DeepSeek + Huawei Just Went After Nvidias CUDA Moat @fahdmirza
DeepSeek + Huawei Just Went After Nvidias CUDA Moat  @fahdmirza
Uploaded September 2026 | Updated November 2026, 3 days ago
DeepSeek and Huawei just open-sourced DeepGEMM Ascend and backed TileLang.

📬Weekly AI Newsletter: fahdmirza.substack.com

🔥 Buy Me a Coffee to support the channel: ko-fi.com/fahdmirza

#deepgemm #deepgemmascend #tilelang #deepjit

PLEASE FOLLOW ME:
â–¶ LinkedIn: / fahdmirza
â–¶ YouTube: / @fahdmirza
â–¶ Blog: fahdmirza.com

0:00 Why this is buzzing today
0:35 The tweet that started it
2:00 Why CUDA is Nvidia's real moat
3:10 DeepGEMM Ascend explained
5:23 The benchmark numbers
6:14 TileLang explained
7:40 Huawei's chips and supernode
8:20 Does this break Nvidia?

RESOURCES:

â–¶ github.com/deepseek-ai/DeepGEMM-Ascend

All rights reserved © Fahd Mirza
DeepSeek + Huawei Just Went After Nvidias CUDA MoatQwen Intelligence is Here: Mobile AI AgentsYuE2 - Open Music Generation Model for Any Language LocallyIt’s Official: Qwen Will Always Be Open-SourcedUnion Alpha Stealth Model Tested: Coding, Vision, ReasoningOpus 5.5 vs GPT-6 Sol: Can AI Actually Simulate a Latte?NeoHorse-1-4B: The Model That Trains Itself - Run LocallyMicrosoft Cooks VibeVoice-ASR: Handle 60-minute Long-form AudioOrnith-1.5-9B: Great on Paper, Struggled in My Tests LocallyWhich AI OCR Model Fits YOUR Use Case? (Ultimate 2025 Guide!)Thomson Reuters Built Their Own AI Lawyer: Run Thomson-1 LocallyQwen3.8-27B Obliterated: Dont Use this Model in Production
Fahd Mirza |

DeepSeek + Huawei Just Went After Nvidia's CUDA Moat

SHARE TO X SHARE TO REDDIT SHARE TO FACEBOOK WALLPAPER