CoreWeave launches another $3bn convertible as $104bn backlog demands more capital

CoreWeave Inc. (NASDAQ: CRWV) is returning to capital markets with a proposed $3 billion convertible senior-note offering due in 2033, along with an option allowing initial purchasers to acquire another $500 million. The Nvidia-backed AI infrastructure provider simultaneously established an at-the-market equity programme covering as many as 35 million Class A shares, creating another potential source of billions of dollars as management attempts to finance extremely rapid expansion while working toward an investment-grade credit profile.

The announcement highlighted a central contradiction in the artificial-intelligence infrastructure boom: extraordinary customer demand can require extraordinary financing. CoreWeave ended the second quarter with approximately $104 billion of revenue backlog and said that figure excluded more than $25 billion of net new customer commitments added early in the third quarter, yet it continues raising debt and equity because delivering that backlog requires enormous upfront investment in GPUs, data centres, power and networking.

How much capital could CoreWeave raise through the new convertible and share programme?

The convertible offering has a base size of $3 billion and could reach $3.5 billion if initial purchasers exercise the full additional option. CoreWeave’s investor materials indicated an expected coupon range of 2.375% to 2.875%, an expected conversion premium of 22.5% to 27.5% and an April 1, 2033 maturity, although those figures were presented as expected terms during the offering process rather than final terms.

CoreWeave intends to use part of the note proceeds to fund capped-call transactions designed to mitigate potential dilution associated with conversion, with the remainder available for general corporate purposes. The structure gives the company financing at a lower coupon than its conventional high-yield debt because investors receive potential upside through conversion into CoreWeave shares.

The separate at-the-market programme covers up to 35 million Class A shares. Reuters estimated that the programme could have generated approximately $2.92 billion if all shares were sold at the September 16 closing price, although CoreWeave can choose whether and when to issue shares based on market conditions.

Taken together, the programmes give management considerable flexibility. They also expose existing shareholders to potential dilution and reinforce how heavily the company relies on external capital to build enough physical infrastructure to meet signed customer commitments.

Why does a company with $104 billion of backlog still need to raise billions?

Revenue backlog is not the same as cash available today. CoreWeave may have customers committed to purchasing computing capacity for years, but the company generally needs to install the underlying infrastructure before much of that contracted revenue can be recognised.

Second-quarter property and equipment reached $46.74 billion, up from $30.56 billion at the end of 2025. Total assets increased to $77.07 billion from $49.30 billion in just six months, illustrating the speed at which CoreWeave’s physical capital base is expanding.

The liabilities side has expanded just as rapidly. Current and long-term recourse debt totalled more than $31 billion at June 30, while non-recourse debt added another approximately $3.7 billion. CoreWeave also carried substantial lease obligations associated with data-centre infrastructure.

This business model effectively requires management to bridge the timing gap between capital expenditure and future customer payments. As long as lenders and equity investors believe the contracted customers will pay and AI computing remains scarce, CoreWeave can potentially finance rapid expansion. If demand weakens or capital markets become less receptive, that same leverage can become a material vulnerability.

Are CoreWeave’s underlying AI cloud economics improving quickly enough?

CoreWeave’s growth remains exceptional. Second-quarter revenue increased to $2.575 billion from $1.212 billion a year earlier, more than doubling, while adjusted EBITDA reached $1.51 billion. Adjusted EBITDA margin was 59%, although GAAP operating results remained negative and CoreWeave posted a $626 million quarterly net loss.

Interest expense is becoming increasingly important. Net interest expense reached $640 million in the second quarter compared with $267 million a year earlier, consuming a meaningful portion of the cash generated before financing costs.

That gap explains why adjusted EBITDA alone can give an incomplete picture of CoreWeave’s economics. AI infrastructure requires so much debt, leasing and capital expenditure that investors need to evaluate interest costs, depreciation, capital intensity and free cash requirements alongside topline growth.

The company’s operating leverage is nevertheless improving in some areas. Revenue more than doubled while adjusted EBITDA also roughly doubled, suggesting strong utilisation of installed computing resources. The question is whether CoreWeave can keep expanding capacity without financing costs absorbing too much of the economic value created by that growth.

What does 4.2 gigawatts of contracted power say about CoreWeave’s ambitions?

CoreWeave said contracted power had increased to about 4.2 gigawatts by August 11 from approximately 3.7GW at the end of June. That scale illustrates how AI cloud competition increasingly depends on access to electricity rather than servers alone.

Power reservations provide a pipeline for future data-centre capacity, but they also imply substantial construction commitments. Each megawatt can require servers, networking equipment, cooling, substations, backup generation and building infrastructure before it becomes commercially productive.

CoreWeave said it had signed short-term third-quarter contracts for computing capacity at approximately $40 million per megawatt on an annualised basis. That pricing indicates extraordinary current scarcity, although short-term rates should not automatically be applied across its entire power pipeline or assumed to persist indefinitely.

The comparison nevertheless helps explain why investors are willing to finance expansion. When productive AI infrastructure can command very high revenue per megawatt, the economic incentive to install additional capacity remains strong even when construction costs and financing requirements are enormous.

Why is CoreWeave trying to reach investment-grade credit quality?

Lower borrowing costs could materially change the long-term economics of the business. CoreWeave’s existing unsecured bonds include coupons ranging from 8.5% to 9.75%, while the expected coupon on the new convertible is much lower because investors also receive conversion rights.

An investment-grade credit profile could eventually allow CoreWeave to finance infrastructure more cheaply and tap a larger universe of institutional fixed-income investors. For a company whose business requires tens of billions of dollars of physical assets, a relatively small reduction in funding costs can become financially significant.

CoreWeave has already raised more than $10 billion of unsecured debt and convertible bonds during its recent expansion, while also using term loans and strategic investments.

The at-the-market equity programme may support that objective by increasing equity capital and reducing some reliance on pure debt. Existing shareholders, however, pay for that balance-sheet strengthening through dilution if large numbers of new shares are ultimately issued.

Why did CoreWeave shares fall while technology stocks were rising?

CoreWeave shares fell roughly 4% to 5% during September 17 trading, even as the Nasdaq Composite advanced around 1.7%. The divergence indicates that investors were focused specifically on the new financing and dilution risk rather than simply following broader enthusiasm for artificial-intelligence shares.

The stock had still gained more than 16% for 2026 before the announcement, showing that investors continue to assign substantial value to CoreWeave’s demand growth.

The investment debate is consequently becoming more nuanced than whether AI computing demand is strong. CoreWeave has already demonstrated extraordinary demand through its backlog and customer commitments. The unanswered question is how much debt and equity capital must be raised before that demand produces sustainable returns for common shareholders.

The next milestones will be final convertible-note pricing, use of the at-the-market programme, capital expenditure, contracted-power conversion and whether interest expense grows more slowly than revenue.

CoreWeave’s problem is one many companies would like to have: more demand than existing infrastructure can serve. Its challenge is proving that financing the solution does not consume too much of the value created by satisfying that demand.