This might lead to duopoly in the market for hardware machine learning chips with Intel and Nvidia competing-with/acquiring new players. This could ultimately stifle innovation like it happened with EDA companies - Synopsys and Cadence.
Novel ML implementations have a lot of R&D overhead and need a (comparatively) lot of incubation before they are effective. Tying that to a commercial success adds magnitudes of complexity that 99% of Venture doesn't like.