AI vendors and hyperscalers seem wedded to their specific rack designs, despite working toward standardization.