Nvidia’s supply chain partners have resolved a series of technical issues that had delayed the rollout of the company’s next-generation “Blackwell” AI servers, marking a major boost to the chipmaker’s efforts to meet surging global demand for artificial intelligence computing infrastructure. The breakthrough, reported by the Financial Times, is expected to significantly accelerate sales and stabilize production heading into the second half of the year.
Key manufacturing partners including Foxconn, Inventec, Dell, and Wistron have begun shipping Nvidia’s cutting-edge GB200 AI racks, which combine 36 Grace CPUs with 72 Blackwell GPUs. These powerful units are connected via Nvidia’s high-bandwidth NVLink system and are designed to handle the enormous processing requirements of large language models and advanced AI workloads.
Production had previously been stalled due to a range of hardware and software problems, including overheating, liquid cooling system leaks, and synchronization errors across the densely packed processors. Engineers from the supply chain confirmed to the Financial Times that they collaborated closely with Nvidia to resolve the issues in recent months.
“Our internal tests showed connectivity problems… the supply chain collaborated with Nvidia to solve the issues, which happened two to three months ago,” one engineer stated.
The resolution comes just in time for Nvidia’s quarterly earnings report, where investors will be watching closely for confirmation that Blackwell shipments are ramping up. The company had announced ambitious revenue goals of $43 billion for the quarter ending in April — a record and roughly 65% year-on-year increase.
Analysts and industry insiders see the recovery in production as critical not only for meeting those targets but also for offsetting challenges in other areas. Nvidia is facing a $5.5 billion revenue hit due to U.S. export restrictions on its AI chips to China, including the H20 model. Analysts at Bank of America noted last week that this could drag gross margins down to around 58%, compared to Nvidia’s earlier estimate of 71%.
Despite the production delays, demand for the Blackwell chips remains robust. Countries like Saudi Arabia and the United Arab Emirates have placed large orders for Blackwell units, underscoring Nvidia’s strategic pivot toward sovereign customers and away from over-reliance on Big Tech “hyperscalers” like Microsoft and Meta.
To improve delivery timelines, Nvidia has also made a key design compromise. While the company had initially planned to introduce a new modular GPU board known as “Cordelia” — allowing for more flexible maintenance and higher margins — it has reverted to the current “Bianca” layout to simplify deployment. However, Nvidia reportedly still intends to roll out the Cordelia design in its next-generation chips.
Speaking at the Computex conference in Taipei last week, CEO Jensen Huang reaffirmed Nvidia’s commitment to pushing the boundaries of AI hardware. He also previewed the upcoming GB300 rack, which will feature expanded memory and support for increasingly complex AI models such as OpenAI’s 01 and DeepSeek R1. The GB300 is expected to launch in the third quarter of this year.
“This technology is really complicated. No company has tried to make this many AI processors work simultaneously in a server before, and in such a short timeframe,” said Chu Wei-Chia, a Taipei-based analyst at consultancy SemiAnalysis, speaking to the Financial Times. “The inventory risk around GB200 will ease off as manufacturers increase rack output in the second half of the year.”
As Nvidia confronts growing regulatory pressures, fierce competition, and geopolitical headwinds, the successful resolution of its rack manufacturing issues positions the company to maintain its dominance in the red-hot AI hardware race — and signals that the next phase of its growth may now be back on track.
