DeepSeek V5 May Train on Huawei Ascend 950 Chips in 2027
Cailian reports DeepSeek V5 would pre-train on Huawei Ascend 950, shifting Ascend from inference into training. What that means for China AI and Nvidia.
·15 min read
Topic
DeepSeek
5 articles
AIGeopoliticsTech IndustryDeveloper ToolsCybersecurityInfrastructureAI InfrastructureAI ModelsSemiconductorsCloud InfrastructureWeb DevelopmentBreaking News
DeepSeek V4.1 Flash: 552B MoE, 1M Context, $0.30 Input, MIT License
DeepSeek V4.1 Flash is a 552B MoE with 1M context, MIT weights and $0.30 per million input tokens. Benchmarks, third-party scores and self-hosting math.
·15 min read
DeepSeek V4 Is Coming: 1 Trillion Parameters, Multimodal, Huawei Ascend-Optimised
DeepSeek's next flagship model is imminent — 1 trillion parameter MoE architecture, multimodal support, 1M token context, trained on Huawei Ascend. Here's what it means for developers and the widening US-China AI stack split.
·8 min read
DeepSeek V4: 1M Context, Multimodal, Coding Benchmarks — What Developers Get in 2026
DeepSeek V4 launch: 1 million token context, multimodal, coding-first. Benchmarks vs GPT-4o and Claude, API pricing, and what developers actually get in 2026.
·6 min read
DeepSeek R1 Explained: What It Is, Why It Shook the AI World, and What Comes Next
DeepSeek R1 matched GPT-4 performance for $6 million — a fraction of what OpenAI spent. Here is a plain-English explanation of what DeepSeek actually is, why Nvidia lost $500 billion in a day, and what it means for developers and businesses.
·8 min read