Ascend 950DT vs B300 Blackwell Ultra vs B200 vs MI400 vs H100

Pyyan / Compare / Ascend 950DT vs B300 Blackwell Ultra vs B200 vs MI400 vs H100

5 of 5

GPU & AI Chips · verified 10 Sept 2026

×HAscend 950DTHuaweicurrent

×B300 Blackwell UltraNVIDIAcurrent

×B200NVIDIAcurrent

×MI400AMDcurrent

×H100NVIDIAcurrent

Specification	Ascend 950DT	B300 Blackwell Ultra	B200	MI400	H100
Summary	144GB on Huawei's own HBM, and the part China is building gigawatts around.	15 PFLOPS dense FP4.	The current volume Blackwell part.	Double NVIDIA's memory capacity.	The workhorse of the previous generation.
Memory	144GB HiZQ 2.0	288GB HBM3e	192GB HBM3e	432GB HBM4	80GB HBM3
Bandwidth	4 TB/s	8 TB/s	—	19.6 TB/s	—
Price	Above ¥250,000 (about $37K), reported	~$40–50K (verify)	$30–50K	—	$25–40K
Vendor	Huawei	NVIDIA	NVIDIA	AMD	NVIDIA
Category	GPU & AI Chips	GPU & AI Chips	GPU & AI Chips	GPU & AI Chips	GPU & AI Chips
Official	Huawei ↗	NVIDIA ↗	NVIDIA ↗	AMD ↗	NVIDIA ↗

Highlighted rows are where these differ.

4 TB/s memory bandwidth and a 2 TB/s interconnect, 2.5x the Ascend 910C
HiZQ 2.0 is Huawei's own memory standard rather than a licensed HBM3 or HBM3e part
Indicated price rose 20% to 50% in two months to September 2026, on a memory shortage

Best for serving models where Nvidia cannot be bought.

Best for frontier training.

Best for general training.

Best for large-model inference.

The workhorse of the previous generation.

Best for value.