|
Over 300 organizations rely on The Information for exclusive insights into public and private companies, including in-depth analysis of the ways in which tech moves markets. Click here to contact our corporate and enterprise team to learn more.
Welcome back! The investment bankers in the middle of the trillion-dollar data center boom are working on their next play: capital recycling. Private equity firms and real estate funds that invest in new artificial intelligence data centers will eventually need cash returns from their investments. Some of the data center operators they own are getting too big to sell, and may have too much debt to get the best price in an initial public offering. Goldman Sachs has floated one alternative: that they sell off their stakes to other investors once long-term tenants are locked in and buildings are “stabilized.” That would spread out the risks and free up more of that initial capital for further data center development and the utilities to power them, according to a recent Goldman Sachs white paper, “Powering the AI Era.” The investment bank said insurance funds and 401(k)s looking for long-term, stable returns would be good candidates to buy the stakes off initial investors. Jason Tofsky, global head of digital infrastructure banking at Goldman Sachs, called it the next turn of the capital “flywheel” in an interview. “There’s a ton of capital invested in this and people are quite excited about the returns. But we need to build the off ramp,” he said. “That’s where the smartest people in the industry are focused right now.” There’s an important hitch, of course. To attract insurance and retirement funds, the industry would need to introduce more certainty and less risk into data center lease contracts. That includes nixing early termination clauses for tenants and extending leases to 17 to 20 years, from 10 to 15 years, according to the white paper. Whether those stricter terms become market standards remains to be seen. The financing risks are particularly pronounced in rare cases when data center developers are building in remote areas without tenants secured, as my colleague Anissa Gardizy wrote earlier this year. And there are a host of technological questions that would affect development, like how many data centers are needed to train models versus run them for customers to use. AI data centers need every dollar of funding they can get. A data center with 250 megawatts of power costs about $12 billion, including everything that goes inside of them, like high-end servers and liquid cooking equipment, according to the investment bank. About 100 times that number of megawatts—25 gigawatts—were added to the global data center supply between 2019 and 2024, totaling hundreds of billions of dollars. “A lot of it hasn't gotten liquidity yet,” Tofsky said. Now, over to Natasha... Venture Capital and Vibe Coding Anysphere, the maker of Cursor, has hired two people from Anthropic’s coding assistant team, Stephanie and I scooped Tuesday. The hires are just the latest moves in the AI talent wars. More unusual are the backgrounds of the two individuals: both previously held roles at VC firms. Boris Cherny, who led the development of Claude Code, previously worked at Coatue Management in a noninvesting role, developing the front end of the investment firm’s “Mosaic” project, a data science tool that Coatue uses to fuel its investment decisions. Cherny was at Coatue for over two years. Meanwhile, Catherine Wu, most recently a product manager for Claude Code, worked as a partner doing AI investments at Index Ventures for nearly three years before joining Anthropic. As the battle for AI talent intensifies, I anticipate we’ll see more crossover between venture capitalists and vibe coders! — Natasha Mascarenhas | 
