NVIDIA CFO Colette Kress pointed at geopolitical troubles between the US and China as hampering her firm's ability to recognize revenue from its H20 AI GPU sales to the Asian country. Kress was speaking at the Goldman Sachs Communacopia + Technology Conference, where she also shared details about the firm's next-generation Vera Rubin AI GPUs.
According to Kress, NVIDIA has received H20 licenses from the Trump administration, and it could account for as much as $5 billion in revenue from H20 sales in the third quarter if tensions smooth out between the world's largest economies.
NVIDIA CFO Comments On Blackwell GB300 Ramp In Q2
Kress started out by commenting on NVIDIA's data center revenues and the growth despite removing the China-specific H20 AI GPUs from the mix. According to her, looking at NVIDIA's revenue, which includes data center and networking, leads to a 12% sequential, or quarter-over-quarter, growth in the second quarter.
She added that "what we're targeting right now for Q3 in our outlook is a 17% growth sequentially," which indicates that NVIDIA is witnessing a surge in demand for its products looking forward.
According to the NVIDIA CFO, apart from seeing demand for the GB200 network racks, her firm was
"also at scale with our GB300 Ultra." This scale-up was surprising to some as she mentioned a lot "of discussion that said, I didn't know that would actually be a big part."
The scale-up was "was seamless, it was a seamless transition that many people didn't understand the amount of scale and volume we were actually able to put into market as well," she added. As a result, Kress believes that her firm will further scale up GB200 and GB300 racks in the third quarter. Her comments fall in line with a recent report which suggests that NVDIA might experience a 300% sequential growth with GB300 in Q3.
The conversation then shifted to NVIDIA's ability to sell the H20 AI GPUs to China. Kress confirmed that NVIDIA has "received licenses for several of our key customers in China." She added that her firms wants the opportunity to complete the shipments and ship the H20 architecture to its Chinese customers.
However, she added that right now "there is still, this position right now, a little geopolitical, situation that we need work through between the two governments." According to Kress, NVIDIA's Chinese customers want to ensure that the "government is also very well received in terms of receiving the H20 to them."
Multiple reports have suggested that the Chinese government is urging local businesses to use local chips instead of NVIDIA's products. However, Kress asserted that the firm believes that "there's a strong possibility that this [the China H20 shipments] will occur." According to her, while it is hard to determine the amount of additional revenue the shipment should add, it could range between $2 billion to $5 billion.
NVIDIA's shares have been under pressure lately due to Broadcom's earnings last week which saw the firm outline a new $10 billion custom AI chip contract. With reports suggesting that the contract could be from OpenAI, and debate in the industry wondering whether a push towards cost-effective chips is the right approach, Kress asserted that power is indispensable when it comes to AI computing.
According to her, individuals generating more tokens and inferencing demands for reasoning models account for a large portion of reasoning model demand. The reasoning models are important since if they execute successfully, they can do work through forms such as agentic AI. As a result, NVIDIA is focused on creating data center solutions on scale.
According to the CFO, her firm is looking to create large AI systems because:
"And so we are looking at our position of creating a data center scale that can do the most performant but the most performant per watt and the most performant per dollars, as well. That wattage is such an important thing. Right now you can decide whether or not capital or power is more important. In respects [inaudible] they both are tremendously important. But when you are purchasing any type of large system as we are, you have to keep in mind, you will be using power throughout the journey of owning that full cluster, four, six years or even further. So having that high performance is going to be very important to make sure you are properly addressing that power that is going to be needed for that. So we stand very strong in terms of how we thought about that transition and moving to a full data center scale solution. Some of them focus at it in terms of it's a rack scale type of capability. Where we put all of the different chips together so that they would be working together, be optimized together, in terms of the right performance. So we feel very good with our plan."
As for NVIDIA's next generation Vera Rubin AI chips, Kress outlined that "Rubin is on a path and that one-year cadence is going to be a journey we're ready to take on with Rubin. So Vera Rubin, six chips, all of them taped out."
Kress added that as part of her firm's efforts to prepare data centers to integrate Rubin, NVIDIA has "already had discussions to where we probably will see several gigawatts needs for our Vera Rubin" and has "pencilled that in so even way before it's even ready to go to market, we already are seeing gigawatts worth of needs as we go forward."
About the author: Ramish is a seasoned technology writer and editor with more than a decade of experience. He specializes in semiconductor fabrication and market analysis. With a background in finance and supply chain management - via his bachelors in Finance and a micromasters in supply chain management from MIT - Ramish combines financial rigor with deep industry insight to deliver accurate and authoritative coverage.
Follow Wccftech on Google to get more of our news coverage in your feeds.
