The cloud computing landscape is witnessing a strategic split over contract durations. Newer cloud providers, such as Nebius and CoreWeave, are pitching the logic of high premiums from short-term contracts to investors, while the industry giant Amazon.com's AWS is doubling down on long-term agreements to secure future revenue. Each path has its own business rationale, but which one will ultimately yield better returns remains an open question.
On August 16, a tech media outlet reported that the previous week saw Nebius and CoreWeave pitch short-term contract businesses to investors, arguing that the tight supply of GPU computing power and surging AI demand allow for significantly higher pricing premiums. The stock prices of both companies subsequently rose sharply.
Simultaneously, AWS CEO Andy Jassy emphasized during an earnings call that the majority of AWS's current AI computing capacity is sold via five-year contracts. This strategy is intended to ensure that investments in data center construction are covered by sufficient revenue. Amazon.com shares have risen over 11% cumulatively since its earnings call on July 30.
The core of this divergence lies in the trade-off: short-term contracts capture high-price benefits from immediate demand but sacrifice future revenue predictability. Long-term contracts, in contrast, secure a stable revenue stream at lower prices, offering investors greater certainty in an era of potential AI investment bubbles. Paul Meeks, Managing Director and Technology Research Head at Freedom Capital Markets, noted, "You need a balance between long-term, stable, high-grade customer contracts and the on-demand market, where you can get opportunistic pricing. It's like having both a tortoise and a hare."
The New Logic Behind High Premiums on Short-Term Contracts
Nebius and CoreWeave both state that short-term contracts command prices far exceeding those of medium-to-long-term deals. Nebius executives disclosed that new contracts for three to six months have an annualized pricing of $40 million to $50 million per megawatt, more than double the annualized price of its one-to-three-year contracts. Marc Boroditsky, Nebius Chief Revenue Officer, stated in a written comment, "We are building a flexible go-to-market process to quickly respond to customer needs. Short-term large-scale training clusters are a real demand for which model developers are willing to pay a premium."
CoreWeave is focusing on the rapid growth of its hosted inference business, which allows clients to pay per use to run models without signing long-term contracts. Jon Jones, CoreWeave Chief Revenue Officer, noted in a statement, "As our customer base expands towards AI application developers and enterprise clients, we find customers increasingly seeking flexibility in how they plan and scale their services." Earlier this month, CoreWeave completed a $2.6 billion, five-year term loan secured against approximately three-year customer contracts. The company stated this financing demonstrates growing lender acceptance for debt backed by short-term, high-price contracts.
Contract structures have subtly shifted, but the change is limited. Reports indicate that a key metric for future cloud revenue, "Remaining Performance Obligations" (RPO), or committed but unrecognized revenue, showed a shift towards near-term obligations at both companies in the second quarter. Nebius disclosed total RPO of $37.5 billion as of June 30, with the portion due within two years rising to 36% from 29% in the previous quarter, while the portion due beyond four years fell to 24% from 32%. CoreWeave's total RPO was $103.7 billion, with the two-year portion rising to 41% from 36% and the four-year-plus portion dropping to 20% from 25%.
However, the shift towards short-term contracts remains limited for both companies. CoreWeave's CFO Nitin Agrawal told analysts that capacity available for renewal is "extremely limited," constraining the company's ability to benefit from the current high-price environment. Nebius stated it allocates only a small fraction of available computing power to the shortest, multi-month contracts, with its primary strategy being one-to-three-year medium-term deals. Nebius executives also cited the results of the company's first auction for unallocated computing power as evidence, stating that the winning bid for Nvidia Blackwell chip capacity was 15% higher than Nebius's previous highest charge, indicating broader pricing upside.
AWS's Contrarian Logic: Locking in the AI Era with Long-Term Contracts
AWS is approaching from the opposite direction. The cloud giant, with a mature on-demand business, has gradually introduced mechanisms for reserving AI capacity and launched "capacity blocks" in 2023, allowing clients to reserve a fixed amount of GPU capacity. The weighted average remaining duration of AWS's long-term contracts jumped to 6.4 years in the second quarter, up from 5.5 years in the previous quarter and well above the recent average of about four years. This jump primarily reflects mega-deals signed earlier this year, including an eight-year, $100 billion commitment from OpenAI and a ten-year, $100 billion commitment from Anthropic.
Amazon.com acknowledges that these large-scale long-term contracts, where clients get lower prices for longer lock-in periods, have weighed on AWS's cloud revenue growth in recent years. However, as the company invests heavily in AI expansion, these contracts are a positive signal for investors. Even in a hypothetical AI bubble burst, AWS has secured a substantial amount of committed revenue. Dave McCarthy, IDC Research Vice President for Cloud and Data Center Infrastructure, stated, "AWS is trying to prove to investors that if you want to bet on AI, you should choose the one with the largest backlog. It reflects customer confidence in their business."
Analysts point out that for emerging cloud providers, it remains uncertain whether they can maintain high premiums after short-term contracts expire. As various cloud providers aggressively expand capacity, increasing market supply, Nebius and CoreWeave may face pricing pressure upon renewal, while giants like AWS could still rely on their large portfolio of long-term contracts. For AWS, over-reliance on long-term contracts could mean missing out on high-premium short-term opportunities and potentially being at a disadvantage when competing for fast-moving AI startup clients.
Notably, a significant portion of the current revenue pipelines for Nebius and CoreWeave still comes from contracts of at least five years with investment-grade clients like Meta Platforms and Microsoft. These long-term contracts are crucial for securing GPU procurement financing and data center construction loans, as lenders require predictable revenue streams. How a continued increase in the share of short-term contracts will affect the future financing capabilities of both companies remains a key variable for the market to monitor.
Comments