“Tree Dao predicts that within the next couple of years, some AI workloads will become multi-silicon, running on chips from various manufacturers instead of just NVIDIA.”