AI war won't be VHS vs. Betamax—more like Apple vs. Android

The AI industry has become consumed by a debate over whether the future belongs to open-weight models—whose weights (internal parameters) are publicly available and can be freely modified—or to closed, proprietary models. This debate spans business strategy, political sovereignty, and safety governance, with valid arguments on both sides.

The case for open models rests partly on their potential cost advantage and ubiquity. Lower-cost open alternatives could take market share from closed rivals like Anthropic and OpenAI and reduce demand for expensive semiconductor chips. Open models also promise broad ecosystem development—similar to how open-source software in the 1980s lowered costs and built a "shared foundation of knowledge," in the words of a letter signed by Nvidia, Google, Meta, Microsoft, and OpenAI. However, critics of open models raise serious concerns: they warn the models lack adequate safety guardrails and could be misused to develop nuclear, biological, or cyber weapons. Additionally, U.S. AI companies have accused Chinese firms of rapidly closing the performance gap with open models via illegal "distillation" (a technique for copying a model's behavior), and warn this could translate to a military edge.

Closed models have their own vulnerabilities. Regulatory restrictions can inadvertently protect incumbents and stifle competition, and tight control leaves users exposed to regulatory risk—the U.S. government effectively banned Anthropic's most powerful Claude models for several weeks over security concerns, highlighting how government intervention can disrupt access. Yet closed providers argue their safety guardrails and quality control justify proprietary restrictions.

Deutsche Bank analyst Adrian Cox reframed the entire debate by invoking historical parallels. He noted the battle between direct current and alternating current in the 1880s during the emergence of electrical technology, and the VHS vs. Betamax format war of the late 1970s and early 1980s. While Betamax was technically superior, VHS prevailed because JVC licensed the format broadly, creating a self-reinforcing ecosystem. Cox argues open-source AI does not have to beat closed rivals on performance alone; it can succeed by being "good enough, cheaper and ubiquitous." However, rather than a winner-takes-all outcome, Cox predicts the future will resemble today's smartphone market, where Apple's controlled premium ecosystem coexists with Google's larger, more customizable Android system. Both have found sustained success.

This coexistence model is increasingly supported by industry leaders. Nvidia CEO Jensen Huang explicitly stated on Friday that the world needs both frontier closed models and frontier open models. The letter signed by Nvidia, Google, Meta, Microsoft, and OpenAI echoed this position, asserting that U.S. AI leadership should be measured not by a single frontier model but by whether the country builds a "broad, open ecosystem." The letter acknowledged that open models carry security risks but argued the answer is maintaining access to them to bolster defenses rather than cutting off access. It emphasized that open weights strengthen competition across model developers, cloud providers, applications, and services, which "spurs innovation, drives down costs, and distributes the benefits of AI broadly across our economy." Cox also noted that open and closed AI will "exert a positive gravitational force on each other"—open models will pressure closed competitors on price and flexibility, while closed models will continue raising the bar on quality and safety.

The debate over open-weight and closed AI models has consumed the tech industry, but it hinges on contrasting visions of how the market will evolve. On one side, open models promise lower costs and broad accessibility, threatening chip demand and the competitive position of closed-model leaders like Anthropic and OpenAI. On the other side, closed providers argue that safety guardrails and quality control justify proprietary restrictions—a concern underscored when the U.S. government effectively banned Anthropic's most powerful Claude models for several weeks due to security issues. This regulatory risk cuts both ways: openness exposes vulnerabilities, but closure also invites government intervention.

Adrian Cox's historical comparison to the smartphone market offers a more nuanced framing than the zero-sum narratives that have dominated. Rather than a format war with a single victor, he suggests open and closed models will exert mutual pressure: open models will force closed competitors to lower prices and remain flexible, while closed models will continue setting performance and safety benchmarks. This coexistence is already supported by powerful voices in the industry—Nvidia CEO Jensen Huang explicitly endorsed the need for both frontier closed and frontier open models, and the letter signed by Nvidia, Google, Meta, Microsoft, and OpenAI reframed open-source development not as a threat but as a cornerstone of American technological sovereignty and innovation. The letter emphasized that competition across model developers, cloud providers, applications, and services drives down costs and distributes benefits broadly.

No comments yet. Be the first to share your thoughts!