In the last 48 hours, the world has turned upside down. Tech stocks lost $1 trillion, the US is no longer the sole custodian or leader of AI, as it claims to be, and India was confronted with the reality that it is not even contesting in the AI arms race—all because of a small research AI lab in China that has shaken Silicon Valley and Washington, D.C. Started a year ago by Liang Wenfeng, DeepSeek has suddenly become a competitor to OpenAI and perhaps the leading AI labs, challenging the AI superpowers and US dominance in tech.
In fact, the way the DeepSeek R1 model has performed in global benchmarks, delivering the same results with fewer resources compared to AI models from US counterparts, has made Wenfeng the Sam Altman of China. Although many cast doubt on how China has made AI advancements despite the United States’ efforts to restrict the supply of high-powered AI chips to the communist nation, citing national security concerns, DeepSeek’s rise shows how innovative techniques and cost-effective solutions may be all you need to create a world-class product.
Unlike Silicon Valley starlets, who have a strong background in tech, Wenfeng comes from the world of finance. The 40-year-old Wenfeng is not the typical founder you come across in tech, and his profile makes him all the more interesting.
After graduating from Zhejiang University, he co-founded the quantitative hedge fund High-Flyer in 2015. Thanks to its unique funding model and his interest in predicting market trends using AI, he was able to pursue AI projects without pressure from external investors, prioritising long-term research and development instead.
However, in 2021, Wenfeng began buying thousands of Nvidia chips as part of a side AI project—well before the Biden administration started limiting the supply of cutting-edge AI chips to China. Nobody would have thought that Wenfeng’s rationale for hoarding graphics processors would eventually make sense. After all, there was no concrete plan in place until Wenfeng launched DeepSeek in 2023.
“When we first met him, he was this very nerdy guy with a terrible hairstyle talking about building a 10,000-chip cluster to train his own models. We didn’t take him seriously,” one of Liang’s business partners told the Financial Times in an interview. “He couldn’t articulate his vision other than saying: ‘I want to build this, and it will be a game change.’ We thought this was only possible from giants like ByteDance and Alibaba.”
Although DeepSeek was initially a side project, Wenfeng was passionate about artificial intelligence and personally involved in the startup, with a major focus on research and development. In fact, Wenfeng envisioned DeepSeek as a homegrown leader in AI that could compete with China’s biggest tech companies as well as US tech majors.
Story continues below this ad
Interestingly, to build a team of AI researchers, Wenfeng recruited top young talent from local Chinese universities and didn’t look beyond China, offering salaries on par with what top Chinese tech companies like ByteDance pay. This strategy helped Wenfeng attract the best talent from China—individuals who may not have years of experience in building AI but possessed the technical skills needed to develop AI from scratch.
DeepSeek’s first AI model, DeepSeek Coder, was released in November 2023 as an open-source model designed for coding tasks. This was followed by DeepSeek LLM, a 67B parameter model aimed at competing with other large language models. In May 2024, DeepSeek-V2 was released, which was well-received due to its strong performance and low cost. In reaction to the release of the DeepSeek-V2 model, there was an uproar in the Chinese AI market, triggering a price war that forced major Chinese tech giants, such as ByteDance, Tencent, Baidu, and Alibaba, to lower their AI model prices to remain competitive.
DeepSeek-V2 was succeeded by DeepSeek-Coder-V2, a much more advanced model with 236 billion parameters. Designed for complex coding challenges, it features a high context length of up to 128K tokens. This model is available through a cost-effective API, priced at $0.14 per million input tokens and $0.28 per million output tokens.
The company’s latest models, DeepSeek-V3 and DeepSeek-R1, further established DeepSeek as a leading AI research lab in China. DeepSeek-V3, a 671B parameter model, offers impressive performance on various benchmarks while requiring significantly fewer resources than AI models from US-based tech giants. However, it was DeepSeek-R1, released in January 2025, that focused on reasoning tasks and challenged OpenAI’s GPT-4 model with its advanced capabilities, making everyone take notice of DeepSeek. Wenfeng’s year-old company stated that its latest AI model, R1, spent just $5.6 million on computing power for its base model, compared to the hundreds of millions or even billions of dollars that US companies spend on their AI technologies. The R1 AI model came out of nowhere, and because the company spent only a fraction of the money on its development (with a team of only 200 people), its low cost of operation shocked Silicon Valley. In no time, the DeepSeek app made it to the top of the US app store, dethroning ChatGPT.
Story continues below this ad
AI is not only cost-intensive but also power-hungry, which is why major US tech companies are spending billions of dollars on developing AI models. In fact, there is no guarantee that these tech companies will ever recoup the investments they are making in AI development. However, DeepSeek has demonstrated that it is possible to develop similar AI capabilities to those of US tech companies at a fraction of the cost and on less powerful chips. In the case of DeepSeek, the company trained its latest model on Nvidia H800 chips, which are significantly less powerful than Nvidia’s Blackwell chips, with the next-generation chips from Nvidia costing anywhere between $30,000 to $40,000 per unit.
Experts already see Wenfeng’s AI strategy as effective, putting China on the global AI map while being cost-effective and aiming to scale AI. DeepSeek’s success can be attributed to something called reinforcement learning, a concept where AI models learn through trial and error and self-improve through algorithms. This approach is very similar to how humans learn through experience. Essentially, DeepSeek’s models learn by interacting with the environment and receiving feedback based on their actions. This, in return, makes AI models get better with reasoning and able to solve complex problems. DeepSeek has also managed to champion the distillation of its large model’s capabilities into smaller, more efficient models.
What has perhaps made everyone notice about DeepSeek is its cost-effective approach, which is unique and different from companies like Meta, which spend millions on training AI models. Because DeepSeek’s techniques require significantly less computing power for training, this has resulted in lower costs. Additionally, it’s open-source, unlike the closed models from OpenAI and Google, which means other companies, especially small developers, can build on top of this model and improve it without paying license fees. The potential is huge. Instead of developing their own models, companies can modify and deploy DeepSeek’s models at a fraction of the cost. And this could drive the mass adoption of AI at scale.
But what’s also helping DeepSeek is its lower API cost, which makes cutting-edge AI models more accessible to small businesses and companies that may not have huge budgets or the tech know-how to deploy proprietary solutions.
Story continues below this ad
DeepSeek’s strategy of using open-source models can have a huge impact on the AI community at large, opening up the AI market and providing access to AI tools for a broad set of users, especially smaller businesses. Because the company is committed to an open-source approach, it can also improve the trust factor and bring accountability to AI development. However, many are suspicious about the timing of the launch of DeepSeek’s R1 model, especially at a time when Donald Trump had just become president of the US. The timing may suggest that China is sending a signal that its AI development is on par with what the US has achieved, despite larger tech companies with infinite resources and talent. But many also question whether DeepSeek’s models are subject to censorship to prevent criticism of the Chinese Communist Party, which poses a significant challenge to its global adoption. At the same time, the rise of DeepSeek and China’s growing presence in the AI landscape also raises the question of where India stands, especially without the presence of an AI lab or startup that matches the capabilities of OpenAI or DeepSeek.
