AI Model Competition May Resemble Apple vs. Android

The ongoing debate surrounding open-weight and closed artificial intelligence (AI) models is often framed as a zero-sum competition, but a more nuanced outcome is likely, according to Deutsche Bank analyst Adrian Cox. Cox draws historical parallels to illustrate how different technological approaches can coexist and thrive, suggesting the AI landscape may evolve more like the Apple versus Android mobile operating system battle than a definitive winner-take-all scenario such as VHS versus Betamax.

Cox's analysis, detailed in a note last week, contrasts the current AI model competition with past technological conflicts. He references the 1880s debate between direct current (DC) and alternating current (AC) during the nascent stages of electrical technology, and the late 1970s and early 1980s format war between VHS and Betamax videocassette recorders. In the case of VHS and Betamax, despite Betamax being considered technologically superior, VHS ultimately prevailed due to JVC's broad licensing strategy, which fostered a robust and self-reinforcing ecosystem around its format. This historical precedent suggests that an AI model's success may not solely depend on peak performance but also on factors like accessibility, cost, and the development of a supportive ecosystem.

The implications of the open versus closed AI model debate extend to business, politics, and safety. Proponents of open models argue they can capture market share from closed rivals such as Anthropic and OpenAI, potentially reducing demand for specialized AI chips. Concerns have also been raised by U.S. AI companies regarding the rapid performance improvements of Chinese open models, which could translate into a military advantage. These companies also accuse China of illicit technology acquisition through "distillation." Conversely, critics of open models highlight the potential risks associated with their lack of stringent safety guardrails, which could facilitate the development of dangerous weapons, including nuclear, biological, or cyber weapons.

However, closed models also present their own set of challenges and potential downsides. The implementation of safety regulations could inadvertently stifle competition, thereby protecting providers of closed models. Furthermore, restricting access to the most advanced AI models can expose users to regulatory risks. An example of this occurred when the U.S. government temporarily banned Anthropic's most powerful Claude models for several weeks due to security concerns. Cox posits that instead of a single format dominating, the AI market is more likely to see a blend of both open and proprietary approaches. He predicts that open-source AI does not necessarily need to surpass closed rivals in performance; being "good enough," more affordable, and widely available could be sufficient for its widespread adoption and success, mirroring the dynamics seen in the evolution of other major technological platforms.