Recently, up-and-coming cloud service providers Nebius and CoreWeave have both been actively pitching short-term business agreements to investors. These two cloud firms have pointed out that, in light of the current scarcity of GPUs, these short-term deals will allow them to capitalize on the skyrocketing prices of AI computing power.

They view GPU computing power as a limited resource with prices poised to climb, and they aim to resell it at a premium through short-term contracts. Nebius and CoreWeave have both indicated that, given the ongoing GPU shortage, short-term contracts spanning three to six months can generate annual revenues of $40 million to $50 million per megawatt.

This figure is approximately more than twice what they earn annually from contracts lasting one to three years.

As a result of this strategy, the stock prices of both companies have experienced a significant uptick. However, there's more to the story, lurking in the fine print of their financial statements. The real backbone of their financial health lies in long-term credit agreements with five-year terms, secured with clients such as Microsoft and Meta.

CoreWeave has placed a strong emphasis on managed inference services, allowing clients to utilize models on demand without being locked into long-term commitments. Jon Jones, CoreWeave's Chief Revenue Officer, noted that AI application developers are increasingly valuing flexibility.

Marc Boroditsky, Chief Revenue Officer at Nebius, mentioned that model developers are prepared to pay a premium for short-term, large-scale training clusters, prompting the company to establish a swift response system.

Earlier, Nebius conducted its first auction of unallocated computing resources, with the winning bidder paying 15% more for Nvidia Blackwell chips than Nebius's previous highest fee.

At the start of this month, CoreWeave secured a $2.6 billion term loan over five years, using client contracts of roughly three years as collateral. This indicates that even lenders now acknowledge the worth of such short-term obligations.

An indicator used to gauge the certainty of future revenue is known as Remaining Performance Obligations (RPO), which represents the portion of signed contracts for which revenue has yet to be recognized.

By the end of June, Nebius reported a total RPO of $37.5 billion, with the share due within two years increasing from 29% to 36%, while the portion due beyond four years decreased from 32% to 24%. CoreWeave's RPO was even higher, at $103.7 billion, with the near-term portion (due within two years) rising from 36% to 41%, and the long-term portion (due beyond four years) declining from 25% to 20%.

Nevertheless, the scale of this shift remains relatively modest. Nitin Agrawal, CoreWeave's Chief Financial Officer, noted that "the current capacity available for renewal is extremely limited," leaving little room to fully exploit the prevailing high-price trend. Nebius also admitted that contracts as brief as a few months represent only a small fraction of its available computing capacity.

The pressure to expand capacity is also palpable. Nebius invested roughly $5.657 billion in capital expenditures during the second quarter, with full-year estimates reaching $23.4 billion, and it raised its signed capacity target from 4GW to 5GW. CoreWeave recently implemented a 25% price hike in July, citing the current demand landscape. Expanding capacity is a prerequisite for leveraging short-term contracts to generate advance cash flow.

While these new entrants are touting the benefits of short-term contracts, AWS is pursuing a diametrically opposite strategy. Andy Jassy highlighted during the earnings call that the majority of AWS's AI computing power is secured under five-year long-term contracts, ensuring that revenue covers the investments in data center construction.

The weighted average remaining term of AWS's long-term contracts extended to 6.4 years in the second quarter, up from 5.5 years in the previous quarter. This surge was primarily attributed to two particularly large deals: an eight-year, $100 billion commitment with OpenAI and a ten-year, $100 billion order with Anthropic. Since the earnings call on July 30, Amazon's stock price has cumulatively surged by over 11%.

Dave McCarthy, Vice President at IDC, remarked, "If you're betting on AI, go with the company that has the largest backlog." This statement aptly captures the rationale behind long-term contracts.

Stock prices for companies adopting both strategies are on the rise. Short-term contracts represent a gamble on current scarcity, whereas long-term contracts are a wager on future stability. CoreWeave itself anticipates a significant increase in supply around 2029, at which point the "demand premium" may diminish.

Once the expanded capacity becomes operational after 2027, it remains to be seen whether the high premiums secured today will still be attainable three years down the line.