After a year, the DeepSeek effect is back! DeepSeek V4 is about to launch, and it might be the main reason behind NVIDIA's stock plunge after last night's earnings report. Multiple media outlets are reporting that DeepSeek V4 will go live within a week. But what's even more critical is that DeepSeek has granted early access to its major update, V4, to domestic suppliers like Huawei to help them optimize processor software and ensure the model runs efficiently on hardware—essentially bypassing NVIDIA entirely. Here's the latest info: the model is codenamed Sealion-lite, featuring a 1-million-token context window. This model is much more user-friendly than the web/app versions and is a native multimodal model. Testing results for DeepSeek V4 Lite show significant improvements compared to version V3.2. Even DeepSeek V4 Lite (non-thinking mode) alone is more powerful than DeepSeek V3.2 (thinking mode), and it's multimodal too. If they truly achieve a small-parameter model with a long context window, it means ultra-low inference costs for extended contexts, which is hugely significant and could potentially reshape the entire industry. If DeepSeek V4 once again demonstrates astonishing efficiency and capabilities, we might see another moment of panic in the U.S. stock market, just like a year ago.