Through the AI boom, the main constraints have shifted from chips to power to memory — and the next potential choke point may be capital. That may seem odd while Anthropic prepares for an IPO valued at more than $1 trillion and data centers continue to expand, but lenders are growing more cautious.
Lenders growing wary
Over the past week, signs of fear and frustration among the banks and financiers that fund data‑center growth have become more pronounced. DBS’s chief executive warned in a Bangkok interview with Semafor reporter Shelly Banjo: “There’s shortages in chips. There’s shortages in power. And soon, there will be shortages of capital.”
A major data‑center financier told the author bluntly: “We’re not an ATM for your dystopian future.” He added that his firm would provide financing if the risk‑return math justified it, but currently it does not. Concerns include the difficulty of enforcing guarantees from hyperscalers and the financing gaps created by construction delays blamed on local protests.
Confidence can crack suddenly
Investor and lender confidence often erodes slowly and then collapses quickly. The article cites several market examples showing shifting lender behaviour: Volta’s $5 billion Norwegian data‑center deal, which months ago might have sailed through, now faces tougher investor terms. CoreWeave’s bonds have plunged. The cost to insure against a SpaceX default has risen about 80% compared with three months ago.
In early pricing talk for SpaceX’s marketed $40 billion loan this week, the loan was said to be roughly 1 percentage point more expensive than the company’s existing bonds. The tone is captured by Elon Musk’s quoted reaction: “Yikes.”
What this means for AI financing
The piece notes that some companies will still likely be able to raise large sums — OpenAI is expected to have little trouble raising the $30 billion it seeks — and equity markets may recover from recent chip‑market turbulence. But historical precedents are a caution: telecom bonds weakened six months before the dot‑com crash, and subprime mortgage bonds peaked eight months before the S&P 500 did in 2007. Equity investors are paid to dream; lenders are paid to avoid losses.
Scale of AI investment
Spending on AI infrastructure remains strong for now. Goldman Sachs projects a roughly $7.6 trillion boom in AI infrastructure investment through 2031.
Conclusion
Even as demand for AI capacity and large funding rounds persist, increasing lender caution could become a real constraint. Tighter terms, higher costs of credit protection and more selective financing may slow project timelines, alter deal structures and raise the cost of capital for major AI and data‑center projects.
