- Blackstone has been investing big in the AI revolution, including billions in data centers.
- Chinese AI model DeepSeek is raising questions about whether these investments will pay off.
- President and COO Jon Gray said the company has spent a lot of time on DeepSeek this week.
A top Blackstone executive on Thursday addressed the threat that AI model Deepseek could pose to the company’s multibillion-dollar bet on AI infrastructure, saying the firm is watching the situation closely.
“We’ve obviously been spending a lot of time the last week looking at the impact of DeepSeek,” president and COO Jon Gray said in a fourth-quarter earnings call, adding: “The real question is what is demand going forward?”
Blackstone, which calls itself the largest data-center provider in the world, has $80 billion in existing data center assets — and has touted more than $100 billion in the pipeline. Gray didn’t address the firm’s pipeline but acknowledged that “the cost of computing is coming down pretty dramatically,” which could change demand for future investments in how AI is powered.
“So we still think it’s a very important segment, and there’s a way to run, but obviously, we’re watching what’s happening very closely,” he said.
Blackstone is the latest company to react to concerns that DeepSeek, an AI model from China, is 20 to 40 times more efficient than OpenAI, according to an analysis by Bernstein. This has called into question both Silicon Valley valuations and trillions in AI infrastructure investment.
Blackstone’s share price, which had opened up nearly 1.5%, dropped at the market open and recently traded down more than 4% to $177.50 a share. Blackstone’s earnings handily beat expectations, with the firm setting a record for fee-related earnings in the fourth quarter.
Gray suggested that lower costs could drive more adoption and, therefore, more demand for infrastructure, echoing statements from tech CEOs like Microsoft CEO Satya Nadella. Gray’s comments reflect Silicon Valley’s new favorite economic theory, the Jevons Paradox, which posits that usage will increase as the price of a resource declines.
Citing Mark Zuckerberg’s comments on Wednesday, he said that data centers might need to be more “fungible” than before, which may mean “less training” of AI models but potentially more “inference,” the practice of using AI to come to predictions or conclusions, as well as increased demand for cloud or enterprise usage.
The larger picture, that “our lives are migrating online” has not changed, Gray said. He also said the firm’s $80 billion in data center assets is tied to existing demand via leases to large, well-capitalized companies.
“We’re not doing things speculatively,” Gray said. “It’s based on the demand signals from our tenants.
Blackstone’s investment in data centers began in earnest when the firm took data center REIT QTS Realty Trust private in 2021 for $10 billion. It’s continued to make big bets like its $16 billion-plus take-private last year of AirTrunk, the largest data center transaction of all time.
During an earnings call last year, CEO and cofounder Steve Schwarzman announced his ambitions for the firm to be the world’s largest investor in AI infrastructure.
He projected $2 trillion in global investment to build and facilitate data centers over the next five years and that the AI revolution would contribute to a 40% increase in electricity demand in the US over the next decade.
This has already spurred the firm to invest in adjacent sectors, like its recent $1 billion purchase of a Virginia power plant in a data center hot spot.
Blackstone has been ahead of the curve in this strategy, and other investors are now jumping in. A KPMG survey found that 40% of US asset manager respondents told KPMG that it’s their top real estate investment priority over the next two years, up from 27% just six months ago.
Although DeepSeek could threaten demand for data centers, China isn’t turning away from investing in AI infrastructure either. China recently announced a roughly $140 billion investment in data centers.
Read the full article here
