The AI Power Struggle

The artificial intelligence industry is increasingly divided between two visions. One argues that AI should be open, widely accessible, and available for businesses, researchers, startups, and governments to build upon. The other believes that the most advanced systems should remain proprietary, controlled by companies able to invest billions of dollars in developing and securing them. Last year, ARC’s Vikram Kalkat wrote on this topic: Is Closed AI Holding Us Back? Why Open Source Might Win the Race; but now the debate has come into sharp focus. 

On July 24, 2026, more than two dozen technology companies and organizations, including Nvidia, Microsoft, Meta, IBM, Hugging Face, Mistral, Palantir, Mozilla, and the Linux Foundation, signed an open letter titled “Open Weights and American AI Leadership.” The signatories urged policymakers not to restrict open AI models and argued that U.S. leadership will depend on an open ecosystem that spreads innovation across the economy rather than concentrating it in a few organizations.

The issue is about more than technology. It is a debate about competition, costs, access, safety, and who ultimately captures value in the AI economy.

Closed AI versus Open Source AI 

Closed AI models are developed and controlled by individual companies. Users access them through APIs or subscriptions, but the underlying technology remains proprietary. Companies such as OpenAI, Anthropic, and Google largely follow this approach with their most advanced systems. The provider controls deployment, upgrades, and pricing.

Open AI takes a different approach. Open-weight models can be downloaded, modified, fine-tuned, and deployed on an organization’s own infrastructure. This lowers barriers to adoption and gives users more control over how AI is integrated into their workflows. While many “open source” AI models are technically open-weight rather than fully open-source, the broader principle is the same: advanced AI capabilities become available to a wider range of users.

Supporters say openness encourages innovation, fuels competition, reduces dependency on a few vendors, and spreads AI across industries and geographies. Critics worry about security risks, misuse, and whether open models could weaken incentives to fund expensive frontier research.

Why Model Weights Matter

Model weights are the learned parameters inside an AI system. A simple analogy is to think of source code as the recipe, training data as the ingredients, and model weights as the finished meal. Access to weights allows developers to run a model independently, adapt it for specialized uses, and build products without starting from scratch.

That is why the July 2026 letter focused so heavily on open models. The signatories argue that broad access to advanced AI creates a foundation on which thousands of innovators can build, much as open-source software helped create today’s internet economy.

The Power of Distillation

Another force shaping the market is distillation, a technique that allows a large model to teach a smaller one. The smaller model learns from the larger system’s outputs and can often perform many of the same tasks while using far less computing power.

This matters because most businesses do not need the most advanced AI model for every task. Frontier models may be necessary for complex reasoning or scientific research, but smaller distilled models can often handle customer support, document summarization, workflow automation, and knowledge retrieval at a fraction of the cost. As distillation improves, the gap between cutting-edge models and smaller specialized systems narrows, increasing pricing pressure across the industry.

Why Token Costs Matter

For all the attention given to model capability, AI’s future may be decided by economics. Every AI interaction consumes tokens, the units used to process and generate language. Technology companies spend enormous sums training foundation models, but they also face ongoing inference costs every time users interact with those systems.

As AI adoption scales from millions to billions of daily interactions, token costs become increasingly important. Open-model advocates argue that organizations should be able to choose the most cost-effective model for each task rather than pay premium prices for every interaction. Open models can be deployed locally, customized for business needs, and optimized to reduce operating expenses.

The Case for Open AI

The coalition behind the July 2026 letter argues that open AI is critical to maintaining American leadership. In its view, open models expand access, encourage competition, lower costs, reduce vendor lock-in, and give organizations greater control over their technology and data. It also argues that leadership should not be defined by one dominant model, but by how widely AI becomes embedded in businesses, schools, hospitals, factories, farms, and public institutions.

The letter also challenges the assumption that closed systems are automatically safer. It argues that concentrating advanced AI within a few proprietary platforms can create single points of failure, while open ecosystems benefit from broader testing, auditing, benchmarking, and security research.

Ed Zitron’s Bear Case for AI

Technology analyst Ed Zitron has become one of the strongest critics of current AI market optimism. His argument is not that AI lacks utility. Rather, he questions whether the enormous capital expenditures flowing into chips, power, data centers, and infrastructure can generate sufficient financial returns.

Zitron’s concern becomes sharper if open models keep improving. If powerful AI capabilities become widely available and distillation keeps lowering costs, premium pricing may be harder to maintain. What is good for adoption may not always be good for margins.

The Bigger Question

Both sides may be right. Advocates of open AI argue that innovation and economic growth thrive when powerful technologies become widely available. Zitron asks where the profits will come from if AI becomes abundant, accessible, and increasingly commoditized.

The future of AI may not depend simply on whether models are open or closed. It may depend on where value ultimately settles: in the models, in the infrastructure that runs them, or in the applications and workflows built on top. As AI matures, the winners may not be the companies with the biggest models, but those that deliver the greatest value at the lowest cost.