Citi has said open-source artificial intelligence models are gaining momentum, as regulators block access to some frontier models and slow the release of others.
The broker argued that demand for inference, the computing power needed to run AI models, continues to outstrip supply.
That is pushing enterprises to seek out alternatives to proprietary frontier models for specific use cases.
Citi said the intelligence gap between proprietary and open-weight models had compressed over the past two weeks.
It pointed to the release of Chinese developer Z.ai's GLM-5.2 model as a driver of that shift.
The bank cautioned that benchmarks for measuring the value of these models to businesses remain limited.
Citi said the move towards open models was fuelling investment in startups offering cost-efficient, open architectures across both software and hardware.
While infrastructure deals continue to cement the industry's long-term direction, the broker flagged constraints that will shape the pace of development.
It said talent shortages and challenges in building data centres would weigh on progress.
That backdrop should favour leaders rich in both AI talent and computing power, Citi added.
The broker also expects companies positioned to benefit from these bottlenecks through pricing to continue outperforming.
