The repricing so far is selective rather than AI-wide: the heaviest pressure falls on companies that need outside capital to build, including Oracle, SoftBank and former bitcoin miners turned AI hosts such as Cipher and TeraWulf, while cash-rich hyperscalers like Microsoft, Alphabet, Meta and Amazon are far less exposed and some of their shareholders may welcome slower capex. The second-order risk sits with Nvidia and the wider chip supply chain, where part of future demand depends on customers that have yet to secure financing. In Asia, any push-out of orders would show up first in SK Hynix, Samsung and TSMC, a meaningful risk for a chip-heavy Kospi after its strong run. At the index level, the AI build-out has been a key support for US earnings and growth, so even a slowdown rather than a collapse trims a major pillar for broader equities and, through risk sentiment, the Australian dollar. Signs that stress is spreading beyond the neoclouds, such as hyperscaler capex cuts or wider credit spreads, would mark the shift from a sector repricing to a broader market problem.---Info via FT, (gated) ---Earlier:CoreWeave bonds trade like distressed debt as AI spending outpaces balance sheets---The AI boom’s binding constraint is shifting from chips to permits, power and credit, and investors are beginning to price the gap between announced data centre capacity and what can realistically be financed and built.Summary:At least 47 large US data centre projects have been delayed, withdrawn or blocked this year, a count described as conservativeAnalysts estimate only about a third of the planned US pipeline is credible, and more than 1,000 communities have sought restrictionsAI infrastructure debt sold off this week, with a CoreWeave bond entering distressed territory and its shares down almost 25% on the weekSB Energy and Firmus have delayed or shelved listings, OpenAI has pushed back its IPO, and Nscale’s $35bn IPO target looks hard near termTenants such as Anthropic hold walk-away rights if partners miss financing or delivery deadlinesUS AI infrastructure spending is projected to exceed $10tn between 2025 and 2032, around 3.6% of GDP a yearAmerica’s AI infrastructure boom, projected to cost more than $10tn, is showing signs of strain as construction delays, local opposition and a sharp turn in credit markets converge on the companies racing to build data centres, the Financial Times reported.At least 47 large US projects have been delayed, withdrawn or blocked this year, according to data from a regulatory intelligence firm that described the tally as conservative because it captures only the biggest sites. Analysts estimated this month that only about a third of the planned US data centre pipeline is credible, while more than 1,000 communities have sought moratoria or other limits on AI infrastructure in 2026.The financing side is now under pressure too. Debt linked to AI infrastructure groups sold off sharply this week, with one CoreWeave unsecured bond sliding into distressed territory and the company’s shares losing almost a quarter of their value over the week. Bonds funding sites leased to CoreWeave also weakened, including a Virginia deal that has dropped around 5% since it priced three weeks ago. Higher interest rates and rising inflation have sharpened questions over whether revenue at leading labs such as OpenAI and Anthropic will be enough to cover the costs.Much of the build-out depends on so-called neoclouds, young data centre providers that scaled rapidly on generous equity and debt markets, and their listing plans are stalling. SoftBank-backed SB Energy, which has yet to complete a data centre, has shelved a planned $50bn listing despite capital needs of around $175bn to fulfil roughly $430bn in contracted future revenue, mostly from OpenAI. Nscale, whose $45bn deal with Anthropic can be cancelled if it misses financing or delivery deadlines at a West Virginia project, had only 25,000 advanced chips online in August against a goal of nearly 500,000. Nebius has said it has financed only about a fifth of the roughly $250bn its planned capacity requires. OpenAI has also pushed back its own IPO.Delays are proving expensive. Cipher agreed last month to absorb around the first $360mn of cost overruns at a Texas site in exchange for more time, a change that required roughly $750mn in extra equity, partly raised through an $850mn convertible bond this week. Several Oracle sites for OpenAI’s Stargate programme have also slipped, and Stargate lost its computing lead this week.AI labs have chosen to risk losing part of their investment rather than face a shortage of computing power, while writing contracts that let them walk away if partners fall behind. With permits harder to win and utilities increasingly refusing grid connections to projects that cannot supply their own power, the coming months will test how much of the boom’s announced capacity actually gets built. This article was written by Eamonn Sheridan at investinglive.com.