Open vs closed AI models may both win, Deutsche Bank analyst says
Adrian Cox says the AI market may resemble Apple and Android more than a single-format tech war, with open and closed models coexisting.
By Daniel Okafor · Business Editor
3 min read
The fight over open vs closed AI models may not produce a single winner, according to Deutsche Bank analyst Adrian Cox. His argument matters because the outcome could shape AI prices, safety rules, chip demand and the balance of power between U.S. and Chinese technology companies.
The current debate spans business, national security and safety concerns. Supporters of closed systems warn that cheaper open models could pull customers from companies such as Anthropic and OpenAI, while also lowering demand for advanced chips.
U.S. AI companies also argue that Chinese open models are narrowing the performance gap with American systems, which they say could carry military implications. They have accused China of using illegal “distillation” to copy technology, according to the concerns described in the debate.
Will open and closed AI models coexist?
Cox said in a note last week that AI may follow the pattern of the smartphone market rather than an older format war. In his view, Apple’s tightly controlled premium system and Google’s broader Android system show how two models can thrive at the same time.
He contrasted that with the battle between VHS and Betamax video players in the late 1970s and early 1980s. Cox wrote that Betamax was widely seen as the stronger technology, but VHS became dominant after JVC licensed it widely and helped create a larger ecosystem around the format.
That history suggests open AI models do not need to outperform closed systems on every measure, Cox wrote. They can gain ground if they are affordable, widely available and strong enough for many users.
Cox also said open and proprietary AI systems are likely to pressure each other in useful ways. Open models could push closed providers to cut prices and offer more flexibility, while closed models could keep setting higher standards for quality and safety, he said.
What are the risks in the AI model debate?
Critics of open models say broader access can make misuse harder to prevent. They warn that systems without strong safeguards could be used in efforts involving nuclear, biological or cyber weapons.
Closed models carry different risks, according to the debate described by Cox. Safety regulations could reduce competition and benefit companies that already control proprietary systems.
Restricted access can also leave users exposed to government action. The U.S. government effectively blocked Anthropic’s most powerful Claude models for several weeks over security concerns, according to the account of the regulatory risk.
Nvidia CEO Jensen Huang made a similar case Friday, saying the world needs both frontier closed models and frontier open models. Huang also posted a letter signed by Nvidia, Google, Meta, Microsoft and OpenAI.
The letter said open-source software in the 1980s reduced costs and created a shared technical base for American engineers and entrepreneurs. It argued that U.S. leadership in AI should be judged by the strength of a broad open ecosystem, rather than by one leading frontier model.
The companies acknowledged that open models bring security risks. Their letter said the response should be continued access that helps improve defenses, rather than cutting off access.
The letter also argued that open weights strengthen competition across model developers, cloud chips, applications and services. That competition, the companies said, can lower costs and spread the economic benefits of AI more broadly.
This story draws on original reporting from Fortune.