Yesterday, AI assets in the U.S. stock market collapsed again.

The trigger was a 2032-maturing bond issued by AI computing power company CoreWeave, whose secondary market yield briefly approached 15%.

This bond was only issued in June this year with a coupon rate of 9.625%. In just four months, the market's required rate of return has surged by more than 5 percentage points.

For context on credit ratings, Moody's rates this bond B1, while S&P assigns a B rating. Both clearly fall into non-investment grade, commonly known as junk bonds.

Bond market concerns quickly spread to the stock market.

On October 14, CoreWeave's stock price plummeted approximately 16%, while AI infrastructure companies like Nebius and Oracle also faced massive sell-offs. From chips and servers to data centers, the entire AI industry chain came under collective pressure.

This reveals a particularly noteworthy contradiction.

On one hand, demand for AI computing power remains robust. CoreWeave holds substantial long-term customer orders and has continuously raised its capital expenditure plans this year, attempting to meet market demand through more aggressive expansion.

On the other hand, capital markets' attitude toward this expansion model has begun to shift. The question looms: Can these AI companies relying on borrowed money generate sufficient profits to cover such exorbitant financing costs?

This issue becomes increasingly pressing amid renewed expectations of U.S. interest rate hikes.

To some extent, CoreWeave's debt crisis exposes a deeper transformation in the current AI computing power industry: As capital becomes more expensive, the financing-driven expansion model for AI computing power is facing a genuine stress test.

Today, Silicon-Based Insider will discuss this matter.

/ 01 / $14 Billion in New Debt in Six Months: CoreWeave's Aggressive Expansion

To understand CoreWeave's debt issues, let's examine several data points first.

Over the past three years, CoreWeave's balance sheet has expanded at an explosive pace.

In 2023, the company's total liabilities stood at $5.109 billion. By the end of 2025, this figure had reached $45.967 billion.

And as of June 2026, total liabilities had further grown to $72.046 billion—approximately 14 times the 2023 level.

Of course, total liabilities don't equate entirely to interest-bearing financial debt, as they also include items like lease obligations and deferred revenue.

When examining interest-bearing debt alone, the growth remains staggering. At the end of 2025, CoreWeave's principal interest-bearing debt stood at approximately $21.615 billion. By June 2026, this had grown to $35.551 billion.

In just six months, new interest-bearing debt increased by nearly $14 billion—a 64.5% surge, averaging over $2.3 billion in additional debt per month.

The pressure from rapid debt growth has already begun reflecting on CoreWeave's income statement.

Throughout 2025, CoreWeave's net interest expenses reached approximately $1.229 billion, accounting for about 24% of annual revenue. By the first half of 2026, net interest expenses alone hit $1.176 billion—nearly matching the entire previous year's total.

On a quarterly basis, net interest expenses were $388 million in Q4 2025, rising to $536 million in Q1 2026, and further climbing to $640 million in Q2 2026. Over just two quarters, single-quarter net interest expenses grew by approximately 65%.

For comparison, CoreWeave's revenue in Q2 2026 was $2.575 billion. This means the company incurs about $25 in net interest expenses for every $100 in revenue.

In terms of cash flow, CoreWeave generated $3.663 billion in net cash inflows from operating activities in H1 2026. However, during the same period, purchases of fixed assets alone consumed $14.117 billion—nearly four times the operating cash inflow.

Calculating free cash flow simply as operating cash flow minus fixed asset purchases, CoreWeave recorded a negative $10.454 billion in H1 2026.

Looking ahead, this deficit may widen further.

In August 2026, CoreWeave raised its full-year 2026 capital expenditure guidance to $35-39 billion. For comparison, the company's full-year revenue guidance for the same period stands at just $12.4-13.2 billion.

In other words, CoreWeave plans to invest nearly three times its expected annual revenue in capital expenditures this year. This is precisely why CoreWeave's debt issues warrant such close attention.

/ 02 / How Did CoreWeave Accumulate $35 Billion in Debt?

An even more pressing question is: How did CoreWeave manage to borrow so much money?

Behind this lies a financing system built around GPU assets and long-term customer contracts. Broken down, CoreWeave's interest-bearing debt primarily consists of three components.

The first and most critical category is DDTL (Delayed Draw Term Loans).

DDTL can be understood as financial institutions providing credit lines to companies, which then draw down funds in batches according to project construction and procurement progress.

For CoreWeave, this means securing long-term computing power orders from major clients like Microsoft and Meta, then using future cash flows generated by GPU equipment and customer contracts as collateral to borrow funds for equipment purchases. The company gradually repays these loans through computing service revenues over the next few years.

CoreWeave has scaled this financing model to a substantial level. As of June 2026, the combined outstanding principal of multiple DDTL projects reached $13.643 billion, accounting for 38.4% of all interest-bearing debt.

However, this financing model comes at a cost—namely, high financing costs.

According to Q2 financial reports, early DDTL loans carried effective interest rates as high as 15% for DDTL 1.0, approximately 11% for DDTL 2.0, and around 9% for DDTL 2.1 and 3.0.

Nevertheless, as business scales and financing structures optimize, CoreWeave has begun securing cheaper funding.

In March 2026, the company secured an $8.5 billion DDTL 4.0 credit facility, with some borrowings priced at U.S. Treasury yields plus 200 basis points. As of June 2026, the effective interest rate for related financing stood at approximately 7%.

The second category is equipment financing.

As of June 2026, CoreWeave's outstanding debt from OEM manufacturer and software licensing financing arrangements had reached $5.102 billion, accounting for about 14.4% of all interest-bearing debt.

This type of financing is relatively straightforward: Financial institutions provide funds for purchasing GPUs, servers, and other equipment, which CoreWeave repays in installments with interest.

For AI data centers requiring massive upfront capital investments, this model helps alleviate cash flow pressure from one-time equipment purchases.

However, equipment financing is also expensive.

According to Q2 reports, CoreWeave's recourse OEM and software licensing financing carried effective interest rates of approximately 11%, while non-recourse portions stood at around 9%.

The third category consists of senior unsecured notes and convertible bonds.

As of June 2026, CoreWeave's senior unsecured notes principal stood at approximately $10.029 billion, with convertible bond principal at $6.588 billion—totaling $16.617 billion and accounting for about 46.7% of all interest-bearing debt.

In April 2026, CoreWeave issued $2.75 billion in senior unsecured notes maturing in 2031 with a 9.75% coupon.

In June, the company issued $1.25 billion in U.S. dollar bonds maturing in 2032 with a 9.625% coupon, along with €2 billion in euro bonds maturing in 2032 with an 8.5% coupon.

In contrast, convertible bonds carry much lower coupons. In April, the company issued $4 billion in convertible bonds with a 1.75% coupon; in September, it completed another $4.2 billion convertible bond issuance with a 2.875% coupon.

The recent market concerns stem precisely from the senior unsecured bonds in this third financing category.

The reason is simple: Unlike secured bonds backed by specific assets, unsecured bonds lack collateral, leaving investors reliant solely on the company's overall solvency for protection. In terms of repayment priority, they also rank below secured creditors with claims on specific assets.

In other words, these creditors face greater exposure to CoreWeave's overall credit risk.

/ 03 / Exposing a Fragile Link in the AI Computing Power Supply Chain

CoreWeave's debt issues triggered such significant market reactions primarily because of its unique role in the AI industry chain.

It serves as both a computing power service provider and a crucial conduit for financial capital entering the AI hardware market.

Over the past two years, a typical capital flow path has emerged in the AI industry chain.

Microsoft, Meta, OpenAI, and other companies continuously increase their computing power demands, signing long-term orders with new cloud providers like CoreWeave.

After securing these orders, new cloud companies turn to banks, bond investors, and other financial institutions for financing. The raised funds purchase NVIDIA GPUs, servers, networking equipment, and finance data center construction.

Once computing power is delivered, clients gradually repay investments through service fees over several years.

Thus, companies like CoreWeave fulfill a critical function: converting future years' AI computing power demand into today's chip and data center orders.

The problem lies in the fact that these companies lack the massive mature business cash flows enjoyed by Microsoft, Google, or Amazon. Their substantial upfront investments rely entirely on external financing.

Currently, high debt levels have become a common characteristic among leading new cloud companies.

Besides CoreWeave, Nebius, IREN, and TeraWulf carry interest-bearing debt of $10.04 billion, $7.71 billion, and $5.725 billion respectively.

Combined, these four new cloud companies hold over $58.9 billion in interest-bearing debt.

Of course, compared to CoreWeave, Nebius and IREN maintain healthier financial positions. As of June 2026, Nebius and IREN held unrestricted cash reserves of $8.04 billion and $5.9 billion respectively, covering approximately 80% and 77% of their interest-bearing debt.

Nevertheless, CoreWeave's debt turmoil has still sparked widespread concerns across the new cloud sector.

The reason is straightforward: For new cloud companies heavily reliant on external financing, real risks may emerge long before actual debt defaults occur.

The first impact is on financing costs.

If bond yields continue rising, companies will face higher borrowing costs when issuing new debt in the future.

As financing becomes more expensive, companies must recalculate investment returns for new projects. Data centers that were profitable under lower interest rates may gradually become unviable, ultimately forcing delays or cancellations.

More troublingly, if stock prices keep declining, companies would need to issue more shares to raise the same amount of capital through equity offerings, causing greater dilution for existing shareholders.

Simultaneously, falling stock prices may reduce the attractiveness of convertible bonds, further driving up financing costs.

This could create a vicious cycle: Rising financing costs depress returns on new projects; slower expansion lowers growth expectations; further stock price declines make subsequent financing even more difficult.

Meanwhile, these impacts ripple through the entire AI industry chain.

NVIDIA isn't just a shareholder in CoreWeave—it's the exclusive hardware supplier for CoreWeave's core business. The billions of dollars CoreWeave has borrowed are essentially used to purchase NVIDIA chips. This overlapping structure of supplier and shareholder creates exceptionally deep financial entanglement between the two companies.

If CoreWeave struggles, its GPU hardware procurement demand will inevitably shrink, directly impacting NVIDIA's revenue. Simultaneously, NVIDIA's massive equity investment in CoreWeave would face impairment risks.

This dual exposure creates a classic game theory dilemma: The cost of a bailout is enormous, but the cost of inaction may prove equally untenable.

From this perspective, while AI computing power demand growth may be far from over, the capital supply supporting this expansion is already facing stricter pricing.

By Arch