NVIDIA is expected to see a surge in the adoption of their Blackwell GB200 AI servers, with projections indicating that 2 million units will be shipped by the year 2025. This growth is fueled by a shift to a new packaging technology called PFLO, which is set to address supply chain bottlenecks that have hindered previous product launches.
Last year, NVIDIA experienced success with their Hopper AI products, which boosted the company’s financial performance but also highlighted issues within the supply chain. The main challenges were related to the supply of HBM & CoWoS, leading to significant delays in fulfilling orders. In response, NVIDIA is making a strategic move to overcome these challenges by transitioning to the new PFLO standard, with 420,000 units expected to be shipped this year and up to 2 million by 2025-2026.
Taiwan Economic Daily reports that NVIDIA’s GB200 AI systems will feature a “panel-level deployable package,” which is seen as a superior alternative to traditional CoWoS packaging. This transition is expected to occur towards the end of 2025, as NVIDIA aims to meet the growing demand for its Blackwell products. By adopting panel-level distribution earlier than planned, NVIDIA is positioning itself for success and ensuring scalability in production.
PFLO, the new packaging standard, integrates multiple individual integrated circuits on separate silicon wafers using materials like laminate or glass. While specific comparisons with CoWoS are not available yet, the incorporation of PFLO with Blackwell products suggests that it offers competitive performance and scalability benefits.
Suppliers for the new packaging standard are currently limited, with Taiwan’s energy technology and Innolux vying for NVIDIA’s orders. The company anticipates shipping 420,000 Blackwell GB200 units in the latter half of 2024, with production estimates ranging between 1.5 million and 2 million for the following year. This significant increase in production underscores the potential revenue growth for NVIDIA with the Blackwell generation.
The cost of each GPU is expected to range from $35,000 to $45,000, with servers priced up to $3 million. As demand for these products rises, NVIDIA stands to benefit financially, especially with Dell poised to launch servers equipped with up to 8 B200 Blackwell AI GPUs. The inclusion of liquid cooling in these servers is expected to make them a popular choice in the market.
In conclusion, NVIDIA’s transition to the PFLO standard for their Blackwell GB200 AI servers signals a strategic move to address supply chain challenges and meet increasing demand. The company’s focus on innovative packaging technologies and collaborations with key suppliers position them for success in the rapidly evolving AI hardware market. With projections of 2 million units shipped by 2025, NVIDIA is on track to capitalize on the growing demand for advanced AI solutions.
Article Source
https://wccftech.com/nvidia-ship-half-a-million-blackwell-gb200-ai-chips-this-year-2-million-in-2025/