US companies are abandoning expensive frontier AI in favour of open weights. A rapid shift is under way toward running models on in-house infrastructure and substantially reducing token-based charges.
US companies are rapidly moving away from high-performance frontier AI models from OpenAI and Anthropic and towards open-weight models that cost far less. According to Financial Times reporting, companies across industries including finance, logistics and manufacturing are considering the use of open weights.
Open-weight models make their weights public, allowing companies to run and fine-tune them on their own hardware. They are considered capable of reducing operating costs compared with closed models.
References to them in corporate earnings increased sixfold, suggesting that moves toward adoption are beginning to lead to actual implementation. Tinder’s increase in AI spending and AT&T’s expansion of its share of open models are taking place against a backdrop of cost management and data sovereignty.
The focus on data security is also strengthening. Digital Realty built an internal chatbot on the assumption that it would run open weights on its own infrastructure. The trend of combining multiple open-weight models for a range of business operations is spreading.