Nvidia’s highly anticipated Blackwell AI chips, which have already experienced delays, are now encountering overheating issues in server racks, according to a report by The Information. The issue is raising concerns among customers about the feasibility of preparing new data centers in time.

Overheating in High-Density Server Racks

The overheating occurs when multiple Blackwell chips, up to 72 in a rack, are connected. Nvidia has reportedly asked suppliers to redesign the racks several times to address the problem, with no suppliers named in the report.

A company spokesperson told Reuters that Nvidia is collaborating with leading cloud service providers as part of its engineering process, describing these iterations as “normal and expected.”

Implications for Customers

The delays and technical challenges could impact Nvidia’s high-profile customers, including Meta Platforms, Google, and Microsoft. These companies rely on advanced chips like Blackwell to power AI-intensive workloads such as chatbot responses and other generative AI applications.

Innovations and Expectations

Unveiled in March, Blackwell chips represent a significant leap in AI processing power, integrating two silicon components into a single unit for a 30-fold speed increase over prior models. Initially slated for shipment in the second quarter, delays have already pushed delivery schedules.

Looking Ahead

While Nvidia works to resolve the overheating issues, the delays may influence cloud service providers’ plans for AI infrastructure, especially as demand for robust AI systems continues to rise. The adjustments in server design will be crucial to ensuring the chips can meet the performance and reliability standards required for deployment.