AI circular financing: Uncomplicating the complicated

ANTHONY SAGLIMBENE – CHIEF MARKET STRATEGIST, AMERIPRISE FINANCIAL

WEEKLY MARKET PERSPECTIVES —  September 8, 2026

AI circular financing. Our attempt to uncomplicate the complicated.

At a high level, circular financing describes arrangements in which the same company sits on multiple sides of the same transaction. For example, a company can act as an investor, supplier, lender, or landlord, so that capital leaving the company can eventually loop back as revenue. In AI's closed loop, for example, chipmakers supply and invest equity capital in cloud providers that buy their processors, and those same providers lease the chipmakers' own data-center capacity to resell as compute to AI developers. Thus, the dollars move in a circular motion and, in several recent arrangements, help one company benefit at several points along the AI infrastructure buildout. In our view, it’s important for investors to recognize that these arrangements are not inherently good or bad as they stand today. At the end of the day, it’s a financing structure that allows multiple companies across the AI ecosystem to operate and build toward growing demand, which, in our view, was evidenced through several key Tech reports over the Q2 earnings season. But because a handful of behemoth technology firms now sit at the center of many of these AI circular financing deals, understanding how the money flows is likely going to increasingly matter to investors, and when trying to judge how much of the AI boom reflects genuine end demand versus capital being recycled through the ecosystem.

AI circular financing in a nutshell: Anthropic, Lambda, and NVIDIA (a recent example)

We believe a recent example of these circular financing deals helps illustrate well the web of interdependencies that now make up the AI space. Recently, The Wall Street Journal reported that Anthropic (provider of the Claude large language models) signed a $35 billion computing agreement with Lambda, a cloud provider. Here’s the structure at a high level:

• Anthropic pays Lambda for computing capacity to train and run its Claude models.
• Lambda is a cloud provider that NVIDIA has invested in, and it buys its GPUs from NVIDIA.
• The compute will run out of a Texas data center developed by Hut 8, on which NVIDIA reportedly holds the underlying lease on that facility.

That places NVIDIA in three roles at once. As an investor in the cloud provider. As a supplier of the chips. And as a data center leaseholder. Thus, Anthropic pays Lambda. Lambda buys and deploys NVIDIA hardware. And NVIDIA's capital and equipment sit beneath the entire arrangement. Notably, this arrangement is not a one-off. NVIDIA has taken large investment stakes across the AI compute landscape, and Anthropic (OpenAI, too) has signed a string of very large-capacity commitments with multiple providers as they race to lock up scarce compute. The Lambda deal is simply a recent example of how tightly the supplier, the financier, and the customer can be woven together in a single transaction.

So, what are the positives of these growing arrangements across the AI ecosystem?

In our view, and when used well, these structures can solve real problems facing the AI buildout today. 1) They can help accelerate the AI infrastructure buildout. Data centers, power, and advanced chips are in short supply. Vendor-backed financing brings capacity online faster than if every AI developer had to fund it themselves. 2) They de-risk capacity for AI developers. Long-term commitments give companies like Anthropic confidence they'll have the compute they need ahead of an anticipated supply crunch. 3) They help smaller players scale. Newer cloud providers can grow faster when a supplier is willing to invest alongside them and guarantee demand for the hardware. 4) They align incentives. A chipmaker that invests in its customers has a direct stake in their success, which can smooth the flow of capital into a fast-growing market.

But these deals come with risks that, to date, are not easily defined.

The same features that make these deals efficient and attractive for companies across the AI ecosystem also create genuine concerns. 1) Demand can look stronger than it is. When a supplier finances its own customers, reported demand may overstate what a fully “arm's-length” market might show. 2) Concentration. One company sitting at the center of many deals means its health becomes central to the whole chain. Thus, problems that develop at the hub can radiate outward across the AI landscape. 3) Layered structures (e.g., investor, supplier, landlord, customer, etc.) make it harder for investors to see how much revenue is genuinely new versus recycled. 4) Fragility risks if demand slows. If AI adoption disappoints, losses across the industry would likely build fast and cascade across interconnected investors, lenders, and operators.

The key drivers investors should be watching over the next 6–12 months, as it pertains to AI circular financing:

• The pace of AI demand, particularly enterprise adoption and tools like AI coding assistants that are driving compute needs today.

• Compute and power supply constraints, which could determine how much capacity is needed and how aggressive the financing becomes.
• Potential public listings of major AI developers (i.e., Anthropic and OpenAI), which would likely bring far more disclosure and market scrutiny to these arrangements.
• The financing environment, since the cost of capital shapes how easily these large, long-dated commitments get funded. While changes in the interest rate environment could affect deal flow here, we suspect that, as long as growth and demand assumptions hold, rates may have less sway over investment decisions. As such, a higher-for-longer rate environment, while still keeping such deals flowing, could pose a greater risk to markets down the road should AI dynamics ultimately become less attractive.

What AI circular financing and auto financing have in common and how they are different

When discussing more complex topics, such as AI circular financing, we believe it can be helpful to draw connections to other financing arrangements that may feel more familiar to investors. In our view, well-established large-automaker financing arrangements fit this bill and can provide a useful reference for framing what’s happening in the AI space currently.

For example, automakers have well-established captive finance arms across the automotive ecosystem, and these arrangements have been the lifeblood of the automotive industry for decades. Think of how a manufacturer (e.g., GM, Ford, Toyota) finances both its dealers' inventory and the end customer's purchase, all in an effort to specifically move their own product. This is a well-understood/accepted practice across the auto industry and is a necessary function for moving the millions of vehicles built every year and for providing dealers and customers with the financing they need to purchase those vehicles. 

In both AI and auto cases, manufacturers use financing to stimulate demand for their products. Both extend capital down the supply chain (in these cases, dealers or cloud providers) and, effectively, to the end user. Notably, both align the supplier's success with its customers' ability to buy.

But we believe the “differences” likely matter more to investors when evaluating AI arrangements today. For example, auto captive finance is mature, heavily regulated, and very transparent to investors and industry watchers. Conversely, AI arrangements are new, lightly disclosed, and continue to evolve on a deal-by-deal basis. In addition, auto lending is backed by collateral with a known resale market and predictable depreciation. On the other hand, AI compute assets carry uncertain useful lives and values that can fall quickly as chip generations turn over/advance. Further, auto demand is proven and relatively steady. AI end demand, while growing fast, is far less established and more uncertain over the intermediate- to longer-term. Notably, auto financing is spread across many independent lenders, while AI financing is concentrated across just a handful of central players.

So, while auto and AI lending mechanisms rhyme, their maturity, transparency, collateral quality, and concentration risks differ meaningfully. That said, we believe this connection is a helpful framework for investors and reinforces why, at least for now, investors shouldn’t necessarily make hard assumptions about whether AI circular financing is good or bad. Of course, any financing arrangements that lead to overcapacity, unnecessary complexity, or breed fraud could cause a major market disruption (given the growing scale of these AI arrangements). But it’s far from clear that’s where the end destination lies. And we believe it’s important for investors to consider this point as the AI infrastructure buildout continues.     

AI financing today is complex and moving fast. Multibillion-dollar arrangements are being struck in weeks, not quarters, and they knit suppliers, financiers, and customers together in ways that can be hard for investors to fully understand. But at the same time, we believe that speed and interconnection are likely helping to build critical AI infrastructure at a remarkable pace, based on history. Think railroads to the Industrial Revolution, to the internet, and to mobile, all built at an ever-faster pace, with increasing complexity and cost. But the more these AI deals concentrate around a few central players, the more the system's health depends on those players, a risk investors shouldn’t ignore as we venture further into the AI era.

Bottom line: For investors, we believe the key takeaway to AI financing moving forward is to keep probing to understand the flow of money. Watch where concentration risks lie, and judge future and existing deals by how much new demand is created and/or whether these complicated arrangements help reduce current bottlenecks. Investors should expect these deals to eventually support a consistent flow of higher revenue for the parties involved. If not, we believe markets are going to increasingly become exposed to a tangled web of risks that may not be easily untied.

The week ahead:

• August CPI on Friday is the week's most important release, with consensus estimates looking for core at +0.2% month over month and +3.4% year over year, unchanged from July levels.
• Other data include August PPI and existing home sales on Thursday, and preliminary September University of Michigan consumer sentiment on Friday.
• The U.S. Treasury will auction $73 billion in 3-year notes, $53 billion in 10-year notes, and $31 billion in 30-year bonds.