$NVIDIA (NVDA.US)$ 💡 NVIDIA's next-generation AI chip strat...
💡 NVIDIA's next-generation AI chip strategy: the advent of B200A and prospects for the Blackwell platform (TrendForce)
1 ️ ︎ B200A Overview and Positioning
According to TrendForce's forecast, NVIDIA $NVDA is scheduled to announce B200A in the first half of 2025 (1H25). The new chip has the following features:
- Uses 144GB HBM3e (36GB*4) memory
- Mainly targets edge AI applications
- Uses Cowos-S packaging technology
- Enables lower TDP (Thermal Design Power) compared to B200
- Can be used with air cooled solutions and GB racks
The B200A is positioned as a simplified version of the B200 and is mainly designed for OEM customers. Thus, NVIDIA responds to diverse market needs and is taking a strategy to expand its product lineup.
2 ️ ︎ Product Strategy and Market Launch Timing
- B100/B200: Launched in the second half of 2024 (2H24), mainly for CSPs (cloud service providers)
- H200: Full-scale shipments will begin in the 3rd quarter of 2024 (3Q24)
- B200A: Shipments to OEMs will begin in the first half of 2025 (1H25) or later
Nvidia considers restrictions on Cowos-L production capacity and has a policy of preferentially supplying B100/B200 to CSP customers with high demand. With the introduction of the B200A, it is expected that delays due to the complexity of liquid cooling designs in 2025 will be mitigated.
3 ️ ECT Platform Transition and Shipment Forecast
- 2024: Hopper platform (H100/H200) will be the main force
-- For North American CSP/OEM: H100/H200
-- For Chinese customers: AI server with H20
- 2025: Blackwell platform becomes the mainstay
-- For high performance AI demands: B200/GB200 racks
-- B100 will gradually be replaced by B200/B200A/GB200 racks as a featured product
According to TrendForce's forecast, the Blackwell platform will account for 80% or more of Nvidia's high-performance GPU shipments in 2025, and the annual growth rate is expected to reach 55%.
📍 Future prospects
NVIDIA is developing a wide product lineup from high-end (B200) to mid-range (B200A) to meet diverse market needs. Thus, market share expansion and diversification of revenue sources can be expected. In particular, the introduction of a B200A using CowOS-S is considered a smart strategy in response to the production capacity restrictions of CowOS-L. Thus, production capacity bottlenecks can be mitigated and more customer needs can be met. Entering the small-scale AI market with B200A can also be thought of as a move to develop growth markets other than cloud AI. An official opinion on the NVIDIA financial results announcement at the end of this month is awaited.
1 ️ ︎ B200A Overview and Positioning
According to TrendForce's forecast, NVIDIA $NVDA is scheduled to announce B200A in the first half of 2025 (1H25). The new chip has the following features:
- Uses 144GB HBM3e (36GB*4) memory
- Mainly targets edge AI applications
- Uses Cowos-S packaging technology
- Enables lower TDP (Thermal Design Power) compared to B200
- Can be used with air cooled solutions and GB racks
The B200A is positioned as a simplified version of the B200 and is mainly designed for OEM customers. Thus, NVIDIA responds to diverse market needs and is taking a strategy to expand its product lineup.
2 ️ ︎ Product Strategy and Market Launch Timing
- B100/B200: Launched in the second half of 2024 (2H24), mainly for CSPs (cloud service providers)
- H200: Full-scale shipments will begin in the 3rd quarter of 2024 (3Q24)
- B200A: Shipments to OEMs will begin in the first half of 2025 (1H25) or later
Nvidia considers restrictions on Cowos-L production capacity and has a policy of preferentially supplying B100/B200 to CSP customers with high demand. With the introduction of the B200A, it is expected that delays due to the complexity of liquid cooling designs in 2025 will be mitigated.
3 ️ ECT Platform Transition and Shipment Forecast
- 2024: Hopper platform (H100/H200) will be the main force
-- For North American CSP/OEM: H100/H200
-- For Chinese customers: AI server with H20
- 2025: Blackwell platform becomes the mainstay
-- For high performance AI demands: B200/GB200 racks
-- B100 will gradually be replaced by B200/B200A/GB200 racks as a featured product
According to TrendForce's forecast, the Blackwell platform will account for 80% or more of Nvidia's high-performance GPU shipments in 2025, and the annual growth rate is expected to reach 55%.
📍 Future prospects
NVIDIA is developing a wide product lineup from high-end (B200) to mid-range (B200A) to meet diverse market needs. Thus, market share expansion and diversification of revenue sources can be expected. In particular, the introduction of a B200A using CowOS-S is considered a smart strategy in response to the production capacity restrictions of CowOS-L. Thus, production capacity bottlenecks can be mitigated and more customer needs can be met. Entering the small-scale AI market with B200A can also be thought of as a move to develop growth markets other than cloud AI. An official opinion on the NVIDIA financial results announcement at the end of this month is awaited.
References
Disclaimer: Community is offered by Moomoo Technologies Inc. and is for educational purposes only.
Read more
Comment
Sign in to post a comment
182057795 : I won't sell it until it's 200 (laughs)