NVIDIA's Data Center Revenue Hits $107.6 Billion: How Far Has Rubin Come on the Roadmap?
The competition for AI computing power has expanded from simply releasing new chips to delivering complete systems to customers.
On November 17, 2026, NVIDIA announced its financial results for the third quarter of fiscal year 2027, ending October 25: revenue was approximately $115.8 billion, up 103% year-over-year; Data center revenue totaled approximately $107.6 billion, up 110% year-over-year. The company also confirmed that the Vera Rubin NVL72 has begun shipping. The term “fiscal year 2027” here refers to NVIDIA’s fiscal year and does not mean that the results pertain to the calendar year 2027.NVIDIA's Official Financial Report
A Few Simple Calculations: How Much of the Total Do Data Centers Account For?
Based on the rounded figures in the announcement, 1,076 ÷ 1,158 ≈ 92.9%. In other words, for every $100 in revenue this quarter, approximately $93 came from the data center business.
This figure illustrates that understanding NVIDIA’s business requires looking beyond consumer-grade graphics cards. The more important questions are: Why are customers continuing to expand their data centers? How do new platforms integrate into existing data centers? And what value do these investments ultimately deliver?
However, this data set does not break out Rubin's revenue separately; therefore, we cannot attribute the entire quarter's growth to Rubin, nor can we use it to estimate sales for a specific model.
“Shipped” goes a step further than the roadmap, but deployment isn't yet complete.
NVIDIA's supply chain announcement further clarifies that the Vera Rubin NVL72 and Vera CPU will begin volume shipments in the third quarter. This represents actual progress in the manufacturing and delivery phases and differs from earlier product plans or future launch schedules.NVIDIA Supply Chain Announcement
For customers, even after receiving the equipment, issues such as system integration, workload adaptation, and stable operation remain. When evaluating an AI infrastructure, one should not focus solely on peak performance but also consider the speed, reliability, and total cost of the application under actual operating conditions.
| Progress Observed | What does this indicate? | What still needs to be monitored |
|---|---|---|
| Product Announcement | Clear Architecture and Product Direction | Delivery Schedule, Configuration, and Full-Scale Supply |
| Begin Mass Shipments | Manufacturing and delivery are already underway | Client-Side Deployment and Continuous Scaling |
| System Put Into Operation | Moving on to the actual usage phase | Stability, Utilization, and Workload Performance |
| Publish Performance Results | Performance under specific test conditions | Actual business costs under the same quality requirements |
What is the difference between "Available in the Cloud" and "Rack in Operation"?
NVIDIA says the cabinets are already in operation at CoreWeave, Google Cloud, Microsoft Azure, Oracle Cloud Infrastructure, and Nebius. This does not mean that all five companies offer computing power to the public in the same way.Oracle OCI has announced that the NVL72 and Vera CPUs are now generally available,Google Cloud This confirms that the A5X instance is officially available. When preparing to migrate to the cloud, you should also verify your account’s available regions, quotas, delivery times, and costs separately.
In the production expansion timeline, which items are still in the future?
The supply chain announcement covers areas such as substrate manufacturing, advanced packaging, and system manufacturing. Taking Kinsus as an example, its expansion plan for substrate production to support NVIDIA is expected to begin in the second half of 2027, with the target for the new capacity to come online set for the third quarter of 2028; SPIL’s expansion, meanwhile, is slated for mid-2029. Neither of these can be considered as supply capacity that has already been established today.Official Production Expansion Timeline
This also explains why news about a chip company frequently mentions its supply chain partners. From the completion of the design to large-scale delivery, coordination is required across various stages of the manufacturing process. An announcement that one stage will increase investment does not automatically guarantee that other stages will simultaneously meet their planned targets.
When reading announcements of this kind, it is recommended to label them separately as “Already Delivered,” “Under Construction,” and “Scheduled to Begin Production.” If timelines and statuses are mixed together, it is easy to overestimate the computing power currently available on the market.
The $130 billion for the next quarter is a guidance figure, not actual revenue.
NVIDIA has provided fourth-quarter revenue guidance of $130 billion, plus or minus 3 percent, which corresponds to approximately $126.1 billion to $133.9 billion. The announcement also notes that the quarter is one week longer than usual and that the guidance does not include data center computing revenue from China. These conditions should be considered in conjunction with the forecast figures.Fourth-Quarter Outlook in the Financial Report
Therefore, while subtracting the guidance directly from the previous quarter’s figures can illustrate the company’s expectations for revenue levels, it does not fully account for the difference in demand growth. The length of the quarter, product mix, and delivery schedule are all factors that warrant further verification against actual results.
For business users, the most useful metric is how much it costs to complete a single task.
If you're procuring AI services or evaluating in-house computing power, you can start by asking yourself these three questions.
- Are the workloads a good match? Large-model training, online inference, and batch document processing have different requirements for latency, throughput, and sustained resource consumption.
- Are the metrics consistent? Given the same model size, output length, and concurrency targets, comparing completion time and total cost is more meaningful than simply looking at hardware generations.
- Does the new equipment deliver tangible benefits? Migration, adaptation, operations and maintenance, and downtime should also be included in the total cost; it is not sufficient to calculate only the unit price of equipment or cloud instances.
The start of Rubin shipments signals that the next-generation infrastructure is entering a more concrete implementation phase. The key points to watch moving forward are whether it can steadily scale up supply and improve utilization efficiency under real-world workloads, and whether customers are willing to continue paying for these improvements.
© Copyright notes
The copyright of the article belongs to the author, please do not reprint without permission.
Related posts
No comments...