While chatting with Professor Yao Dong earlier, he mentioned an interesting point: open source is another form of talent monopoly.

He gave the example of compression algorithms. Current general compression algorithms, like LZ77, LZW, and Huffman Coding, are all open source. However, the top researchers capable of implementing these algorithms are mostly concentrated in a few countries. Other countries find it hard to catch up even in terms of finding people.


The field of large models is following this path too. Researchers are concentrated in China and the U.S., making it difficult for other countries to assemble teams of similar scale. Mistral is a rare exception, but 65% of its funding still comes from France. Looking at the top open-weight models, the top five have all been domestic models since last month. The latest ultra-large models only have openai-oss-120B left.


Some might argue that compression algorithms are not hot topics, and the example is not suitable. But the issue isn’t about whether something is trending; it’s about talent reserves. Once a field forms a technical barrier, latecomers can’t even get a ticket to enter.

This is a bit like industrialization. China’s industrialization path has objectively raised the entry barrier for others. It’s not that they can’t do it, but the cost is too high.

Some say that hiring top talent with money can solve border issues. But top talent isn’t isolated; they need an ecosystem. Labs, data, peer interactions—these aren’t things that can be moved just with money.

The realization of AGI in the future will likely only happen between China and the U.S. It’s not a prophecy, but a look at the current distribution of talent and resource investments. Other countries aren’t without opportunities, but the window is shrinking.

Open source promotes innovation, which is true. But it also silently sets the rules of the race. The players who can run on this track are always the same few.