AI risk remains elevated: financing and power frictions persist, while China’s chip-software push broadens competition.

The underlying risk picture is broadly unchanged. Strong infrastructure demand remains visible in NVIDIA, Oracle and HPE results, while Project Jupiter’s power-related contractual dispute and selective credit markets continue to test whether large AI capacity can be financed and delivered on schedule. China added a material competitive development: DeepSeek and Huawei released Ascend-focused programming infrastructure, which may lower switching costs to domestic chips but has not yet demonstrated broad adoption or changed global AI revenue capture.

• Project Jupiter shows that power availability can become a contractual and financing issue, not merely a construction delay; lenders are reassessing risk allocation for similar campuses.
• AI-linked borrowing is moving into riskier credit segments, where investors are demanding higher yields, stronger protections and greater certainty of contracted cash flows.
• High-grade investors are also seeking concessions and limiting exposure as hyperscaler and affiliated-project debt supply rises.
• China’s Ascend software ecosystem is becoming more usable through DeepSeek-Huawei tooling; if it scales, lower-cost domestic compute could compress some incumbent economics and redirect profits geographically.

• NVIDIA reported Data Center revenue of $89.0 billion, up 117% year on year, indicating that accelerator demand remains exceptionally strong.
• Oracle reported large AI-related contracted demand, including customer-prepaid or customer-supplied GPU commitments that reduce its own funding requirement.
• HPE reported higher Cloud & AI revenue, improved segment profitability and positive free cash flow, supporting the view that parts of the supply chain are converting demand into earnings.
• US Department of Energy-backed grid projects are intended to make more than 23 GW of incremental capacity available, a partial medium-term response to the power constraint.

• Whether Project Jupiter’s delay changes tenant payments, construction milestones, lender protections or debt syndication beyond the current notice.
• Whether AI-linked high-yield and project-finance issuance clears with widening concessions, weaker covenants or refinancing stress.
• Whether contracted cloud backlog converts into revenue and free cash flow rather than further capacity commitments alone.
• Whether DeepSeek-Huawei Ascend tooling gains measured developer adoption, production workloads and sustained utilisation outside announced partnerships.