Open-source artificial intelligence (AI) models are gaining traction among businesses as tighter regulatory restrictions and slower releases of frontier AI systems drive demand for more affordable and accessible alternatives, according to a new research note from Citi.
The investment bank said enterprises are increasingly turning to open-weight and open-source AI models as they contend with the rising costs of proprietary systems and growing limitations on access to some of the industry’s most advanced models.
Citi noted that the performance gap between proprietary and open-source models has narrowed significantly following the recent launch of Z.ai’s GLM-5.2 model. The development, it said, has strengthened the case for businesses to deploy open-source AI solutions for a range of specialised applications.
According to Citi’s latest model rankings, several open-source systems developed by Z.ai, DeepSeek, MiniMax and Moonshot AI are now competing more closely with proprietary offerings in terms of intelligence, while also delivering greater cost efficiency.
The bank highlighted accelerating growth among companies focused on open AI architectures. It said AI startup Cohere has tripled its internal annual recurring revenue projection for 2027, while OpenRouter’s share of open-source tokens processed increased to 65 per cent in June, up from 34 per cent in January.
Fireworks also recorded strong growth, with its open-source token volume roughly doubling between April and June to reach 30 trillion tokens, according to industry data cited by Citi.
The report also pointed to increased investor interest in technologies designed to reduce the cost of AI deployment. Citi cited Baseten’s US$1.5 billion funding round, Upscale AI’s US$190 million capital raise, and Qualcomm’s planned acquisition of Modular as evidence of growing support for open and flexible AI software and hardware ecosystems.
Despite the momentum behind open-source AI, Citi warned that the sector continues to face significant challenges. The bank identified shortages of skilled AI researchers and engineers, alongside infrastructure constraints, as major obstacles to sustained growth.
It also noted that opposition to new data-centre developments is becoming more widespread across the United States. More than 300 local data-centre bans or moratoriums have been enacted since 2023, with over 275 introduced since the beginning of this year.
Citi said these constraints are likely to benefit established AI companies with extensive talent pools and greater access to computing infrastructure, even as open-source models continue to expand their share of the market.
