More from this creator
Other episodes by Kitty Cat.
More like this
If you liked this, try these.
Transcript
The full episode, in writing.
Picture this: It's July 17, 2023, and in Hangzhou, Zhejiang, a new player enters the AI arena. The company, Hangzhou DeepSeek Artificial Intelligence Basic Technology Research Co., Ltd., better known as DeepSeek, is founded by Liang Wenfeng. Now, just shy of three years later, DeepSeek's influence is rippling across the tech world. Let's dig into how this all came to be and where DeepSeek is headed.
Liang Wenfeng is not new to the game. His journey with AI began when he co-founded High-Flyer back in February 2016. At that time, High-Flyer was all about AI-driven stock trading. By late 2017, most of High-Flyer's trading was managed by AI systems. Liang's foresight didn't stop there. In 2021, he started acquiring NVIDIA GPUs, anticipating the AI boom and future restrictions on chip sales to China. This decision has proved pivotal in DeepSeek's journey.
Fast forward to 2024, DeepSeek releases DeepSeek-V2, a model that sets the stage for a price war in the AI market. Competitors like ByteDance, Tencent, Baidu, and Alibaba were forced to cut prices on their models. Just months later, DeepSeek-V3 and DeepSeek-R1 hit the scene, marking DeepSeek's entry into the international market and challenging the notion of U.S. dominance in AI development.
The DeepSeek-R1 model especially turned heads. It was comparable to OpenAI's GPT-4, yet it was trained at a fraction of the cost—about $6 million compared to GPT-4's hefty $100 million price tag. This disrupted the global AI market, triggering significant stock value declines for major U.S. tech companies and forcing a strategic rethink.
So, what's DeepSeek up to now in April 2026? In February, they launched DeepSeek-V4 but made an unconventional move—they withheld this model from U.S. chipmakers like Nvidia and AMD, choosing instead to provide early access to domestic suppliers such as Huawei. This was a break from the norm, as pre-release versions are typically shared with chipmakers for performance tweaks.
Interestingly, despite U.S. export controls, DeepSeek trained its latest AI model on Nvidia's advanced Blackwell chips. This move has raised questions about potential violations of U.S. export restrictions. Moreover, they've significantly enhanced their flagship model by expanding its context window from 128,000 tokens to over 1 million, enabling it to handle more complex prompts. The model's knowledge cut-off has also been extended to May 2025, ensuring it's equipped with more recent information.
However, not everything has been smooth sailing. Earlier this month, DeepSeek experienced a significant service outage lasting over seven hours, affecting millions of users. The company restored service but has remained tight-lipped about the cause of the disruption.
DeepSeek's rise is reshaping the competitive landscape. Their cost-effective AI models are pushing U.S. companies to reassess their strategies and investments in AI development. As a result, the market is buzzing with competition, driving innovation and, no doubt, a bit of tension.
DeepSeek has shown that a combination of strategic foresight, cost-efficiency, and bold moves can shake up the AI industry. Liang Wenfeng's vision is steering the company toward becoming a formidable competitor on the global stage, and its impact is palpable—both in China and internationally.
So, as DeepSeek continues to chart its course, it's clear that this is a company not just keeping pace with the AI race but actively redefining it. Where they go from here will undoubtedly be watched closely by industry leaders and competitors alike. Stay tuned—this story is far from over.