It's possible with central tracking servers if you de-hype-ify that number to say <500ms. With a decentralized network it's not possible. Even trying to achieve this would result in completely insane exponential bandwidth overhead and consummate issues around vulnerability to amplification DOS attacks.

This is why we have no huge-scale meshnet. As the network's size increases the bandwidth required to maintain the network's routing tables increases exponentially, not linearly, until no bandwidth remains for actual traffic.

If this is solvable it will either require a creative redefinition of the problem or a fundamental innovation in mathematics, probably in the realm of graph theory.