TweakiOS

We are very close to the launch of the latest Chinese AI model, DeepSeek-V4. There is much speculation and some unconfirmed leaks about the efficiency of the upcoming model compared to the American models ChatGPT and Cloud. According to the leaks, the cost of input via DeepSeek's API will be $0.27 per million tokens. If this is true, it will be more than 50 times cheaper than its competitors, whose models cost $15 per million tokens. Of course, the cost varies due to the different operating mechanisms of the models. It is not only the tech community that is eagerly awaiting the release of the new model, but also the markets. The emergence of a model that offers competitive efficiency at such low prices will put significant pressure on American AI companies. The model is scheduled to launch this month, and there are speculations that it will be released in the coming days, coinciding with the start of the Chinese New Year.

This is the core message I’ve been wanting to deliver. It's all about Rotation. Even within the AI Data Center, capital follows a 'Value Chain of Constraints' to solve each emerging bottleneck. As we know, it got started with GPUs for compute power, then rotated to HBM to overcome the memory wall, and then to Energy as power became the primary limit. After addressing storage through NAND, the focus is now shifting to Photonics and Interconnects because copper has reached its physical distance and bandwidth limits. Ultimately, the money moves in a loop—from GPU to Memory, Energy, Storage, and finally to Photonics—because each sector must be unlocked to scale the next.