Software Firms Shift to Open-Weight AI Models

PYMNTS reports that software firms are increasingly replacing proprietary U.S. tools with open-weight models from Chinese labs, driven by rising costs for proprietary subscriptions. The transition allows companies to build on their own data, reducing reliance on third-party vendors. However, this approach is not feasible for all organizations, as it requires high upfront capital, specialized personnel, robust computer infrastructure, and substantial proprietary data.

Data privacy remains a concern when adopting Chinese models, with clients citing specific worries about information handling. Even firms that successfully deploy their own open-weight stacks often revert to large AI lab products when they need the highest tier of capability. Large labs have shifted from flat-rate subscriptions to token-based billing as consumer usage moves from chatbots to autonomous agents. Chinese labs charge less because their models run more efficiently and regional energy costs are lower.

AT&T cut costs for coding and advanced AI tasks by up to 56% by directing employee requests to less expensive models when suitable. The company stated that this method reduced performance quality by only 2%. AT&T intends to increase the portion of requests handled by open-source models from 40% today to a range of 60% and 70% in future years. For middle market companies, the choice balances savings and control against the burden of managing infrastructure, including computing capacity, storage, and cybersecurity controls.