Just a month after sending shockwaves through tech inventory and wiping out more than $1 trillion in US stock market value, Chinese AI startup DeepSeek is already preparing to do it again. Now, an exclusive report from Reuters suggests that the company could arrive even earlier in less than three months after releasing its next-gen R2 model.
Deepseek is moving fast. After disrupting the global market with the budget-friendly inference model R1, the Hangzhou-based company is now accelerating plans for its successor, R2. According to a Reuters report, the sources say the company is pushing for previous releases. If R1 thinks that the internet is burning and everyone is talking, R2 is ready to make an even greater impact. The new model aims to enhance coding capabilities and extend reasoning skills beyond English, making it an even bigger competitor in the AI race.
The Hangzhou-based company doesn’t share details, but insiders say the goal is to improve coding capabilities and extend reasoning skills beyond English. Deepseek remains silent, but expectations are rising.
“The Hangzhou-based company is accelerating the launch of its successor to the R1 model in January, according to three people familiar with the company. Deepseek will release the R2 in early May. We had planned it, but now we want it as soon as possible, two of them said without providing details,” Reuters reported.
R1 caught the industry off guard. It was built with less powerful Nvidia chips and was able to compete with models developed by the US tech giant at much more cost. If R2 fulfills its promise, the impact could be even greater.
Vijayasimha Alilughatta, COO of Indian technical services provider Zensar, considers this a turning point. “The launch of Deepseek’s R2 model could be a pivotal moment in the AI industry,” he said. Deepseek’s approach to building a cost-effective model “is likely to spur companies around the world to accelerate their efforts. This is a tension among a small number of dominant players on the ground. I’ll break it.”
This could raise alarms in Washington, where AI leadership is a national priority. Meanwhile, Chinese companies have already integrated Deepseek’s model into their products, and Beijing appears to be paying close attention.
Deepseek’s AI rise
Deepseek’s success was a surprise to many AI experts in the West. Talking about Deepseek has grown after the V3 model surpassed Meta’s Llama 3.1, Openai’s GPT-4O, and Alibaba’s Qwen 2.5 on third-party benchmarks. In late January, Deepseek easily surpassed ChatGpt on the App Store.
Deepseek operates unlike most Chinese tech companies. The company does not follow the strict top-down management structure that is common in major companies in the country. Instead, it’s more like a lab than a traditional business. Employees explain an environment where young talent is given serious responsibility and working hours is not a requirement.
Liang Wenfeng, the founder of Deepseek, is a quiet figure in the Chinese tech scene. The former Quant Trader built his fortune through hedge fund high flyer before focusing on AI. He shuns media attention and describes his former colleague as being modest but deeply involved in his work. “Lian gave us control and treated us as experts. He always asked questions and learned with us,” said former researcher Benjamin Liu. .
The rise of Deepseek was not completely smooth. Chinese regulators were wary of the company’s large-scale chip purchases. Before it was founded, High-Flyer stocked up Nvidia A100 chips. This is an extraordinary move for hedge funds. Officials wanted to know why. After considering the plan, authorities decided to retreat. This is the result that proves important when the US later banned exporting A100 chips to China. By that time, DeepSeek had already had a strong computing infrastructure in place.
Industry response to Deepseek’s success
Deepseek’s success has led American tech CEOs to alarm over breakthroughs in China’s AI. At Davos’ World Economic Forum, a well-known CEO raised concerns about China’s rapid progress in AI. Deepseek is a frequent topic of discussion, with some leaders highlighting the potential geopolitical risk of falling behind this critical technology. ”
“If the US cannot lead this technology, we will be in a very bad geopolitical place,” warned one CEO.
Their comments highlight the growing interests of AI’s US leadership, an area that is increasingly tied to national security and global influence.
The well-known figures have placed emphasis on Deepseek’s achievements.
Venture capitalist Mark Andreessen calls it “one of the most amazingly impressive breakthroughs I’ve ever seen.” Journalist Holger Zschaepitz suggests that Deepseek’s efficiency could pose a major threat to the US stock market, undermining the usefulness of the billions spent on AI capital expenditures. Garry Tan, CEO of Y Combinator, saw the development as an opportunity for US competitors, saying that cheaper model training could accelerate demand for AI inference and real-world applications.
Strategic AI Development
Deepseek’s strategy is not to throw money into problems. Unlike the Chinese tech giants who rushed to launch ChatGpt-style apps, DeepSeek focused on improving the quality of their models first. Startups also use techniques such as mixtures (MOEs) and multi-head potential attention (MLA) to keep costs low. These methods make the model more efficient and require much less computing power than traditional architectures.
Other companies, including French Mistral, experimented with MOE, but Deepseek leaned forward more aggressively. result? A model that rivals those built on a significantly higher budget. Bernstein analysts estimate Deepseek’s prices are 20-40 times cheaper than Openai’s equivalent model.
The impact is already shown. Openai has recently cut prices, while Google’s Gemini has introduced a cheaper access layer. Since the R1 debut, Openai has also released an O3-MINI model that uses less computing power. This shift suggests that DeepSeek is already affecting AI pricing strategies beyond China.
The future of Deepseek
Beijing has embraced Deepseek’s success, but officials have kept the company on a tight string. Liang is asked to keep things inconspicuous to avoid attracting too much attention. Chinese government agencies and state-owned companies have begun rolling out their Deepseek models at a pace rarely seen in new AI companies. Meanwhile, some foreign governments, including South Korea and Italy, have removed Deepseek’s products from the app store due to privacy concerns.
With R2 on the way, Deepseek is trying to test how much disruption a single company will cause in an industry dominated by a few powerful players. Whether that leads to more regulatory scrutiny or forces them to rethink their competitors’ strategies, Deepseek’s impact cannot be denied. The only question I have right now is how far it can go.
Source link
