Private credit is moving deeper into the AI infrastructure stack. Apollo and Blackstone are preparing a roughly $36 billion debt deal for Anthropic that shows how chips are becoming financeable assets, not just scarce technology.
Apollo Global Management and Blackstone are working to bring additional investors into a roughly $36 billion debt package tied to Anthropic's AI buildout, and the interesting part is not only the size. It is the structure. The financing would help buy custom Google tensor processing units, known as TPUs, which Anthropic would then lease rather than own outright.
According to Reuters, citing Bloomberg, Broadcom is backstopping payments on the largest portions of the transaction. That matters because Broadcom helps Google develop those chips, giving lenders a supplier-linked support layer in a market where model companies are still spending far ahead of profits. For private credit, that turns a volatile AI story into something closer to an asset-backed financing.
The deal is still being marketed, and the terms could change. Investors were reportedly being asked to place orders this week, with a closing possible next week. Even so, the direction is clear. AI companies are no longer raising money only to hire researchers, train models, and acquire customers. They are now building balance sheets around the physical infrastructure required to stay competitive.
Compute is becoming the new collateral
Private credit has spent years expanding beyond traditional buyout loans into asset-backed finance, consumer credit, infrastructure, and other areas where banks have pulled back or where borrowers want more flexible capital. The AI boom gives those lenders a new category to underwrite: high-value chips attached to long-term demand for compute.
That does not mean the risk is simple. TPUs are not office buildings or aircraft. Their value depends on whether demand for specific AI workloads keeps rising, whether Google's chip architecture remains attractive, and whether Anthropic can convert Claude's adoption into durable revenue. But compared with an unsecured bet on a startup's future equity value, hardware financing gives lenders something concrete to structure around.
For Anthropic, leasing the chips changes the capital equation. A direct purchase would consume an enormous amount of cash at a time when frontier labs need flexibility for training runs, talent, safety work, enterprise sales, and product development. Leasing moves the burden into recurring infrastructure payments, which is easier to match against revenue if demand continues to grow.
The timing also explains why this deal is drawing attention. Anthropic said this week that it raised $65 billion at a $965 billion post-money valuation, while its annualized revenue run rate has crossed $47 billion. Those figures put it ahead of most private technology companies by a wide margin, but scale does not eliminate the compute bottleneck. In AI, revenue growth and chip access are now tied together.
The pressure moves to every other lab
OpenAI, Google, xAI, Meta, and Anthropic are all competing for the same scarce ingredients: chips, power, data center capacity, and engineering talent. Equity funding helps, but it does not automatically secure supply. A financing package tied to specific hardware can move faster than a general-purpose fundraising round because it gives investors a defined use of proceeds and a clearer claim on the underlying assets.
That is why this could become a template. If Apollo and Blackstone can sell down the debt while keeping meaningful stakes, they may be able to recycle capital into the next AI infrastructure transaction. Other lenders will be watching the pricing, the investor demand, and the way Broadcom's backstop is treated by the market. If the deal lands cleanly, it will not be the last of its kind.
The more difficult question is what happens if AI economics shift. Training costs may fall. Inference could move toward different chips. Enterprise customers may demand lower prices as the tools become more common. Any of those outcomes could change how lenders think about the residual value of specialized hardware. Private credit is not just financing the AI boom here. It is making a call on how long today's compute shortage lasts.
For now, Anthropic is signaling that it does not want infrastructure to decide its ceiling. That is the real market implication. The companies that lock up stable chip supply first will not merely train larger models. They will set the pace for product releases, enterprise contracts, and platform reliability. The models get the public attention, but the balance sheet is becoming just as important.
Also read: Dell's AI Server Boom Sends Stock Up 40% • Anthropic Just Dethroned OpenAI • First thing you see when Googling "OpenAI Codex app" is a fake malware website
