On October 27, 2026, NVIDIA and Oracle announced plans to expand Oracle Cloud Infrastructure (OCI) AI superclusters to multi-gigawatt (GW) scale using Vera Rubin NVL72. As of the announcement, Vera Rubin NVL72 and Vera CPU are generally available on OCI. The companies did not disclose a contract value for the expansion. But using the "roughly $40 billion per GW" revenue opportunity that NVIDIA cited on its earnings call as a benchmark, a single gigawatt would equal more than 40% of Oracle's fiscal 2027 capital expenditure target.
Of course, that $40 billion does not translate directly into what Oracle will pay. Even so, expanding to multiple gigawatts could place a considerable burden on Oracle if it had to be funded from operating cash flow alone. Indeed, Oracle's latest earnings materials suggest the company is advancing its AI infrastructure investment by using customer prepayments, customer-supplied equipment, and equity issuance. The pace of future expansion will likely depend on how much external funding and customer-side equipment Oracle can secure.
Vera Rubin Now Generally Available, With Multi-Gigawatt Expansion Planned
The announcement distinguished between what is already available and what is planned. Generally available now are Vera Rubin NVL72 and Vera CPU on OCI. Vera Rubin NVL72 is a system that connects 72 GPUs in a single rack via NVLink. Still in the planning stage are the multi-gigawatt expansion of Vera Rubin NVL72 within OCI superclusters and the introduction of HGX Rubin NVL8, an x86 server configuration.
HGX Rubin NVL8 will pair eight Rubin GPUs with two 64-core Intel Xeon 6 processors, 4TB of DDR5 memory, and 6.4Tbps of network connectivity. Alongside the large-scale AI compute expansion with Vera Rubin NVL72, the companies plan to bring Rubin-generation GPUs to x86-based server environments as well.
One use case Vera CPU targets is AI agent processing. AI agents do not just run model inference; they frequently perform tasks handled by the CPU, such as calling external tools and executing code. When these tasks take a long time, GPUs sit idle in the meantime, lowering overall system efficiency. NVIDIA and Oracle say they aim to remove this bottleneck by boosting CPU-side processing power. Uday Ruddarraju, CTO of Compute at OpenAI, also said the company will deploy Vera CPU in production on OCI to support the fast-growing volume of sandboxes for AI agents.
The announcement did not disclose the contract value for the expansion or how it will be financed. Reuters also reported that the deal value was not disclosed and that Oracle has not explained specifically how it will raise the funds. On the same day, Oracle also said it will offer Google's TPUs for the first time. According to Reuters, it also plans to offer inference-focused AI chips from SambaNova and Cerebras. While expanding its cooperation with NVIDIA, Oracle is also diversifying the suppliers of the semiconductors used in its AI compute infrastructure.
$40 Billion per GW Equals More Than 40% of Oracle's Annual Capex Target
A useful reference for gauging the scale of the expansion is the figure NVIDIA presented on its fiscal 2027 second-quarter earnings call on August 26, 2026. The company said the revenue opportunity per GW of AI data center power capacity was about $18 billion for the Hopper generation and about $25 billion for Blackwell, and has grown to about $40 billion for the Vera Rubin generation.
The $40 billion for the Vera Rubin generation includes Vera CPU, Rubin GPU, NVLink, InfiniBand or Ethernet networking products, Groq LPU, and more. NVIDIA explains that this growth in revenue opportunity reflects a rising share of its own products in total AI data center investment. The key point is that the $40 billion figure represents the revenue NVIDIA expects to capture, not the total construction cost of a data center or the total amount customers pay.
Comparing this figure with Oracle's capital expenditure plans shows how large the project is. According to Reuters, Oracle has maintained its fiscal 2027 capex target at roughly $90 billion to $95 billion. NVIDIA's $40 billion per GW equals about 42% to 44% of that annual target: $40 billion divided by $95 billion is 42.1%, and divided by $90 billion is 44.4%.
If "multi-gigawatt" is interpreted as 2GW or more, NVIDIA's revenue opportunity on the same basis would be $80 billion or more, reaching about 84% to 89% or more of Oracle's annual capex target. However, this is a comparison of scale using figures the two companies have published, and it does not mean Oracle will actually pay $80 billion or more. Because the deployment timeline and contract terms have not been disclosed, the funds needed for the expansion will not necessarily be concentrated in the single year of fiscal 2027.
So how is Oracle funding its massive capital spending? According to its fiscal 2027 first-quarter earnings materials, released on September 10, total capital expenditure for the quarter was $28 billion. Net cash outlay, after subtracting short-term financing and customer prepayments with a financing element, was only $18 billion. The $10 billion difference includes short-term financing as well as customer prepayments, so it cannot be attributed entirely to customer funds.
The same materials also state that most of the increase in remaining performance obligations (RPO) was tied to customer prepayments or customer-supplied hardware. RPO is a measure of contracted amounts not yet recognized as revenue, and the balance at the end of the quarter reached $664 billion. In the first quarter, Oracle also completed $20 billion in equity issuance through an at-the-market (ATM) program. Reuters also reported that Oracle explained in its September earnings commentary that customers are increasingly bearing the cost of AI chips themselves.
These figures show that Oracle is not advancing its capital spending using only cash generated from its own operations. It appears to be supporting its large-scale AI infrastructure expansion through a combination of customer prepayments, use of customer-owned equipment, and equity issuance. However, customer-supplied equipment differs from funds Oracle raises itself, and should be distinguished as a mechanism that reduces the capital expenditure Oracle itself must bear. This division of financing and equipment burden is likely to matter for the multi-gigawatt Vera Rubin expansion as well.
Actual Payments Unknown: How to Assess Reliance on External Funding
This view comes with important caveats. First, the $40 billion per GW that NVIDIA cited represents its own revenue opportunity, not Oracle's actual payment. The amount Oracle bears will vary depending on contract pricing, equipment ownership structure, and whether customers bring their own equipment. In addition, because the deployment period has not been disclosed, there are limits to comparing a multi-year investment with a single-year capex target.
Another counterargument is that Oracle can secure sufficient funds through equity and bond issuance. Indeed, in addition to the $20 billion equity issuance mentioned above, Oracle was reported to have laid out a plan in June to raise about $40 billion through bonds and equity in fiscal 2027. A large-scale expansion therefore does not immediately imply a funding shortfall, nor can it be said that Oracle depends only on customer funds. What matters for this analysis is that means other than Oracle's own operating cash flow, including customer prepayments, shared equipment burden, and capital market financing, appear to be supporting the expansion plan.
Still, the view that external financing and customer-borne equipment will be important is grounded in Oracle's own earnings materials, because the company ties most of the first-quarter RPO increase to customer prepayments and customer-supplied equipment. This fact holds regardless of whether NVIDIA's $40 billion figure can be treated as Oracle's payment. However, it has not been established that the first-quarter contract mix applies unchanged to this Vera Rubin expansion.
An opportunity to test this view is Oracle's Investor Day, scheduled for October 28. If the company explains there that it can fund the Vera Rubin expansion from operating cash flow without relying on equity or bond issuance or customer prepayments, the view that it depends heavily on external financing would need to be revised. Another point to check is how the breakdown of the RPO increase is described in the next earnings materials. If the explanation that customer prepayments and customer-supplied equipment account for most of the increase disappears, the contract mix may have changed. However, the mere disappearance of the language would not prove reduced dependence, so any judgment should also consider actual financing amounts and the breakdown of capital expenditure.
For Japanese companies that use OCI, it also matters when the Vera Rubin-based compute infrastructure will become available in regions including Japan. The announcement covered the start of general availability and the large-scale expansion plan, but specific availability timing by region was not disclosed. Going forward, attention will focus not only on Oracle's financing and capital spending progress but also on how far the scale of usable compute resources and the range of regions where they are offered will expand.