Upfront Capital: CoreWeave Turns to Customer Prepayments and Shorter Contracts to Fund AI Growth

AI cloud provider CoreWeave is overhauling how it pays for its rapid infrastructure expansion. During its third-quarter earnings call, the company announced plans to rely heavily on upfront customer prepayments, contract-backed loans, and shorter rental agreements to generate cash while avoiding expensive corporate debt markets. By adjusting how it finances data center construction, CoreWeave aims to squeeze more revenue out of existing computing capacity while keeping pace with surging AI compute demand.

CoreWeave reported strong top-line growth for the third quarter, bringing in $3.66 billion in revenue, representing a 168 percent increase compared to the same period last year. However, massive infrastructure buildouts come with steep upfront financing costs. The company posted a $709 million net loss alongside $846 million in quarterly interest expenses. Chief Executive Officer Michael Intrator addressed those borrowing costs directly on Tuesday, stating that CoreWeave refuses to issue new unsecured corporate debt at current high market yields.

Instead of tapping expensive bond markets, CoreWeave plans to fund construction using loans tied directly to signed customer contracts. This targeted borrowing model allows the company to draw down capital in installments only when needed to build out its massive $132.5 billion contract backlog. Alongside contract-backed debt, CoreWeave expects to collect roughly $9 billion in customer prepayments and deposits over the next three months. More than $3 billion of those upfront funds come from new agreements for computing infrastructure that CoreWeave already built, meaning the company generates immediate cash without spending extra capital on new hardware.

This shift reflects a broader trend across the specialized AI cloud sector. Competitors like Nebius and IREN recently secured major prepayments from Microsoft to finance their own data center expansions. With computing capacity running tight globally, major tech companies are increasingly willing to pay cash upfront to lock in guaranteed access to high-end processors. While upfront customer cash reduces CoreWeave’s immediate need to borrow money, it also locks the company into fixed pricing schedules that could limit upside if market rental rates continue rising.

To capture rising market rates, CoreWeave is actively steering available computing capacity toward shorter-term contracts. The company raised prices across its service offerings by about 25 percent in July and added another 10 percent price hike later in the quarter. Intrator noted that prices continue climbing in the current quarter, easily offsetting hardware and energy cost increases. Short-term AI compute capacity currently sells for more than $40 million per megawatt annually, nearly double the $20 million to $25 million per megawatt rate commanded by standard four- and five-year contracts.

Supporting these shorter agreements required creating dedicated financing tools. In August, CoreWeave closed a $2.6 billion loan facility named DDTL 5.5, structured specifically around short-term customer contracts. The facility lets CoreWeave draw funds incrementally as short-term capacity comes online. However, shifting toward shorter deals carries clear trade-offs. Traditional lenders strongly prefer multi-year contracts because predictable cash flows lower default risks, whereas short-term deals expose CoreWeave to renewal uncertainty if AI demand cools.

While equity investors reacted positively to CoreWeave’s revenue growth, bond markets remain cautious. Yields on several CoreWeave corporate bonds climbed above 15 percent last month, signaling concern among debt investors over the massive capital required to build data centers before those facilities generate steady returns.

Finding creative ways to fund expansion without taking on high-yield unsecured debt buys CoreWeave critical operational flexibility. Yet relying on customer prepayments and premium short-term pricing leaves the company heavily dependent on sustained, uninterrupted demand for AI computing power.