Recently, the emergence of the Chinese AI startup Deepseek has created significant waves in the tech industry, particularly affecting companies like OpenAI. Deepseek introduced its R1 model, which is both highly efficient and powerful, causing a stir in Silicon Valley. This has led to concerns about whether companies like OpenAI are overvalued and if their substantial investments in computing power are justified.
Deepseek’s R1 model has reportedly caused internal conflicts at OpenAI. Employees are worried that OpenAI might fall behind Deepseek due to internal power struggles between its research and product departments. The latest model from OpenAI’s research team, known as o1, has received much public attention. However, some former employees believe that OpenAI has not prioritized its chat capabilities sufficiently.
A former OpenAI researcher mentioned that Deepseek used a similar reinforcement learning approach for its R1 model but implemented it with better data and a cleaner architecture. This has given Deepseek an edge, leaving OpenAI’s o1 model in an experimental phase.
Deepseek’s success raises questions about whether investors are overpaying for companies like OpenAI, which primarily focus on expanding data centers to enhance their AI models. OpenAI, led by Sam Altman, aims to raise an additional $40 billion, which would increase its valuation to an impressive $340 billion. A portion of these funds is intended for infrastructure expansion, a plan now under scrutiny due to Deepseek’s efficiency gains.
Despite US export restrictions on advanced AI chips, Deepseek managed to develop its R1 model, unsettling competitors and investors alike. The company demonstrated that AI models could be developed with less computational effort and more efficient data structures, challenging OpenAI’s current strategy.
In response to Deepseek, OpenAI made its own lighter and more resource-efficient model, o3-mini, available for free to all users. This move appears to be a clear attempt to compete with Deepseek’s freely accessible open-source models.
The pressure is mounting on US AI companies to improve the efficiency of their models. Deepseek has shown that it is possible to achieve this, prompting investors to push for similar advancements from US developers. However, OpenAI and other US companies benefit from not being subject to government censorship or considered security risks, which could be a significant advantage in the international market.
Italy has already blocked Deepseek due to privacy concerns, and France and Ireland are engaging in discussions with the Chinese startup for the same reasons. Despite the focus on efficiency, the expansion of data centers is likely to continue. As former OpenAI employee and AI expert Miles Brundage explains, while less computing power is needed per intelligence unit, the demand for more units persists due to ongoing scaling efforts.
