Companies across finance, logistics, and telecommunications are increasingly adopting lower-cost, open-source and open-weight AI models instead of proprietary systems. References to these models on US corporate earnings calls rose sixfold year over year in August and September, according to AlphaSense. Businesses are prioritizing cost savings, data control, and customization.
AT&T now runs 40% of its AI workloads on open-source models. It aims to raise that share to 70% within a year to manage rising costs. Vercel reported that open-weight models made up 56% of all tokens processed through its AI Gateway in August, up from 7% in December.
Many of the cheaper alternatives are developed by Chinese companies. Their growing adoption poses a competitive threat to revenue growth at major AI labs, including OpenAI and Anthropic.