Corporate America is shifting from expensive proprietary AI models to open-weight alternatives to curb rising token costs and secure proprietary data. AT&T is leading this transition, routing its 45 billion daily tokens to achieve cost savings of up to 90% in key applications. While highly efficient Chinese models from developers like Moonshot have surged in popularity, security concerns are driving demand for American open-weight alternatives from tech giants like Meta and Nvidia.
Corporate open-weight AI adoption
- ▪Gartner analyst Chirag Dekate projects that open models will underpin more than 50% of business AI use cases within two years, up from less than 10% in August 2026.
- ▪Data from fintech provider Ramp released on August 12, 2026, shows that 6.1% of businesses spending on AI used platforms offering open-weight and Chinese-developed models in July 2026, up from 4.5% a year earlier.
- ▪Corporate America is increasingly substituting proprietary artificial intelligence models with open alternatives to reduce operational costs.
Chinese AI models competition
- ▪American companies view Chinese AI models with suspicion due to concerns regarding data security, training biases, and the fact that they are not fully open source.
- ▪Running Chinese open-weight models on American cloud platforms helps mitigate data-security concerns by keeping customer prompts and information within the cloud instead of sending them to Chinese developers.
- ▪Open artificial intelligence models from Chinese firms, including Moonshot and Z.ai, have gained significant popularity in Silicon Valley due to their low cost and code-generation capabilities.
AT&T token cost reduction
- ▪AT&T Chief Data and AI Officer Andy Markus stated that open models power about 25% of the telecommunications company's overall AI usage, with plans to increase that share to 70% or 80% over time.
- ▪Switching from closed, proprietary AI models to open models has resulted in cost savings of 80% to 90% for AT&T in certain applications.
- ▪AT&T customized an open model called OTel using telecom-specific data to support AI agents that detect the root cause of network issues.
- ▪AT&T uses an average of 45 billion AI tokens each day and built a 'smart router' to automatically direct user prompts to the most cost-effective model.
Open versus proprietary models
- ▪General Catalyst managing director Marc Bhargava expects proprietary models from OpenAI and Anthropic to retain an edge for the most demanding tasks, such as coding.
- ▪Proprietary models from companies like OpenAI, Anthropic, or Google are not freely available for users to download and modify, whereas open-source and open-weight models are.
- ▪True open-source models allow full access to training data and code, while open-weight models typically share only the underlying numerical parameters, or weights.
Meta Nvidia open model push
- ▪Nvidia released a small open system and is developing a larger Nemotron-4 open-source model to rival leading open-weight models.
- ▪Meta Platforms announced a new open-weight model called Muse Glimmer on August 10, 2026, and plans to release the weights of its advanced Muse Spark 1.2 model.
- ▪Meta Platforms plans to spend up to $145 billion on AI infrastructure in 2026 to scale its computing capacity.
Story comments
Loading comments…