Gordon Moore’s famous law observed that the number of transistors on a chip doubled roughly every two years, leading to a relentless decrease in the cost of computing. DeepSeek’s latest release suggests we may be seeing the emergence of a new “Moore’s Law” for AI costs, with the price of intelligence dropping at a similarly rapid and predictable rate.
The engine driving this new law is not silicon, but architectural innovation. DeepSeek’s Sparse Attention is a prime example of a design breakthrough that dramatically increases the “efficiency density” of an AI model, allowing it to do more computational work for less cost.
The 50% price cut is the first major data point for this new law. It’s a sudden, step-change reduction in the cost of AI, suggesting a pace of price decline that is far faster than most had anticipated. If this trend continues, the cost of advanced AI could halve every 18-24 months.
This has profound implications for the future. Just as the original Moore’s Law enabled the PC revolution and the internet, a Moore’s Law for AI costs would unleash a torrent of innovation, making it cheap enough to be embedded in virtually every product and service.
DeepSeek, by championing and enabling this rapid cost decline, is positioning itself as a leader in this new era. Its “intermediate step” is a promise that the pace of innovation will continue, ensuring that the cost of artificial intelligence continues its relentless and world-changing march downward.