Pimco Cut AI Debt Stakes by Half and Skirted Losses

Pimco halved its stakes in the Meta and Oracle data center bonds it anchored, selling into strength before prices fell, Bloomberg reported.
2 min read
Pimco Cut AI Debt Stakes by Half and Skirted Losses
  • Pimco now holds about 50% less of the roughly $18 billion of Meta-related debt and $10 billion of Oracle’s Michigan data center deal it backed, filings show.
  • It flipped about $3 billion of Meta’s Beignet bonds above par, while Oracle’s Michigan notes, issued at 98.7 cents, have slid to around 88 cents.
  • The pullback shows how quickly large lenders are repositioning as AI infrastructure debt reprices, a signal for data center developers who rely on the same buyers.
Key Takeaways

Pimco helped kick off the AI debt boom, and it has been quietly taking chips off the table. The bond giant cut its stakes in two of the biggest AI data center financings by about half and has so far dodged the losses now hitting other holders, Bloomberg reported exclusively.

Pimco anchored about $18 billion of the Meta financing and $10 billion of Oracle’s Michigan campus deal. It then trimmed both while prices held up.

Selling Into Strength

The Meta trade shows Pimco’s strategy. The firm sold about $3 billion of Beignet bonds within six to eight months, while they traded above par, according to a person familiar with the trades. After pricing, the notes climbed above 110 cents on the dollar.

Oracle’s deal followed a different path but reached a similar outcome. The Michigan bonds priced at 98.7 cents and now trade near 88 cents on Trace. Pimco joined as an anchor investor and paid well below the official price. As a result, the decline had less impact.

Still Holding Plenty

Pimco has not walked away, however. The firm remains one of the largest holders of both the Louisiana and Michigan financings. Both have fallen in recent weeks. Meanwhile, Broadcom’s AI deals trade near 96 cents.

Meta’s five-year credit protection costs hit an all-time high last week, Bloomberg reported. Investors expect the company to borrow more. Mark Zuckerberg had just told investors that Meta would increase its AI spending even further.

Picking Its Spots

Pimco has taken a selective approach elsewhere. The firm ranks among private credit’s loudest critics. It has warned about lax underwriting as the market has grown beyond $1.8 trillion.

The firm passed on several major deals. These included a jumbo financing for Broadcom and Anthropic, an approximately $11 billion SoftBank junk bond, and Oracle-linked loans for a New Mexico data center. However, Pimco took a $4 billion share of a SpaceX financing, Bloomberg reported Monday. A Pimco representative declined to comment.

Why It Matters

For CRE, pricing offers the clearest signal. The trend aligns with Bloomberg’s findings on double-digit yields across at least 18 data center financings. We covered that story earlier this week. One example was a $5 billion loan for neocloud Volta, which priced above 11%.

Data center developers and campus owners compete for the same lenders. Tougher pricing can therefore raise their borrowing costs. Pimco’s trades show that one of the market’s largest buyers is already managing that risk.

Still, not everyone sees a bubble. Blackstone’s Jon Gray argued on Bloomberg Television that AI builders are racing to meet demand, not simply betting that demand will emerge.

What’s Next

Watch whether the recent price declines on the Michigan and Louisiana bonds stabilize. Also track Meta’s credit protection costs as investors assess the company’s plans for additional borrowing.

CRE Daily - No Cap

podcast

No CAP by CRE Daily

No Cap by CRE Daily is a weekly podcast offering an unfiltered look into commercial real estate’s biggest trends and influential figures.

CRE Daily Newsletters

Join 65k+
  • operators
  • developers
  • brokers
  • owners
  • landlords
  • investors
  • lenders

who start their day with CRE Daily.

The latest news and trends in commercial real estate delivered to your inbox. Get smarter about what matters in just 5-minutes or less.