Nvidia's Blackwell AI Servers Face Potential Delays Due to Technical Challenges

Nvidia's Blackwell AI Servers Face Potential Delays

Nvidia, the leading AI chip manufacturer, may be facing significant challenges with its next-generation Blackwell AI servers. According to a report by TrendForce, the mass production and peak shipments of Blackwell machines, including the B200 and GB200 platforms, could be postponed until mid-2025, representing a delay of nearly six months 1

1

2

Technical Challenges

The primary issues causing the potential delay are:

1. Overheating: The Blackwell GPUs are reportedly prone to overheating in servers equipped with 72 processors, even at high power consumption levels 1

   1

   .

2. Power Consumption: The power requirements for Blackwell-based servers have increased significantly. An Nvidia NVL72 rack based on the GB200 platform with 72 B200 GPUs is now expected to consume 140 kW of power, up from the previously reported 120 kW 1

   1

   2

   2

   .

3. Interconnection Optimization: TrendForce claims that Nvidia needs to optimize its interconnections, particularly the high-speed NVLink technology used for GPU-to-GPU communication 3

   3

   .

Cooling Solutions and Power Management

The extreme power consumption of Blackwell servers necessitates advanced cooling solutions:

• Liquid cooling is essential for Blackwell servers, as traditional air cooling is insufficient for the high thermal loads 2

  2

  .
• Current sidecar coolant distribution units (CDUs) can only handle 60 kW to 80 kW of thermal power 1

  1

  .
• Cooling system providers are working to optimize cold plate designs and increase CDU capacity, with liquid-to-liquid in-row CDUs expected to exceed 1.3 mW performance 1

  1

  3

  3

  .

Related Stories

Nvidia Overcomes Technical Hurdles, Accelerates Production of Blackwell AI Servers

NVIDIA's Blackwell GB200 AI Servers Set for December Deployment

NVIDIA's AI AI server evolution: GB300 production ramps up as Vera Rubin looms on the horizon

NVIDIA's GB200 AI Servers Face Overheating Rumors and Order Reductions Amid Conflicting Reports

Market Impact and Future Outlook

The potential delay could have significant implications for the AI hardware market:

• Limited quantities of Blackwell-based servers are expected to ship in 2024, with Dell already shipping some Blackwell server racks 1

  1

  .
• The GB200 NVL72 model is projected to be the most widely adopted in 2025, potentially accounting for up to 80% of total deployments 2

  2

  3

  3

  .
• The delay may affect the launch timeframe and availability of other Blackwell-based products, including the B200A and the refreshed B300 and GB300 machines 1

  1

  .

Industry Response

The AI industry is adapting to the challenges posed by these high-performance servers:

• Cloud Service Providers (CSPs) are accelerating the adoption of liquid cooling solutions to manage the increased thermal loads 3

  3

  .
• Suppliers of coolant distribution units are working to improve cooling efficiency by increasing rack sizes and developing more efficient cold plate designs 3

  3

  .

As Nvidia works to overcome these technical hurdles, the AI hardware landscape continues to evolve, with power efficiency and thermal management becoming increasingly critical factors in the development of next-generation AI infrastructure.