Story
September 7, 2026
Open AI Models Are Eating Into Corporate America’s Closed-Model Bill
Corporate buyers see open AI as a route to lower costs and greater control, while advocates of tighter safeguards and companies handling sensitive data still see limits to an anything-goes shift. The emerging consensus is not a clean break with closed models, but a more competitive, mixed market.
Since ChatGPT’s arrival in 2022, corporate America largely followed the path of closed AI systems from OpenAI and Anthropic: powerful tools, sold at a price and with their underlying code kept private. Open-source alternatives remained largely a developer’s pursuit.1
That began to change last year, when China’s DeepSeek released DeepSeek-V3, a model that approached the capabilities of leading closed systems while using less computing power. As open models improved, the business case sharpened: companies could download and modify them rather than pay subscription and usage charges.2
AT&T’s shift captures the turn. Facing soaring AI costs this year, the telecom company moved toward open models for customer service, call transcription and coding. Open models represented 20% of its AI use in May; that share has since reached 40% and could rise to 60%, according to its chief data and AI officer, Andy Markus. He said the company had cut AI costs by as much as 80% from earlier in the year: “We believe it could go much, much higher.”1
The change is broader than one company. Airbnb and Deloitte are also adopting models that are cheaper and easier to tailor. On OpenRouter, open models accounted for 58% of U.S. usage last month, up from 10% a year earlier. Nvidia’s $12.9 billion purchase of Hugging Face underscored the stakes.2
For open-model proponents, the appeal is plain. Atlas Cloud’s Jerry Tang said some Chinese models deliver 80% to 90% of the performance of OpenAI and Anthropic systems for roughly 20% of the cost. “A Mazda can get you to your destination just fine,” he said, while open source is “more cost-effective.”2
Yet the market is not abandoning closed AI. Companies still favor those systems for heavy coding and image or video generation, while regulatory and data-privacy worries make some hesitant to use Chinese models. AT&T is studying them but using U.S.-developed alternatives such as Meta’s Llama and Google’s Gemma instead.2
The divide now reaches Washington and Silicon Valley: Anthropic’s Dario Amodei argues for strict controls over national-security risks, while Nvidia’s Jensen Huang and Meta’s Mark Zuckerberg frame openness as essential to a more democratic AI ecosystem. The result is a challenge to closed-model giants—not their immediate replacement.3