BREAKING
DeepSeek Building Own Inference AI Chip
The Plan So Far
1
Hire chip engineers
↓
2
Talk to foundry & memory
↓
3
Inference-only chip
↓
4
Cut Nvidia reliance
0
%
Nvidia premarket drop
0
B
China AI chip market
0
%
Huawei's share
H800 to Ascend 950
R1
Jan 2025
●
Trained on Nvidia H800
●
Low-cost, high performance
V4
Apr 2026
●
Optimized for Ascend 950
●
Huawei aided training
0
$
per M input tokens
0
M
token context
A Shift Toward Hardware Control
AI NEWS BLITZ
Reuters reports DeepSeek is developing its own AI chip for inference.