"Accepting High Interest Rates of Over 10%" Fear of 'AI Debt Investment' Fueled by Semiconductor Battle, Overlapping Financial Crisis Signs Amid High Interest and Securitization

Large-scale Borrowing Expansion by AI Companies Amid Semiconductor Competition
Increase in Financial Sector Investment Through Loan Receivable Trading and Securitization
Potential for Joint Damage to Companies and Financial Sector if Business Underperforms

The competition among artificial intelligence (AI) companies to secure semiconductors has led to massive borrowings, accumulating risks of default in the financial market. Anthropic has raised funds by offering an annual yield exceeding 10%, while CoreWeave and xAI have also secured large sums using semiconductors and future income as collateral. Moreover, with the addition of loan receivable trading and asset securitization, the scope of investors betting on the success or failure of AI businesses is expanding. However, if expected returns fail to materialize and the value of the chip collateral decreases, the risk of repayment disruptions spreading to financial sector losses rises significantly. This could recreate the vicious cycle observed during the global financial crisis, where loan defaults lead to financial losses, causing tightening liquidity that further weakens companies' repayment abilities.

Anthropic Borrowing Interest Rate Rises 1.5% Points in Four Months

According to the Financial Times (FT) on the 18th (local time), Anthropic plans to raise USD 9.28 billion from Blackstone to secure additional semiconductors. A Blackstone-managed fund will invest in a subordinated loan secured by Anthropic's future semiconductor rental payments over the next five years. The nominal loan interest rate is 10% per annum, and the total investor yield exceeds this rate when discount issuance and fees are included. Previously, Anthropic raised USD 4.54 billion at an 8.5% interest rate in June through the 'Project Big Sky'. In just four months, Anthropic has again undertaken borrowing, with the nominal interest rate increasing by 1.5 percentage points.

This fundraising is part of the 'Project North Star', totaling USD 61.9 billion. The project aims to secure Google's custom-developed Tensor Processing Units (TPUs) to establish a computing capacity of 1.5 gigawatts (GW), with the raised funds allocated to semiconductor volumes supplied in 2027. Anthropic will begin paying rental fees after receiving the semiconductors. The required funds are planned to be raised via USD 43.3 billion in senior loans and USD 18.6 billion in subordinated loans. The senior loans are guaranteed by Broadcom, co-developers of Google's TPU, with Bank of America, Citigroup, and Morgan Stanley managing financing. Blackstone will acquire half of the subordinated loans worth USD 9.28 billion, with the remaining USD 9.28 billion planned to be sold to other investors over the coming months.

Anthropic's choice of Blackstone as a major funding partner reflects ongoing business collaboration this year. Since March, the two have discussed establishing a corporate AI service joint venture with private equity firm Hellman & Friedman, officially announcing its founding in May with Goldman Sachs' participation. The plan is to integrate Anthropic's technology and development personnel with investor capital and client networks to introduce the AI model Claude into core corporate operations. The primary targets are investor-owned companies and external mid-sized companies lacking in-house development capabilities, offering tailored AI system construction and operational support per client task. To secure specialized personnel for this, acquisitions of technology service and consulting firms are underway.

Table 1. Status of Borrowing and Equipment Financing of Major AI Companies

Growing Dependence on External Funds for USD 500 Billion Investment Plan

Anthropic's massive demand for investment funds underpins its expanded cooperation with Blackstone from customer acquisition to large-scale financial support. Typically, companies use cash reserves and operating funds for investment and cover shortages with loans or bond and stock issuance. Anthropic reported revenues of USD 17.3 billion in Q3 this year and posted consecutive adjusted profits for two quarters, but external funds remain critical for massive semiconductor investments. Planned semiconductor and infrastructure investment over the next decade exceed USD 516 billion, requiring external funding to proceed ahead of public capital inflows, with borrowing increases adding rental and financial cost burdens to future cash outflows.

However, covering required investments with private loans increases costs due to credit risks to the company. The private credit market supplies funds to companies unable to access bank loans or public bond issuance, reflecting investment recovery uncertainties in interest rates. In June, Anthropic received a credit rating one step below investment grade from KBRA. Considering similar-rated companies' public bond yields are around 7%, Anthropic's offered yield above 10% to Blackstone implicitly includes substantial additional costs. Lenders also face risks; Blackstone’s acquired subordinated loans are positioned last in repayment priority during default and lack Broadcom’s guarantee attached to senior loans, increasing exposure to losses. If Anthropic delays rental payment, loans funded by these payments may default, causing losses to funds and investors holding the assets.

AI Companies Raise Large Sums Using Semiconductor Collateral

Such default risks accumulate across the AI sector as competing firms take on high-interest private loans. AI cloud provider CoreWeave has continuously borrowed large sums using semiconductors and customer contracts as collateral. In 2023, it raised USD 2.37 billion secured by GPUs and others, followed by a USD 7.73 billion loan agreement in May involving Blackstone, BlackRock, and Carlyle. To expand OpenAI-supplied computing infrastructure, an additional USD 2.68 billion was secured in July last year. This year saw loan agreements of USD 8.77 billion in March, and USD 3.20 billion and USD 2.68 billion in May and August respectively. Separately, convertible bond issuance terms for USD 3.61 billion and USD 3.82 billion were finalized in April and September. As of mid-month, CoreWeave's total debt reached USD 51.6 billion.

xAI, founded by Elon Musk, also secured funds via bonds and loans for data center expansion and AI model development. In June last year, it arranged USD 5.16 billion through Morgan Stanley, consisting of secured bonds and term loans. The funds supported data centers and Grok AI service development. In January this year, xAI's equipment leasing business received substantial financial backing. Apollo lease funds and affiliates supplied capital to Valuer Equity Partners' infrastructure fund, which purchased computing equipment worth USD 5.57 billion, including Nvidia GB200, to lease to xAI subsidiaries.

Irene also financed GPU purchases for computing services supplied to Microsoft through external borrowing. In June, it agreed to USD 3.76 billion in financing secured by GPUs and contractual cash flows. Arranged by Goldman Sachs and JP Morgan, the deal combined USD 2.17 billion private bonds and a USD 1.60 billion drawdown loan at a weighted average interest rate of 6%. In late August, it additionally secured USD 2.48 billion for purchasing Nvidia Blackwell Ultra for the Mackenzie data center in British Columbia, Canada. Participating were Blue Owl and PIMCO, structuring the deal with senior secured loans and bonds each at USD 1.24 billion, fixed interest rate set at 9%.

Long-term Fund Betting on Chip Rental Fees and Residual Value

With increased borrowing secured by semiconductors, loan receivables are being traded to other investors. In July, some of the USD 36.1 billion private loans prepared for Broadcom and Anthropic’s semiconductor procurement were traded through Wall Street banks. Asset-backed securities (ABS) issuance backed by chip-related loan receivables or rental rights is also underway. These securities are sold to investors who earn returns from loan principal and interest as well as rental income. Financial firms recover existing loans through securities sales, enabling further funding. Infrastructure financing projects involving Nvidia and financial companies are planning to use this method to expand participation from insurers and pension funds, designing investment products based on chip rental income and residual value to attract long-term capital.

However, the chip rental fees and residual values underpinning these investment products carry inherent uncertainties due to technological changes. The introduction of newer chips depresses demand for existing equipment, lowering rental income and used equipment prices. Credit rating agency S&P Global cited rapid GPU obsolescence as a risk factor, noting shorter economic lifespans complicate recovery estimates during defaults. Transactions requiring repayment of remaining debt via re-leasing or equipment sales post-lease are especially vulnerable to price declines. Concurrent asset disposals by multiple firms facing financial strain could flood the market, further dropping used prices. Consequently, loan defaults paired with collateral value drops could amplify losses and increase burdens on ABS principal and interest payments.

Funds Chased for High Returns Flock to AI—Financial Crisis Could Recur

The 2007-2008 global financial crisis was also sparked by a chain reaction between collateral value declines and securitization failures. Financial institutions bundled subprime mortgages—loans to low-credit borrowers—into mortgage-backed securities (MBS), further structured into collateralized debt obligations (CDOs). While housing prices rose, borrowers unable to repay could sell homes or refinance, but falling prices blocked refinancing and forced foreclosure simultaneously, leading to increased defaults, plummeting security prices, and credit ratings. Funding through security sales halted, restricting new loans, dampening housing demand, and triggering a vicious cycle of collateral value drops.

The high-yield expectations that supported risky loans then are echoed in today’s AI finance. According to the U.S. Federal Reserve, initial rates on subprime hybrid adjustable-rate mortgages were about 8%, with non-preferred rates 2-3% points higher. While offering high returns to investors, borrowers bore rising repayment burdens amid rate hikes. Anthropic’s offered yield above 10% encompasses investor repayment risk and company financial costs. Increased investor anxiety over repayment ability post-investment leads to more bond sales and further yield rises caused by price drops. CoreWeave saw some bond yields spike to about 15% this month amid investor sell-offs. If such anxiety triggers lending contractions, equipment investments by borrowing-dependent companies will inevitably be delayed. Delayed operations reduce timely revenue, weakening capacity to repay existing debt. This cycle of default fears causing liquidity crunches, thereby worsening defaults, exposes the AI investment market to similar risks.