🚨 🇺🇲 The US bond market is starting to say the trillions of dollars promised to build its infrastructure could become impossible to finance under current conditions and I think we are witnessing a genuine regime shift. 📊 For three years, markets have rewarded companies announcing the most spectacular investment plans, but now investors are finally starting to ask the right questions. Who is financing these projects? At what cost? With what guarantees? For what return? 47 major data center projects have been delayed, canceled, or blocked in 2026, while only 35% of the US data center pipeline is actually credible according to Bernstein. We are talking about a potential investment wave exceeding $10 trillion between 2025 and 2032 - FT 📉 Companies developing these infrastructures must spend billions upfront to purchase GPUs, build data centers and secure electricity supplies, while the associated revenues may not materialize for several years. Their entire business model depends on their ability to keep raising capital at reasonable costs, but that assumption is currently being challenged. CoreWeave lost 25% last week while some of its bonds suffered a sharp sell-off as the market is wondering whether these projects are financially viable at all. ➡️ Few other examples to understand. Cipher agreed to absorb the first $359 million in cost overruns on one project, creating an additional equity requirement of $750 million. Nscale has a contract potentially worth $45 billion with Anthropic but Anthropic can walk away if certain financing or delivery commitments are not met. Meanwhile, SB Energy needs around $174 billion in capital to fulfill contracts representing $430 billion in future revenue, despite not having completed a single data center. Some companies are being valued on enormous contracts without having demonstrated that they can actually finance and deliver the infrastructure behind them. ⚠️ We also need to revisit the circular financing structure of this industry which I have discussed before. Nvidia invests in companies that buy its chips, AI labs sign contracts that allow developers to raise debt, and SPVs or private credit funds finance infrastructure based on expected future revenues. As long as capital keeps flowing, the system works, but if investors suddenly demand much higher yields, the entire mechanism can start breaking down. Projects get delayed, GPU orders are postponed, developers face financial difficulties, and creditors reassess their exposure. A financing crisis can trigger an investment crisis before demand for AI even starts slowing down and many people tend to forget that. Moreover, that risk is concerning given how important AI-related investment has become as a driver of US economic growth. 📈 I do not think we are witnessing the collapse of the entire industry. Instead, we may be entering a much more brutal selection process between financially viable projects and those built on unrealistic assumptions. Lower Fed rates will not solve the problem as if credit spreads widen faster than risk-free rates decline, the overall cost of financing will continue to rise. 💸 I maintain that we need to return to a cost of capital and valuation levels that are consistent with the actual economic profitability of these projects. The technological success of AI does not guarantee financial success for the companies funding its infrastructure. History has already taught us this through railroads, telecommunications, and the internet. A technology can revolutionize the economy while destroying enormous amounts of investor capital. The current tensions in the bond market are a serious warning sign ⚠️ *FT link: ft.com/content/9ea5a37d-15c9…

🇺🇲 Microsoft Slashes Internal Claude Spending by a Third - The Information theinformation.com/articles/…