The $500 billion revenue forecast is too conservative! Nvidia CFO says "it will definitely be higher," Jensen Huang says demand from Chinese customers is strong

Nvidia executives have repeatedly sent out optimistic signals, expressing confidence in the sales prospects of Blackwell and the next-generation Rubin chips.

On Tuesday, the 6th Eastern US time, Nvidia CFO Colette Kress stated at a JPMorgan-hosted event that, due to strong demand, Nvidia is now even more optimistic about its data center business. By the end of 2026, Nvidia's data center chip revenue is "definitely" expected to exceed the $500 billion forecast given last October.

The $500 billion Kress mentioned refers to Jensen Huang's statement at the GTC conference more than two months ago that by the end of 2026, Nvidia's existing and future data center chips will bring in about $500 billion in revenue. At the end of October, Goldman Sachs pointed out that this forecast was about 12% higher than the market consensus at the time.

Kress's remarks continued the optimistic mood set by Nvidia CEO Jensen Huang at the Consumer Electronics Show (CES) on Monday. Nvidia officially announced the launch of the Rubin chip on Monday. In his keynote speech at CES, Jensen Huang revealed that the new generation Vera Rubin platform has been fully put into production, reducing inference costs to one-tenth of the Blackwell platform.

Jensen Huang also emphasized in a media interview on Tuesday that Nvidia's new chips perform 10 times better than the previous generation, and said that demand from Chinese customers is strong.

Despite the positive signals from both Kress and Jensen Huang, Nvidia's stock price has not reversed its downward trend. After rising more than 2% in early trading on Tuesday, Nvidia shares continued to fall, turning slightly negative at midday and closing down nearly 0.5%, marking two consecutive days of decline. In 2025, Nvidia has risen about 39% so far. Although it remains a main driver of US stock market gains, the increase is far less than the more than 170% surge in 2024, mainly due to concerns about an AI bubble and competition from Google TPU.

Revenue Expectations Continue to Rise

In her speech on Tuesday, Kress made it clear that since the $500 billion forecast was given in October 2025, customer interest in Nvidia has continued to increase. Speaking on data center chip revenue expectations, she said, "The $500 billion (forecast) number will definitely be even higher."

Kress noted that Nvidia's optimistic outlook on AI applications is not only driven by AI demand itself, but also by enterprise data processing needs, which are promoting the growth of next-generation computing demand and will help total investment reach trillions of dollars by the end of 2030.

This statement echoes Jensen Huang's forecast at the GTC conference on October 28, 2025. At that time, Jensen Huang revealed that the company had "visibility" to achieve a cumulative $500 billion in data center business revenue between 2025 and 2026, a forecast that includes Blackwell and next-generation Rubin architecture products. According to Goldman Sachs, this target is 12% higher than the $447 billion consensus reflected by Visible Alpha Consensus Data and 10% higher than Goldman Sachs's own forecast of $453 billion.

In early December 2025, Kress said that the $500 billion in bookings did not include any work Nvidia was doing for OpenAI's next phase agreement, and revealed that the previously announced $100 billion investment agreement with OpenAI had not yet been finalized. She stated at the time that the company "definitely has the opportunity to receive more orders on top of the announced $500 billion."

At Tuesday's event, Kress was asked about the potential role of the Chinese market in Nvidia's revenue. Kress stated that the US government is working to approve license applications, Nvidia has received orders from customers, but it is still unclear what the specific outcome will be.

Rubin Platform Fully in Production

On Monday at the Las Vegas CES show, Jensen Huang announced that the new generation Vera Rubin AI platform is now fully in production. He stated that, through the integrated design of six new chips, the platform has achieved significant leaps in inference cost and training efficiency, with the first batch to be delivered to customers in the second half of 2026.

Jensen Huang described the Rubin GPU as "a huge beast" and elaborated on its design logic: "The cost of AI inference must drop by 10x every year, while the number of tokens generated by AI 'thinking' grows by 5x every year." He emphasized that this leap in performance, which exceeds Moore's Law expectations, comes from "extreme co-design"—a complete restructuring from CPU, GPU, networking chips to cooling systems.

According to the introduction, the Rubin GPU achieves 50 PFLOPS inference performance at NVFP4 precision, which is 5 times that of Blackwell; training performance reaches 35 PFLOPS, 3.5 times higher than the previous generation. The cost of generating inference tokens can be reduced to as little as one-tenth of the Blackwell platform. Each GPU is packaged with eight groups of HBM4 memory, with a bandwidth of up to 22 TB/s. Microsoft's next-generation AI super factory will deploy hundreds of thousands of Vera Rubin chips.

To address the AI "memory" bottleneck, Nvidia has built an inference context memory storage platform based on BlueField-4 DPU, adding an extra 16TB of high-speed shared memory to each GPU on top of the existing 1TB, connected by 200Gb/s bandwidth, thus solving the long-text "VRAM wall" problem.

In terms of energy efficiency, the Rubin NVL72 rack achieves 100% liquid cooling and supports an inlet water temperature of 45 degrees Celsius, meaning that data centers can dissipate heat without the need for high-energy-consuming chillers. Jensen Huang stated that this will save 6% of electricity for data centers worldwide.

Jensen Huang’s Speech Drives Storage Stocks Higher, Data Center Cooling Stocks Plunge

Jensen Huang's speech at CES had a significant impact on the stock prices of related industries. In his presentation, he emphasized the need for memory and storage in AI systems, stating, "This is a market that never existed before and will likely become the largest storage market in the world, essentially carrying the working memory of global AI."

Driven by these remarks, storage chip giant SanDisk became the best-performing S&P 500 component stock on Tuesday, with its share price closing up nearly 27.6%, the largest single-day gain since February 18, 2025.

SanDisk has risen more than 40% in the first three trading days of this year, and has soared about 1050% from the low in April last year. Storage device manufacturers Western Digital and Seagate Technology also recorded double-digit increases on Tuesday.

On the other hand, data center cooling system manufacturers saw their stock prices plunge. Johnson Controls' stock fell as much as 11% on Tuesday, closing down more than 6.2%. Modine Manufacturing dropped 21% intraday, before recovering more than half the loss to close down nearly 7.5%, while Trane Technologies and Carrier Global closed down 2.5% and 0.5%, respectively.

The decline in these companies' stock prices was triggered by Jensen Huang's claim that Rubin chips can be cooled with warm water, eliminating the need for chillers. At CES, Jensen Huang stated that server racks equipped with the new Rubin chips can be cooled at temperatures that do not require chillers, with airflow requirements basically equivalent to racks equipped with Blackwell chips.

Industry research points out that chillers are the "main" equipment provided by companies like Trane and Johnson Controls to data centers.

Baird analyst Timothy Wojs wrote in a report that Jensen Huang's comments "raised some questions and concerns about the long-term positioning of cooling equipment in data centers, especially as liquid cooling technology becomes increasingly popular."

Barclays analyst Julian Mitchell pointed out, "Given Nvidia's dominant position in the entire AI ecosystem, its remarks, though seemingly exaggerated, should not be underestimated."