Listening to Nvidia CEO Jensen Huang's speeches at the San Jose GTC in March and Taiwan's Taipei GTC in June confirmed that Nvidia's power lies not in one GPU product but in its ability to constantly expand algorithms, software, semiconductors, networks, developers, and startups into one ecosystem. Huang repeatedly emphasizes the co-design of software and hardware, with the goal of 'token throughput per watt', generating more tokens with the same power. The bottleneck of AI lies in data movement rather than computation itself, as much more energy is consumed moving data between memory and the computing unit. Nvidia has been reducing this inefficiency by closely connecting GPUs, CPU, memory, and network. Korea is a memory semiconductor powerhouse, but that is not enough in the competition for AI hegemony. The government is in the right direction to invest in Sovereign AI by utilizing semiconductor excess tax, but concentrating large amounts of money in one or two places can create risks such as AI's 'overfitting'. Uncertain futures require portfolio strategies. The solution is a nationwide MPW, a significant expansion of the multi-project wafer program, to reduce the speed difference between algorithms and semiconductor production.