Milstein Real Estate In the News

In Forbes, Columbia Business School Professor Stijn Van Nieuwerburgh looks at the surge in high-yield debt behind the AI data center buildout. So far in 2026, data center developers and AI cloud providers known as "neoclouds" have raised $80 billion in data center bonds. Of that, $37 billion, nearly half, is rated junk, more than four times the 2025 total. A Forbes analysis of FactSet data found these bonds finance 16 data center projects leased to or backed by companies such as Amazon, Meta, Nvidia, Oracle, and Jane Street, along with fast-growing neoclouds like CoreWeave and Fluidstack. Investors are now pulling back. Average yields on these bonds climbed to nearly 10% in a single week, and about $4 billion of CoreWeave's corporate debt is trading below 80 cents on the dollar. "Bond investors don't benefit from the upside. They only worry about the downside," Van Nieuwerburgh says. "The yields are beginning to reflect an increased probability of a downside scenario." Equity markets are showing the same strain, as heavily indebted developers Firmus, SB Energy, and AirTrunk have postponed or canceled their IPOs.

In a MarketWatch report on the International Monetary Fund's latest AI outlook, Columbia Business School Professor Stijn Van Nieuwerburgh's research helps show how big the AI buildout is. In a paper published by the Brookings Institution, Van Nieuwerburgh estimates that investment in AI infrastructure and specialized chips will total $10.3 trillion between 2025 and 2032. That's about 3.63% of US gross domestic product each year. The article says AI spending is likely to exceed what was spent building the railroads, the electricity grid, and the internet. Economists at ING estimate AI will account for a third of US growth this year. IMF Managing Director Kristalina Georgieva says AI could raise global growth by up to half a percentage point a year over the next decade.

In a Quartz report on Asia's debt-fueled AI infrastructure boom, Columbia Business School Professor Stijn Van Nieuwerburgh's research helps show how much borrowing is behind data center growth. In work for the Brookings Institution, Van Nieuwerburgh estimates that about $1.4 trillion of debt has gone into financing data centers, and that banks hold roughly half of it. The article covers Singapore-based Princeton Digital Group's push for a three-year syndicated loan of up to $1.2 billion, citing Bloomberg. The money would fund a 198.8-megawatt data center campus in Johor, Malaysia, and refinance existing debt. HSBC, Maybank, OCBC, Standard Chartered, and UOB are leading the deal. Earlier this year the company raised about $856 million for data centers in Indonesia, and it has set a $5 billion debt target for regional expansion.

POLITICO's Decoded newsletter looks at how Wall Street leaders are weighing the financial risks of the AI boom, and it cites Columbia Business School Professor Stijn Van Nieuwerburgh's research on the scale of the spending. In work published through the Brookings Institution, Van Nieuwerburgh estimates the cost of the AI buildout could exceed $10.3 trillion through 2032. Relative to the size of the economy, that's more than the US invested during its canal, railroad, electrification, highway, and telecommunications booms.

WebProNews cites Columbia Business School economist Stijn Van Nieuwerburgh in its look at how the AI data center boom is creating demand well beyond the server halls. Van Nieuwerburgh calculates that annual AI-related construction investment could exceed 3% of US GDP between 2025 and 2032, a larger share than historical spending on railroads, the electric grid, or highways. That scale is creating a side economy for the thousands of construction workers who spend years on each site. National Jobsite Catering serves about 1,500 meals a day at each of its 12 data center sites and has grown its revenue tenfold in four years. WhereiPark leases temporary parking lots near builds.