Major U.S. corporations are rapidly shifting away from expensive proprietary AI systems toward open-source and open-weights models, driven by dramatic cost savings and improved performance from competitors like Meta and Chinese startups. AT&T has increased its use of open models from 20% to 40% of its AI workload in just months, with plans to reach 60% soon, while saving up to 80% on AI costs compared to earlier in the year.

Why Are Companies Abandoning Expensive AI Models?

The shift reflects a fundamental change in how corporate America approaches artificial intelligence. When AT&T's chief data and AI officer Andy Markus examined the company's AI spending, he discovered that proprietary models from companies like OpenAI and Anthropic, which charge subscription fees and keep their underlying code private, were becoming prohibitively expensive as AI adoption scaled across the organization.

Open-source and open-weights models offer a compelling alternative. These systems allow developers to download the underlying code or numerical calculations (called "weights") that power the AI without paying licensing fees or seeking approval from the model creators. Instead, companies pay primarily for the computing servers needed to run the models. This architectural difference has made open models dramatically cheaper while their performance has caught up to closed alternatives.

The trend extends far beyond AT&T. Companies including Airbnb and Deloitte are also adopting open AI models for their flexibility and cost-effectiveness. According to data from OpenRouter, a platform that lets users select different AI models for tasks, open models accounted for 58% of AI use last month, up from just 10% a year ago.

How Are Open Models Performing Against Premium Alternatives?

The performance gap between open and closed models has narrowed dramatically. Some Chinese open models, including those from companies like DeepSeek and Alibaba, now deliver 80% to 90% of the capability of leading closed models from OpenAI and Anthropic while costing as little as 20% of the price, according to Jerry Tang, CEO of Atlas Cloud, a startup providing access to various AI models.

"Mazdas do a fine job getting you where you need to go, and Maseratis are simply not worth the investment," Tang said, comparing premium closed models to luxury sports cars. "Open-source models offer a better bang for your buck."

Jerry Tang, CEO at Atlas Cloud

The performance improvements stem partly from technological breakthroughs. DeepSeek's open model, released last year, stood out for being nearly as advanced as leading closed models while requiring significantly less computing power to operate. Since then, open models from Chinese companies have increasingly matched the performance of closed systems from Anthropic, OpenAI, Google, and others.

Accessibility has also improved dramatically. Many open models can now run on a phone or laptop, according to David Stout, CEO of WebAI, an AI startup. This accessibility makes the technology cheaper and more practical for a wider range of applications.

How Are U.S. Companies Choosing Between Open and Closed Models?

Most U.S. corporations are not abandoning closed models entirely. Instead, they are adopting a hybrid approach, using different models for different tasks based on cost and capability requirements. Closed models remain superior for computationally intensive work like coding and image or video generation, while open models excel at simpler, specialized tasks.

At AT&T, the company is taking open-weights models and customizing them to build new tools for call transcription and customer service, demonstrating how enterprises can adapt open models to their specific needs. However, AT&T has chosen not to use Chinese open models due to concerns about regulation and data privacy, instead working with alternatives from U.S. companies like Google's Gemma and Meta's Llama models.

Steps to Evaluate Open-Source AI for Your Organization

• Assess Your Use Cases: Determine which AI tasks are mission-critical and require maximum performance versus routine tasks where good-enough performance suffices, since closed models excel at complex work while open models handle specialized applications efficiently.
• Calculate Total Cost of Ownership: Compare not just licensing fees but also the cost of computing infrastructure, customization, and maintenance, as open models may require more technical expertise but offer significant savings on per-use costs.
• Evaluate Regulatory and Privacy Requirements: Review your industry's compliance obligations and data handling policies, as some organizations may prefer open models from domestic companies over international alternatives due to regulatory concerns.
• Test Model Performance: Run pilot projects with candidate models on representative workloads to measure real-world performance, latency, and accuracy before committing to large-scale deployment.

What Does This Shift Mean for AI Giants?

The rise of open models poses a significant challenge to OpenAI and Anthropic, both of which are preparing for major initial public offerings. These companies depend on subscription revenue and per-use fees to fund billions of dollars in research and development and computing infrastructure. As enterprises increasingly adopt cheaper open alternatives, the companies face pressure to justify their premium pricing.

The trend also reflects a broader philosophical debate in Silicon Valley and Washington about who should control AI technology. Dario Amodei, CEO of Anthropic, has argued that AI should be tightly regulated to address national security concerns. In contrast, tech leaders like Nvidia CEO Jensen Huang and Meta CEO Mark Zuckerberg have championed open-source technology as essential for a democratic, balanced AI ecosystem.

Meta has positioned itself strategically in this landscape by converting its leading AI system to an open-weights model and offering it at lower cost than other top U.S. models. Zuckerberg stated: "I do not believe restricting access to foreign open-source models is an effective solution. Our goal should be for American open-source models to be the best globally".

Zuckerberg

The corporate shift toward open models accelerated after Chinese startup DeepSeek released its advanced open model last year, demonstrating that open-source systems could match closed models in capability. This breakthrough prompted other U.S. companies to develop their own open alternatives, intensifying competition and driving down costs across the industry.

Nvidia's announcement that it is acquiring Hugging Face, a major library of open AI models, for $12.9 billion underscores the strategic importance of open-source AI infrastructure. The deal signals that even hardware giants view open models as central to the future of AI deployment.