center compute revenue from Q3 guidance.
• Evidence on data-centre power delivery, construction timing and cancellations; 4.25-GW-plus projects increase execution and grid dependency risks materially even when chip demand is strong.([sec.gov](https://www.sec.gov/Archives/edgar/data/1045810/000104581026000069/nvda-20260817.htm))