A New York Times report describes a broader shift among large US companies toward open-source AI models, moving away from exclusive reliance on proprietary offerings from firms like OpenAI or Anthropic. The trend is driven by a combination of factors: the need to control inference costs at scale, a desire to keep sensitive data in-house, and the flexibility to fine-tune models for specific business needs.
Open-weight models from providers such as Meta, Mistral AI, and Chinese labs including DeepSeek and Alibaba are increasingly being run directly on corporate infrastructure rather than accessed through third-party APIs. This deployment model reduces dependence on a single vendor and limits data exposure to external servers, a particularly sensitive concern in regulated industries like finance and healthcare.
The shift does not amount to an abandonment of closed models. Many companies are adopting a hybrid strategy, reserving the most advanced proprietary systems from OpenAI or Anthropic for tasks requiring the strongest reasoning capabilities, while routing routine or data-sensitive workloads to open models that are cheaper to run at volume. This segmentation reflects a maturing enterprise AI market, where model choice increasingly reflects economic and strategic trade-offs rather than raw performance alone.
The report also notes that this trend could reshape the balance of power between closed frontier labs and the open-source community, which is gaining traction as industrial players seek to diversify away from a handful of dominant providers. It remains to be seen whether this momentum will translate into greater investment in the US open-source ecosystem, given that several of today's most competitive open models originate from Chinese labs.