If the long-awaited WWDC conference two weeks ago made one thing clear to us, it’s this: Hardware manufacturers such as Apple will be the gatekeepers for language models, while language models such as ChatGPT, Claude or Gemini will become interchangeable commodities.
OpenAI CEO Sam Altman was still optimistic when Apple announced the partnership with the young start-up from San Francisco, but the news that Apple was also in negotiations with Meta and Anthropic to integrate their language models was immediately dampened for Altman. Apple Intelligence is therefore not the same as OpenAI for Apple.
The reason is logical: even if ChatGPT is the most powerful language model today, the others follow in its footsteps. For end users, the differences will be barely perceptible and will be shaped more by personal preferences and habits. Just as some people today use Google, Bing or DuckDuckGo, or some choose Google Office and others Microsoft Office.
Much more important will be which of these language models the hardware providers specify as the standard and which will have access to Apple device data and apps. In the case of search engines, we have now learned that Google pays Apple twenty billion dollars a year to be the default search engine on Apple devices.
The gatekeeper function will hand Apple even more market power, and they won’t even have to develop their own language models. Not only do they take a lot of effort to create and operate, but their hallucinations can also quickly threaten a company’s reputation, as Google and OpenAI have had to learn.
Apple, on the other hand, can position itself as a neutral arbiter, giving all language model providers the opportunity to operate on their devices under certain conditions. If violations are detected, i.e. if a language model contravenes its own rules, the language model can be quickly excluded. This is similar to what we are already experiencing with mobile apps in the app store.
Of course, it makes sense for Apple to also develop its own voice model, which is designed for special tasks and will have access to all data on Apple devices. This will run more in the background and be seamlessly integrated and could serve as a kind of “intermediate model” to other models and apps.
Other hardware providers such as Tesla with its cars or the upcoming Tesla Bot Optimus, or a startup like Figure, will also find themselves in a similar position. This is because they determine which AI software will be used in their robots. And their robots will become a new category of products for end consumers.
Some software companies will also take on a similar role to the hardware providers. Namely those like SAP, who have installed widespread enterprise software at their customers. SAP’s AI strategy provides for neutrality in the language models that companies can use, as the German software giant has already done with databases and operating systems. Even if a neutral position is taken, the company’s own software is often created and optimized first and foremost on a few databases and operating systems, with the others following later.
The market power will remain with the hardware providers, the providers of voice models will take on the role of commodity providers, in which they will be constantly challenged to differentiate themselves from the others.