The point is that 100nm GaAs is already peforming at a faster speed than the fastest silicon available. How many billions and years has Intel spent on their 14nm Si Broadwell which is still not in production?

The Poet process is providing the industry with a new much cheaper Moore's Law roadmap. GT is saying they can go down to 40nm in their own lab. Intel needs 10 billion dollar fabs to reduce by another 5-10 nm. The big speed increases in GaAs will not just be dependent on size but also Poet's ability to integrate optics into the same die. That just doesn't exist in a workable manner in silcon.