Recently, NVIDIA announced the Rubin platform at CES, with Jensen Huang calling it the most aggressive chip upgrade in history. Just looking at the hardware metrics is enough to be shocking: 6 chips working together, inference performance skyrocketing to 5 times that of Blackwell, and training performance increasing by 3.5 times. But what is the real killer feature? Inference costs are directly cut by 90%.

What does this mean for the entire crypto ecosystem? The cost of computing power drops significantly, directly lowering the threshold for AI application deployment.

From a crypto perspective, at least three tracks are worth paying attention to. First is the AI Agent direction (projects like $VIRTUAL, $FET), where cheaper computing resources mean more on-chain AI applications can run; second is decentralized computing networks ($RENDER, $AKT), which are not competitors to NVIDIA but rather complementary; third is the data storage sector ($WAL, $FIL), as AI model training relies on vast amounts of data supply.

But here’s a thought-provoking paradox: the more powerful NVIDIA becomes, the higher the degree of centralization in chip supply. Can the story of decentralized computing power continue? The answer may depend on the performance of these projects in 2026.