Bombshell Breakdown: The "Capital Internal Circulation" in the U.S. AI Scene

Bombshell Breakdown: The "Capital Internal Circulation" in the U.S. AI Scene

Morgan Stanley’s latest warning states that as the investment cycle for artificial intelligence computing accelerates, the AI industry’s capital intensity is reaching unprecedented levels. However, outside the hyperscalers, the entire ecosystem’s capital remains limited. To support massive infrastructure expansion, the US AI sector is evolving a highly complex, interwoven “new financing structure” and internal capital circulation system.

On March 23, according to Chasing Wind Trading Desk news, Morgan Stanley said in its latest research report that in the next few years, AI-related investment will account for about 50% of the large-cap stocks’ total capital expenditure, and its capital intensity will surpass that of the previous internet bubble era. This investment is massive and highly front-loaded, resulting in a severe mismatch between recent capital demand and AI revenue realization. This mismatch has spawned the emergence of various new financing structures throughout the ecosystem.

The report uses OpenAI as a core case, drawing a capital flow map involving Nvidia, Microsoft, Oracle, CoreWeave, Amazon, AMD, Disney, and others.

Morgan Stanley points out that the current AI boom is built on vendors and clients mutually supplying funds, warrants, and off-balance-sheet guarantees. While this model greatly accelerates infrastructure expansion, it may also exaggerate surface contract pricing and hide real economic leverage and risk off the balance sheet, potentially causing “sky-high contracts” featured in headlines to be overestimated.

“Suppliers nurture customers, customers nurture suppliers”— Panorama of Circular Financing Structure

The core finding of the report is: The AI ecosystem has formed a highly interconnected structure where suppliers fund customers and customers feed back to suppliers. These arrangements include:

Supplier financing and favorable termsLong-term purchase commitments (Take-or-Pay contracts)Revenue sharing arrangementsSupplier buyback agreementsHigh customer concentrationThird-party guarantees and endorsementsIP licensing in exchange for model accessEquity investment for computing power commitment

Morgan Stanley believes these are essentially financing mechanisms that allow many ecosystem participants to expand infrastructure well beyond what their own cash flow could support. The bank’s report explicitly warns:

While these financing agreements accelerate data center construction, they may also preemptively exhaust future demand and redistribute risk among counterparties.

Core Case: $100 Billion Capital Flows in the OpenAI Ecosystem

Morgan Stanley says, taking OpenAI’s ecosystem as the most transparently disclosed case, we can clearly see the astonishing scale of this internal capital circulation:

1. Nvidia’s deep involvement as compute king

  • Direct investment: In February 2026, Nvidia finalized plans to invest $30 billion in OpenAI.
  • Support for key customers: Nvidia holds about 9% of CoreWeave (CRWV) shares (worth about $4 billion) and has added another $2 billion investment.
  • Buybacks and leasing: Nvidia agreed to buy back up to $6.3 billion in unsold compute capacity from CRWV and signed a $1.3 billion agreement to rent back GPUs supplied by CRWV over four years. Additionally, Nvidia itself has $22.7 billion in future lease obligations (mainly data centers).

2. Microsoft’s comprehensive penetration

  • Equity and compute commitments: Microsoft’s $13 billion investment in OpenAI is now valued around $135 billion (about 27% on a fully diluted basis). OpenAI has committed to buying an additional $250 billion in Azure services.
  • Revenue sharing and hardware procurement: The two have a revenue-sharing agreement, expected to bring Microsoft $6.1 billion in revenue in calendar year 2026. Microsoft is estimated to spend about $45 billion in fiscal 2026 on Nvidia chips and has signed a $10 billion server lease agreement with CRWV. Microsoft’s own additional leasing agreements reach $155.1 billion.

3. Oracle and AMD’s massive orders

  • Oracle: Has signed contracts worth about $40 billion with Nvidia (for about 400,000 chips) and another totaling $5.7 billion for GPU deployment with AMD. OpenAI agreed to purchase $300 billion worth of compute from Oracle over about 5 years. Oracle’s data center lease commitments are as high as $261 billion.
  • AMD: Announced a $100 billion strategic deal with OpenAI (deploying 6 GW of AMD GPUs). As part of the agreement, AMD issued up to 160 million warrants to OpenAI.

4. Amazon and Disney’s cross-industry entry

  • Amazon: Pledged to invest $50 billion in OpenAI ($15 billion upfront, $35 billion subject to milestones). The scale of their partnership expanded by $100 billion over 8 years. Amazon currently has $96.4 billion in uninitiated leases.
  • Disney: Invested $1 billion in OpenAI (with additional warrant options) to obtain usage rights to OpenAI’s models, while OpenAI secured a 3-year license to use Disney IP. This model of exchanging IP for compute is essentially a disguised financing that doesn’t require imminent cash outlay.

Morgan Stanley states that in this AI capital network woven from hundreds of billions of dollars, new funding often covers only part of the total compute commitment, while the fulfillment of the remaining contracts highly depends on future revenue growth or the next round of fundraising. Investors must beware that this “left hand to right hand” capital cycle could pose systemic fragility when pricing AI mania.

Meanwhile, Morgan Stanley explicitly lists seven potential risks of this circulating structure in the report:

1. Warrants distort real pricing

Customers exchange long-term purchase commitments for supplier warrants, making headline contract amounts unable to reflect repeatable true pricing levels.2. Off-balance-sheet guarantees hide real leverage

Third-party guarantees by cloud suppliers support data center construction, but usually don’t appear on the guarantor’s balance sheet, causing significant divergence between reported leverage and actual economic leverage.3. IP licensing masks true operating costs

Content creators license IP to access AI models on favorable or non-cash terms, underestimating the true operating costs and cash requirements of AI labs.4. Supplier equity investments amplify debt risks

Supplier equity investments back up cash flow for other suppliers selling to the same customer, enabling them to take on more debt and causing chain effect, further boosting capacity expansion.5. High customer concentration amplifies counterparty risk

Revenue growth increasingly relies on the success of a few AI labs, significantly increasing concentration risk.6. Revenue sharing arrangements blur true demand

Revenue sharing among multiple parties may allow several entities to record the same revenue under US GAAP, making it hard to judge true AI demand.7. Buyback agreements may inflate demand

Supplier buyback agreements shift risk back onto suppliers and provide downside protection to customers, potentially artificially inflating surface demand figures.

 

~~~~~~~~~~~~~~~~~~~~~~~~

The above excellent content originates from Chasing Wind Trading Desk.

For more detailed interpretation, including real-time analysis and frontline research, please join [Chasing Wind Trading Desk ▪ Annual Membership]

Risk Notice and DisclaimerThe market carries risk and investments require caution. This article does not constitute personal investment advice and does not take into account the specific investment goals, financial situation, or needs of individual users. Users should consider whether any opinions, views, or conclusions herein suit their particular circumstances. Investment is at your own risk.