Verdreifachung in drei Monaten. Hinter der Milliardensteigerung von Oracle hat sich die Landschaft des KI-Clouddienstes verändert.
Last night, Oracle's stock price soared 41% on Wednesday, marking its largest intraday gain since 1992.
The wealth of its co-founder, Larry Ellison, skyrocketed by $101 billion in a single day, setting a record for the largest single-day wealth increase in history. He overtook Elon Musk to become the world's richest person.
The cause of Oracle's stock price surge is the sharp increase in its cloud computing business. The company disclosed that as of August, its remaining performance obligations (RPO) reached $455 billion, not only three times that of three months ago but also far exceeding analysts' estimates of $180 billion.
The company also expects its cloud infrastructure revenue to reach $18 billion in fiscal year 2026 and accelerate to $144 billion in the next four years.
This surge is not accidental. The real logic behind it is that under the leadership of NVIDIA, the cloud computing industry is undergoing a complete reshuffle. Companies that can secure GPUs in advance have become the "hard currency" in the eyes of investors.
These emerging players have a name - Neocloud. Neocloud refers to GPU-only clouds: they do not engage in general computing but focus solely on high-performance GPU clouds, mainly serving computing-intensive scenarios such as AI model training, inference, and rendering.
Excluding Oracle's recent increase, the representative Neocloud companies CoreWeave and Nebius have risen 192.55% and 237.15% respectively this year.
So, why do investors have high hopes for these Neocloud companies?
01
Tripled in 3 months, with a contract scale higher than Microsoft's!
Ultimately, it is AI computing power that supports Oracle's sharp increase in performance.
The most astonishing data is the terrifying growth of RPO. As of the end of August, the company's remaining performance obligations (RPO) reached $455 billion. You know, at the end of May this year, this figure was only $138 billion.
In other words, Oracle tripled its on - hand cloud business orders in just three months.
RPO can be understood as "money that has been sold but not yet recorded as revenue". Whether it can turn into revenue depends on data center construction, computing power delivery, and energy consumption allocation. And Oracle's current RPO has even exceeded that of its long - standing rival, Microsoft - Microsoft's latest commercial RPO is $368 billion.
The company explained that this mainly comes from four super - large contracts signed with three customers in the first quarter, and it is expected that there will be more similar large orders in the next few months. RPO is likely to exceed $500 billion.
The explosion of orders has directly boosted performance expectations. The management's guidance is that Oracle's cloud infrastructure revenue will grow 77% this fiscal year, reaching $18 billion; in the next four years, it will climb to $32 billion, $73 billion, $114 billion, and $144 billion respectively.
Even more astonishing is that almost all of this is due to OpenAI. According to people familiar with the matter, Oracle signed a $300 - billion, 5 - year super contract with OpenAI, accounting for 94.6% of the newly added RPO ($317 billion) this quarter.
This cooperation is closely related to the "Stargate" project promoted by Masayoshi Son and Sam Altman.
At the beginning of this year, when Masayoshi Son and Sam Altman announced an investment of $500 billion at the White House, Larry Ellison was the key partner standing on the stage. According to the disclosure, this contract, which will be implemented starting in 2027, involves a data center capacity of up to 4.5 gigawatts, equivalent to the electricity consumption of 4 million households.
The risks are also huge. OpenAI's annual revenue in 2024 was approximately $3.7 billion, and the annualized revenue disclosed in June this year was only $10 billion. But starting from 2027, they will have to pay Oracle $60 billion per year for 5 years.
For OpenAI, this is a huge gamble; for Oracle, it is a super bet that locks in future revenue in advance.
02
GPU stirs up a reshuffle in cloud services
Oracle's recent surge is not accidental but driven by a larger trend: in the AI era, those who have NVIDIA's support will dominate the market.
In this computing power distribution game centered around NVIDIA, the landscape of cloud services is being reshuffled.
In the past decade, AWS (Amazon Web Services), Azure (Microsoft Cloud Services), and GCP (Google Cloud Services) have been the stable "Big Three". However, after the explosion of generative AI, computing power has become a scarce resource, and the situation has changed overnight.
According to Coatue's data, in 2025, the Big Three accounted for up to 93% of cloud revenue share. However, in the procurement share of NVIDIA GPUs, they only took 70%. Instead, new players like Oracle and CoreWeave, with a revenue share of less than 7%, bought 30% of the GPUs.
That is to say, smaller manufacturers actually have an advantage over the giants in terms of GPU resources. Under this misalignment, a group of new cloud providers are rising rapidly, and they are called Neocloud.
The representative companies are CoreWeave and Nebius. After falling behind in the general IaaS competition, Oracle also chose to "go all - in on GPUs". In the narrative of AI clouds, its positioning is becoming more and more similar to that of Neocloud.
The capital market is extremely enthusiastic about this track. In addition to Oracle, CoreWeave has risen 192.55% this year; after Nebius recently signed a $19.4 - billion super contract with Microsoft, its stock price soared 70%. As of now, its cumulative increase this year has reached 237.15%.
Investors are voting with real money, but what they are actually betting on is not the current revenue but the future right to occupy GPU production capacity.
Why does this misalignment occur? It's simple: the giants are buying NVIDIA GPUs while also developing their own chips - AWS has Trainium, Google has TPU, and Microsoft has Maia.
For NVIDIA, these customers may "change their minds" at any time, so it is more willing to allocate its limited production capacity to "loyal customers" who only buy its products and do not engage in self - research.
It goes without saying that NVIDIA has a direct stake in CoreWeave. CoreWeave has always been the leading Neocloud company strongly supported by NVIDIA and has been named the first deployer of the Blackwell/Hopper series multiple times.
Oracle is also not far behind. In 2024, it placed a $40 - billion order for Blackwell - class GPUs with NVIDIA at once. This is the largest GPU procurement in the industry to date, directly locking in the priority production schedule for the next few years.
Behind this reshuffle of the cloud service landscape, emerging cloud players and traditional cloud providers are taking two completely different paths.
In terms of business models, traditional cloud providers regard GPUs as "raw materials" and first process them into AI services such as Copilot, Bedrock, and Vertex. What customers ultimately pay for are actually API and SaaS functions, not bare GPUs.
This means that after the GPUs are invested, the revenue will be gradually released only after the services are up and running.
On the other hand, Neocloud and Oracle sell GPUs as "finished products" directly. As soon as the GPUs are put on the market, they can be monetized immediately. Customers can rent as many GPU hours as they need.
More importantly, most of their customers are super - large customers like OpenAI, xAI, and ByteDance. They sign contracts for an entire GPU cluster for 5 to 10 years. A single such order is often equivalent to the new orders of a traditional cloud provider in a quarter.
Coupled with their relatively low revenue base, each large order can show an "explosive growth" in the financial statements, sending a strong signal of revenue surge to investors.
Under this model, the expansion logic is also completely different from that of traditional clouds.
However, in the AI era, it is completely different. GPUs themselves have become the most valuable assets, as scarce as oil. NVIDIA's supply logic is also very clear: "Pay first, then deliver the goods". Those who can pay the money first can lock in the production capacity in advance.
For Neocloud companies like CoreWeave and Nebius, whose scale is much smaller than that of AWS and Azure, if they want to avoid being squeezed out of the production schedule, they must lock in the production capacity faster and more aggressively than the traditional giants.
So, Wall Street quickly developed a new approach for them - GPU - backed debt.
In this model, Neocloud first signs a large order with NVIDIA, then uses the GPUs for sale - leaseback or packages them into ABS, sells the future rental cash flow to financial institutions to get the money in advance, and then buys more GPUs, forming a cycle of 'GPU → financialization → buy more GPUs'.
In 2024, CoreWeave obtained billions of dollars in debt financing from Goldman Sachs and Blackstone in the form of GPU - backed debt. They first signed a large order with NVIDIA, then packaged the future GPU rental cash flow and sold it to financial institutions, and used the money to continue to increase their investment.
Nebius is also taking the same path. After winning the super - large contract with Microsoft, the company mentioned that it expects to use the cash flow generated by the contract with Microsoft and the newly issued contract - backed bonds to pay for the relevant capital expenditures.
Therefore, neither Oracle's surge nor the sharp increase in the stock prices of CoreWeave and Nebius this year is accidental. The real signal behind this is the reshuffle of the cloud computing landscape led by NVIDIA.
In this game, computing power is the chip. Those who can grab more GPUs from NVIDIA will gain the right to speak in the next AI industry cycle.
Both the "veteran" Oracle and the "up - and - coming" Neocloud companies have seized this trend and managed to ride the wave of the times.
This article is from the WeChat public account "Crow Intelligence Talk", author: Lin Bai. Republished by 36Kr with permission.