Major US corporations are shifting toward open-weight AI models as they seek lower-cost alternatives to closed, flagship systems from providers such as OpenAI and Anthropic. The move is driven by rapidly rising IT costs and the desire to avoid per-token billing on third-party platforms, the Financial Times reports.
Open-weight models expose their parameter files publicly, enabling companies to download, run and fine-tune them on their own hardware without paying token-based fees to an external developer.
Measurable uptake across reports and platforms
Data from research platform AlphaSense show that references to open-weight or open-source models in corporate earnings reports and investor presentations rose sixfold in August–September 2026 compared with the same period a year earlier.
Platform metrics also reflect the shift: Vercel’s AI Gateway reported that open-weight models accounted for 56 percent of tokens processed in August 2026, up from just 7 percent in December 2025. On the OpenRouter platform, Chinese-developed open-weight models dominate the top ten by tokens processed.
Industry examples and concrete figures
The trend reaches beyond pure technology firms into finance, telecommunications and other sectors. AT&T already runs about 40 percent of its AI workload on open models and aims to reach 70 percent within a year. Andy Markus, AT&T’s head of data and AI, said the company processes 45 billion tokens per day, and at that scale costs become a critical factor. AT&T fine-tunes open models using its own data, achieving performance comparable to or better than closed alternatives for specific tasks.
Scott Wallace, solution architecture lead at data-center operator Digital Realty, said the firm built an internal chat system on open-weight models and never sends customer data to closed, external top-tier models. Depending on task sensitivity and complexity, they employ a mixed approach; besides cost savings, data security and data sovereignty are key considerations.
Pressure on closed-model providers
This migration to open models poses a direct threat to revenue growth for companies that rely on closed models. Anthropic is preparing for an IPO with cited valuations of at least $2 trillion in some reports, while OpenAI has been linked to a private funding round valuing it at about $1.2 trillion. Both companies released cheaper versions of their flagship models last week, signaling an intensifying price war with Chinese rivals.
Why it matters
Switching to open-weight models addresses immediate economic pressures around high-volume token processing, reduces dependency on per-token pricing, and gives companies more control over data handling and sovereignty. These practical advantages are reshaping procurement and supplier strategies across industries.
(This article was prepared with assistance from an AI; the final text was edited and verified by our journalist.)
