Nvidia's $50 billion forecast isn't enough! Colette Kress confirms: it will be higher, Huang emphasizes strong demand from Chinese customers

Nvidia CFO Colette Kress sent a strong signal at an event hosted by JP Morgan, stating that the company’s previous revenue target of $50 billion for data center chips “will definitely be surpassed.” This statement indicates that since Jensen Huang announced this forecast at GTC last October, customer enthusiasm for Nvidia products has far exceeded expectations.

Revenue Targets Continue to Rise, Optimism Grows

Kress clearly pointed out that current customer procurement intentions have significantly strengthened compared to when the forecast was released. She emphasized in her speech that this $50 billion target “will definitely be higher.” According to Goldman Sachs analysis, this goal was already about 12% above market consensus at the time, and Kress’s further upward revision again confirms Nvidia’s strong momentum.

Kress noted that Nvidia’s optimism about the prospects of AI applications is driven not only by the explosive demand for AI itself but also by the rapid growth in enterprise data processing needs. These factors combined will propel the growth of next-generation computing demand, with total related investments expected to reach trillions of dollars by the end of 2030.

It is worth noting that Kress specifically stated that this $50 billion order volume does not include the latest collaboration with OpenAI, and the previously announced $10 billion investment agreement with OpenAI has not yet been officially finalized. This means Nvidia still has room to secure additional orders beyond the existing forecast.

Huang Announces Rubin Chips Fully in Production at CES

Jensen Huang officially announced at the Las Vegas CES that the new generation Vera Rubin AI platform has entered full production. The first batch of customer deliveries is expected to begin in the second half of 2026.

Rubin chips feature an integrated design of six new chips, achieving leapfrog improvements in inference cost and training efficiency. Huang described it as a “huge monster,” and explained the core design philosophy—reducing inference costs by 10 times annually, while AI-generated tokens increase fivefold each year.

In terms of performance metrics, Rubin GPUs reach 50 PFLOPS of inference performance (using NVFP4 precision), equivalent to five times that of Blackwell; training performance is 35 PFLOPS, a 3.5x improvement over the previous generation. More importantly, the cost of token generation during inference can be reduced to one-tenth of the Blackwell platform.

Each GPU is equipped with eight HBM4 memory modules, with bandwidth up to 22 TB/s. To address the “memory” bottleneck in AI systems, Nvidia developed an inference context storage platform based on BlueField-4 DPU, providing an additional 16 TB of high-speed shared memory per GPU, connected via 200 Gb/s bandwidth, specifically to tackle storage wall issues in long-text processing.

Energy Efficiency Breakthrough Reshapes Data Center Cooling Solutions

In terms of energy efficiency, the Rubin NVL72 rack has achieved a 100% liquid cooling solution, supporting an inlet water temperature of 45℃. This means data centers no longer need to rely on power-hungry chiller units. Huang estimates this could save about 6% of global data center electricity consumption.

Chip Remarks Trigger Industry Chain Shakeup

Huang’s emphasis on the importance of storage and memory at CES had immediate impact. He pointed out that this has created an unprecedented market, potentially becoming the world’s largest storage market—essentially the “working memory” for global AI applications.

Driven by this, storage chip giant SanDisk became the best-performing stock in the S&P 500 on Tuesday, closing up nearly 27.6%, marking the largest single-day gain since February 18, 2025. Western Digital and Seagate also recorded double-digit gains on Tuesday.

Conversely, data center cooling system manufacturers’ stocks suffered heavy declines. Johnson Controls fell as much as 11% during the day, ultimately closing down over 6.2%; Modine Manufacturing dropped 21% at one point, later narrowing to a 7.5% decline; Trane Technologies and Carrier Global fell 2.5% and 0.5%, respectively. The reason for the decline is directly linked to Huang’s remarks—Rubin chips can be cooled with warm water without the need for chillers.

Baird analyst Timothy Wojs wrote that Huang’s comments “raise some questions and concerns about the medium- to long-term positioning of data center cooling equipment, especially as liquid cooling technology becomes more widespread.” Barclays analyst Julian Mitchell believes that given Nvidia’s dominant position in the entire AI ecosystem, while such statements should not be overinterpreted, they should also not be taken lightly.

Attention on China’s Market Potential

Regarding the potential impact of the Chinese market on Nvidia’s revenue, Kress stated that the US government is advancing the approval process for license applications. Nvidia has received orders from Chinese customers, but the specific outcome remains uncertain. Huang, in an interview, particularly emphasized the strong demand from Chinese customers, indicating that this market is crucial for Nvidia’s growth prospects.

Despite frequent positive signals from management, Nvidia’s stock performance on Tuesday was not ideal. It rose over 2% in the early trading session but ultimately closed down nearly 0.5%, marking two consecutive days of decline. Since 2025, Nvidia’s cumulative increase has been about 39%, far below the over 170% surge in 2024, mainly due to concerns over the AI bubble and competition from Google’s TPU.