After years of US embargoes aimed at stifling Chinese advancements in artificial intelligence, the technology is re-emerging stronger than ever. The latest iteration, known as DeepSeek R2, is anticipated to revolutionize the industry once more, much like its predecessor that impacted Western stock markets significantly.
According to insider leaks, DeepSeek R2 boasts an impressive 1.2 trillion operations per second, effectively doubling the computational power of its previous version. This positions it as a formidable competitor against industry giants like GPT-4 Turbo and Google Gemini 2.0 Pro. Moreover, cost efficiency is a key advantage of DeepSeek R2, with operational costs slashed by 97 percent. Specifically, the cost per million tokens for input is $0.07, while production costs are marked at $0.27 per million tokens, rendering it the most affordable AI model currently available.
In a strategic shift, DeepSeek R2 is no longer reliant on Nvidia’s accelerators. Instead, it predominantly utilizes the Huawei Ascend 910B accelerator for 82 percent of its operations. This move not only circumvents recent US embargoes but also ensures robust performance and vertical integration, marking a new chapter in AI technology.
Yorum Yap