DeepSeek, Which Triggered 'AI Price Cuts', to Raise Prices

China's DeepSeek, which sparked a wave of cost reductions in the artificial intelligence (AI) industry, is set to implement sweeping price hikes.

According to foreign media including Bloomberg on the 6th, DeepSeek announced in a notice released the previous day that it would raise prices significantly, urging users to prepare accordingly.

DeepSeek did not disclose the specific margin of the price increase in the notice.

For its flagship "V4 Flash" model, DeepSeek currently charges $0.14 per 1 million input tokens (a unit of AI usage) and $0.28 per 1 million output tokens.

This is merely 0.5% to 1.4% of the rates charged by U.S. company Anthropic's latest "Fable 5" model, which costs $10 for 1 million input tokens and $50 for 1 million output tokens, respectively.

The "Kimi K3" model from Moonshot AI, a Chinese startup that recently gained attention in the global industry for high-efficiency AI, also costs $3 and $15 per 1 million input and output tokens, respectively, which is much higher than DeepSeek.

In January of last year, DeepSeek introduced "DeepSeek R1," achieving extreme cost-effectiveness without expensive AI hardware. This shattered the market consensus that advanced AI development and operation must incur high costs, sending shockwaves through global stock markets.

Since then, DeepSeek has maintained unusually low prices for its services, triggering the "DeepSeek death zone" phenomenon, in which a large number of companies lacking price and quality competitiveness were weeded out of the AI industry.

Industry watchers analyze that DeepSeek's latest decision may indicate that its strategy of enduring low profitability to secure market share has reached its limits as the company grows.

Bloomberg noted that as DeepSeek prepares for an initial public offering (IPO) as early as within the year, the need to achieve profitability that meets investors' expectations has grown.

It is also highly likely that AI infrastructure costs have played a role in the price hike.

DeepSeek is pushing forward with the construction of a 1-gigawatt (GW) AI data center in China's Inner Mongolia Autonomous Region. Building a 1GW-class facility costs approximately $50 billion (71 trillion won) by U.S. standards.

While infrastructure costs such as land are generally lower in China compared to the United States, the financial burden is expected to be immense for DeepSeek, which is still at a startup stage.

Meanwhile, Bloomberg forecasted that if DeepSeek raises its service prices excessively, competing firms could operate DeepSeek's AI models independently and offer even lower prices.

Under its open-source policy, DeepSeek discloses the architectural structure and weight (computation pattern) data of products like V4 Flash to the outside, allowing them to be used freely.

Because many companies improve DeepSeek's AI and operate it on their own servers, rivals could exploit excessive hikes in DeepSeek's own pricing by offering these modified models as alternatives to absorb customers.

(Photo: Getty Images)
※ Please note: This article was translated by AI and may contain errors.