Bi/multilingual learners in U.S. classrooms navigate the dual challenge of learning academic content while acquiring a new language. Artificial intelligence-based tools (e.g., those powered by Multilingual Large Language Models (MLLMs)) can facilitate new forms of languaging in these classrooms. We ran a survey on teachers of bi/multilingual learners (n = 232) across the U.S. and analyzed their perceptions of AI using a mixed-methods approach. We found that teachers desire multilingual abilities and feel positively towards AI to promote engagement, personalized support, and equity. They worry about the dangers of overreliance, mistakes/misinformation, and the loss of language nuance when using AI for bi/multilingual learners. Educators are divided on the potential social isolation caused by AI and confidence in supporting their students in a foreign language. Contextual factors (e.g., language background, grade level) also affect teachers’ views on AI. These findings can inform the future development of AI used to support bi/multilingual students. The fears, desires and contextual factors can indicate what changes and designs may be most impactful in different classrooms.