His point was about scalability within centralized (logical) services, which people often confuse or understand them as a single point of failure or not scalable. But that's not true, as he pointed out there.