DeepSeek has dropped the price of its AI services by 75%, a move that's already shaking up the competitive landscape and putting new pressure on companies like Anthropic. The cut, which took effect this week, is one of the steepest in the industry's short history.
The price drop and what it covers
DeepSeek didn't detail which specific products or tiers are affected, but the company confirmed the 75% reduction applies across its core AI offerings. The move comes as the AI arms race intensifies, with providers racing to attract developers and enterprise customers. By slashing prices so aggressively, DeepSeek is betting that volume and market share gains will outweigh the immediate revenue hit.
Why Anthropic is feeling the squeeze
Anthropic, a rival AI firm known for its safety-focused models, has seen its valuation come under pressure as investors reassess the economics of the sector. The 75% cut from DeepSeek makes it harder for competitors to justify premium pricing, especially when many customers are still testing which platform to commit to long-term. For Anthropic, which has raised billions and is valued in the tens of billions, the price war raises questions about how quickly it can turn a profit.
The impact isn't limited to Anthropic. Other AI companies, including those offering similar large language models, are now facing a choice: match the cut, innovate to justify a higher price, or risk losing customers. The market's response has been a flurry of internal reviews, though no major competitor has announced a matching price cut yet.
What happens next
DeepSeek's move is fresh, and the full fallout isn't clear. Anthropic hasn't commented publicly on the price cut, and its investors are likely weighing whether the company's focus on safety and alignment can command a premium in a market that's rapidly commoditizing. The next few weeks will show whether other players follow DeepSeek's lead, or try to differentiate on features rather than price. Either way, the cost of AI is coming down—and for some companies, that's a harder problem than any technical challenge.
