DeepSeek AI, relatively new to the scene, has rapidly become a significant player, causing ripples that have even shaken the US stock market. The company’s release of DeepSeek AI, a ChatGPT rival, went viral just weeks ago. What’s particularly striking is that DeepSeek achieved this impressive feat despite facing US sanctions that limit its access to cutting-edge GPUs, vital for AI development. Not even Google has matched this speed.
One of the most remarkable aspects of DeepSeek’s success is its ingenuity in the face of adversity. Lacking access to the latest hardware, the company developed software tools to train its AI models at a fraction of the cost of competitors like OpenAI. This innovation led to concerns in the US stock market, with investors worried that the usual reliance on cutting-edge hardware was not the only path to AI dominance, causing a $1 trillion dip. While the concerns might have been a bit overstated, DeepSeek doesn’t seem to be slowing down. They’re planning a major upgrade to their reasoning model, DeepSeek R2, with a target release date in May.
China still faces restrictions on accessing the newest chips, which means the development of DeepSeek R2 will depend on the company’s stockpiles, and software optimization will be crucial. However, DeepSeek might excel not just in software and hardware but also in its unique company culture, which could give it an edge against OpenAI and other Western AI firms. Reports suggest the company’s internal values were instrumental in making the DeepSeek R1 breakthroughs happen.
OpenAI has been releasing new reasoning models, including o3-mini and o3-mini-high, since DeepSeek R1’s launch. OpenAI’s GPT-4.5 is also reportedly on the way, with a larger GPT-5 upgrade to follow. This may explain the pressure on DeepSeek to accelerate its own upgrades.
According to Reuters, the R2 model is now slated for release before May, which is a few weeks earlier than initially expected. The new model should provide improved coding capabilities and support for multilingual reasoning. DeepSeek R2 is expected to use the software innovations already used for its existing models—particularly the Mixture-of-Experts (MoE) technology, which selectively activates parts of an AI model needed for a specific task, and Multihead Latent Attention (MLA), which enables the AI to process multiple aspects of a prompt at once. This all happens without access to the newest hardware. Analyst estimations point out that DeepSeek’s pricing might be, astoundingly, 20 to 40 times cheaper than ChatGPT tools. This price advantage puts serious pressure on OpenAI and Google to lower prices for ChatGPT and Gemini.
As DeepSeek navigates the market, it continues to operate under restrictions, unable to purchase the same Nvidia chips that rival firms can access. The US may even tighten existing chip bans in the months ahead, and it is possible that the DeepSeek R2 upgrade will further fuel future ban decisions.
The company’s internal culture may be a huge factor in its success so far: Reports from Reuters portray a company where employees genuinely enjoy their work due to the forward-thinking management of Liang Wenfeng. Wenfeng, a 40-year-old billionaire, initially employed AI as the foundation for High-Flyer, his quantitative hedge fund. High-Flyer reinvested 70% of its profits into AI research even before ChatGPT became a well-known product. A few years ago, the company invested in two AI supercomputing clusters, including Fire-Flyer II, which included 10,000 Nvidia A100 chips. Those chips were banned from purchase in China in 2022, and DeepSeek used them to train its models. Rumors suggest that the company may have smuggled tens of thousands of additional chips since then. However, that is something that’s unlikely to be confirmed.
Wenfeng has designed DeepSeek as a research lab rather than a profit-focused firm and instituted a different management style than those of its competitors. This is a notable difference when compared to the traditional “996” work culture prevalent in other tech firms. (meaning 9 AM to 9 PM, six days a week). DeepSeek researchers work 8-hour days.
One can’t help but wonder if working 8-hour days is enough for the DeepSeek R2 development, especially considering the company’s hastened release schedule. Furthermore, the report reveals that Wenfeng recruits young engineers fresh from school, working side-by-side with them, supporting them and allowing them to take ownership of DeepSeek research initiatives. These engineers are also well-compensated, with senior High-Flyer data scientists potentially making somewhere around 1.5 million yuan (approximately $206,000) annually, nearly doubling their competitors’ rates.
This isn’t to imply that ChatGPT engineers do not enjoy their work, or do not receive generous compensation. We have already heard about dozens of senior OpenAI executives and former co-founders who left the company to start their own AI companies. That being said, we shouldn’t expect the same transparency from Chinese companies. The Reuters report may likely highlight the company’s successes more than its challenges. The Reuters report does note that DeepSeek has quickly become a success story in China, one that Beijing fully embraces. It’s not just the DeepSeek engineers who may love the firm. The government may have investigated the large AI chip purchases a few years ago, including that 10,000-chip cluster, but DeepSeek is immensely popular now. DeepSeek AI is already being integrated in various areas, with 13 local city governments and 10 state-owned energy companies using it. Tech giants, such as Baidu, Lenovo, and Tencent, are increasingly adopting it as well. Despite any lack of confirmation of the Reuters story, it sure looks like DeepSeek is growing in popularity with Chinese companies and the government, and that sort of support can further improve the firm’s ability to compete against OpenAI, Google, and other big AI firms.
Meanwhile, the Western world is ready to propose bans on DeepSeek. That’s not surprising. DeepSeek might have gone viral, and Reuters paints a great picture of the company’s inner workings, but the AI still has issues that Western markets can’t tolerate. Countries like Italy and South Korea have already announced bans on DeepSeek AI. The US government is also evaluating plans for a wider ban. The ban centers around user data privacy, with all of DeepSeek’s data going to China. DeepSeek also has other issues, including widespread censorship of China-related topics and general AI safety concerns.
With this in mind, it’s likely that the DeepSeek R2 release by May won’t cause such a market shock as its predecessor caused. Still, it will be interesting to see how R2 stacks against ChatGPT, Gemini, and other innovations this spring.
