This post was originally published on this site.
The shift from public bond issuance to bespoke private financing suggests the cost of the AI buildout is outrunning what traditional debt markets can comfortably absorb, which puts private credit giants at the centre of the next phase of spending. For chipmakers such as Broadcom and Nvidia, lender-backed purchases support demand visibility well beyond what buyers could fund from their own balance sheets. Off-balance-sheet lease structures may ease pressure on borrowers’ credit metrics, but they also obscure true leverage, which credit investors are likely to scrutinise as deal sizes grow. Any wobble in AI-related revenue would now ripple through a wider set of lenders than before.
—
Earlier:
—
The AI arms race has grown too expensive for corporate cash flow and bond markets alone, so its biggest players are now asking Wall Street’s private lenders to help pay for the chips.
Summary:
- Broadcom is working to arrange more than $50 billion in financing for OpenAI’s custom AI chip, with Apollo and Blackstone among the lenders approached, according to the Wall Street Journal (gated)
- Oracle is in talks with Apollo and Goldman Sachs to fund a large chip purchase, likely through a separate entity that buys the chips and leases them to Oracle
- SpaceX has discussed a $40 billion financing for Nvidia chips with lenders, a story first reported by the Financial Times
- Talks are at an early stage and deal sizes could change; Broadcom and Oracle are aiming to close before year-end
- Cloud providers have issued hundreds of billions of dollars in bonds for AI, and newer buyers such as OpenAI and Anthropic lack the balance sheets to buy hardware outright
Some of the biggest names in artificial intelligence are lining up financing deals worth tens of billions of dollars each to pay for computing hardware, as the cost of building data centres strains conventional funding routes, the Wall Street Journal reported in an exclusive, citing people familiar with the discussions.
Broadcom has spent recent weeks working to arrange more than $50 billion in financing for the custom AI chip it is developing with OpenAI, according to the report. Apollo and Blackstone are among the lenders Broadcom has approached about taking part. The talks are at an early stage and the size of the deal could change, though it is expected to close before the end of the year and could cover several gigawatts of OpenAI chip capacity.
OpenAI’s chip programme, known internally as Nexus, uses names drawn from types of peppers, with its first and second-generation chips called Jalapeño and Serrano. The financing would follow the partnership the two companies announced a year ago to jointly develop 10 gigawatts of OpenAI’s custom chips using Broadcom’s networking technology, with deployment planned from the second half of 2026 through the end of 2029.
Separately, Oracle is in talks with Apollo and Goldman Sachs to fund a large chip purchase and also hopes to finalise a deal this year, while still talking to several potential partners. The financing is designed to cover the gap between paying for hardware and collecting cloud revenue from it. The structure would likely see investors fund a separate company that buys the chips, which Oracle would then lease over time. That would let Oracle avoid adding more debt of its own as it competes with larger, cash-rich rivals. The number of chips involved is unclear, but the Journal noted that Nvidia chips for a single 1-gigawatt data centre would cost tens of billions of dollars.
SpaceX has also spoken to lenders in recent days about a $40 billion financing for Nvidia chips, according to the report, which said the Financial Times first reported those talks.
The wave of deals reflects how expensive AI infrastructure has become. Cloud providers such as Amazon Web Services and Oracle have historically paid for hardware from their own cash flows, and for the AI buildout they issued hundreds of billions of dollars of bonds, which the Journal said pushed public debt markets to their limits. A newer group of buyers, including OpenAI and Anthropic, lacks the financial firepower to buy hardware outright. Having long rented most of their computing capacity from cloud providers, these labs now want to own more of it to cut costs and reduce their reliance on others. How these private financings are structured, and on what terms, will show how much further the buildout can be stretched.
This article was written by Eamonn Sheridan at investinglive.com.
