Two of the biggest players in the alternative investment space, Apollo Global Management and Blackstone, are preparing debt financing of approximately $36 billion to support the expansion of computing infrastructure startup Anthropic. The debt is tied to the purchase of specialized AI chips from Google, which Anthropic will then use and lease as computing capacity based on tensor processing units, or TPUs.
The structure of the transaction shows how quickly AI infrastructure is becoming a separate asset class around which gigantic financing packages are forming, similar to project finance in the energy or telecommunications industries. For investors, this means that a significant part of the value of generative AI is no longer just the software itself, but also the commercially structured debt and infrastructure projects behind it.
How the $36 billion debt financing is supposed to work
Apollo and Blackstone are reportedly approaching institutional investors to participate in a debt package that is expected to be around $36 billion. The collateral for this debt is to be the infrastructure associated with the purchase and operation of custom TPU chips from $GOOG, which will then be used by Anthropic and further monetized by providing AI computing power to third parties.
Both investment groups plan to sell off some of the debt to investors and retain some on their own balance sheets. This syndication structure is typical of large infrastructure transactions where the original arrangers want to reduce risk concentration but retain a significant share of the proceeds. The debt will be spread over time and tied to long-term contracts related to the use of computing capacity.
Simply put: the debt capital finances the purchase of a computing "hardware stack" from Google, Anthropic uses this power for its models while reselling it to customers, and the cash flow from this operation pays the interest and principal. For Apollo, Blackstone and other investors, it's a way to get exposure to the growing demand for AI computing power without having to invest directly in AI company stocks.
Broadcom's role and securing payments
Broadcom $AVGO, which is a co-developer of TPU chips for Google $GOOG, is playing a key role in the deal. Broadcom translates Google's design architecture into physical silicon and ensures that the TPUs are manufacturable in the required volume and quality.
As part of this financing, Broadcom provides payment guarantees on the largest parts of the transaction. In practice, this means that a portion of the payment obligations associated with chip shipments are covered by guarantees from an established semiconductor player, reducing the riskiness of the debt package for investors. It is this combination:
long-term contracts for computing capacity
the involvement of Google as a supplier and partner
and guarantees from Broadcom
makes the deal a structured infrastructure project rather than a purely speculative financing of a single startup. Investors are encouraged to place their orders later this week, with the aim of closing the deal in the following week, although details are still subject to change, according to market reports.
Anthropic: valuation heads into the stratosphere
The debt megabuck comes shortly after Anthropic announced a new round of funding. The firm raised $65 billion at a valuation of around $965 billion after cash, leapfrogging rival OpenAI, according to published figures. The round was led by funds such as Altimeter Capital, Dragoneer, Greenoaks and Sequoia Capital, supplemented by big strategic players from the tech industry.
At the same time, Anthropic said its annual revenue has already surpassed $47 billion, which puts it in the league of tech giants, although it still technically remains pre-IPO. The round also includes about $15 billion in previously agreed investments from hyperscalers, including about $5 billion from Amazon.
In addition to financial investors, strategic hardware and memory partners such as Samsung, SK Hynix and Micron have also joined the funding. This shows that Anthropic is no longer just an AI software company, but is becoming a central node in a broader ecosystem of chip, memory and cloud infrastructure suppliers.
TPU as the backbone of AI infrastructure
The Tensor Processing Unit is a specialized chip designed by Google for machine learning-related computations, particularly for working with matrix math, which is the basis for training neural networks as well as running models in production. The TPU belongs to the category of application-specific integrated circuits, optimized for a specific type of computation, and thus different from general GPUs and CPUs.
Google began using TPUs internally in 2015 and has gradually made them available to third parties through the Google Cloud TPU service within Google Cloud Platform since 2018. For customers, this means they can use TPUs as a dedicated compute resource, similar to virtual servers, but with significantly better efficiency for AI tasks.
The expanded collaboration between Google, Anthropic and Broadcom includes Anthropic's access to approximately 3.5 gigawatts of compute capacity by 2027, representing a huge increase in available power. Already last fall, Google and Anthropic announced a deal worth "tens of billions of dollars" for one million TPUs to bring over 1 gigawatt of AI capacity later this year. These contracts are the cornerstone for structuring the current debt financing.
Anthropic, OpenAI and SpaceX: the race for AI and space valuation
Anthropic finds itself in direct competition with OpenAI not only for technological leadership, but also for the attention of the capital markets ahead of its expected IPO. OpenAI announced a funding round in March that valued the company at roughly $852 billion. Anthropic has now surpassed that level, reaching a valuation approaching one trillion dollars.
In the broader technology context, SpaceX also belongs in this club, which was valued at around $800 billion last year and has climbed to around $1.25 trillion following its merger with Musk's AI firm xAI. SpaceX is also heading towards its own IPO and is targeting a valuation of around $2 trillion, according to market reports.
Anthropic was founded in 2021 by former OpenAI executives and is now seen as one of the most serious competitors in the field of big language models. Anthropic, OpenAI and SpaceX are expected to become publicly traded companies within a few years, even though all three are currently spending more money than they are making.
The product side of this story is also part of the story: Anthropic recently introduced a new model, the Claude Opus 4.8, which promises vastly improved capabilities in coding and other professional tasks over previous versions. Technological advances are thus running parallel to the financial escalation in valuations and infrastructure capacity.
What's in it for investors
For investors who follow the AI sector, this deal delivers several key signals:
AI infrastructure is becoming a new asset class. The $36 billion debt package to buy TPUs shows that computing capacity can be financed in a similar way to power plants or data centers.
The role of private equity in AI is growing. Apollo and Blackstone are not just looking at traditional buyouts, but are stepping in to fund the very backbone of the AI ecosystem.
Anthropic is moving to be one of the main "anchors" of the AI investment universe. A valuation approaching $1 trillion, more than $47 billion in annual revenue, and a close tie to hyperscalers make it one of the key players alongside OpenAI and SpaceX.
For the public markets, this is a harbinger. Once these companies go public, investors will value not just software, but also access to infrastructure, long-term contracts and the ability to finance computing capacity, much like energy projects.