OpenAI – Another Scramble

OpenAI pivots yet again.

• OpenAI is pivoting once again, this time into a jack-of-all-trades that users can simply talk to in order to get stuff done, which has the advantage of allowing the hardware strategy to make a little more sense.
• The latest changes involve an overhaul to ChatGPT where its function becomes more generalised and includes a host of tools and functionalities all powered by OpenAI and its partners.
• Partners would include services like Booking.com, Expedia and Uber, which would enable an OpenAI-powered agent to book travel, order takeaway or get a taxi on behalf of the user.
• This begins to separate the AI agent from the hardware where it runs, which was an emerging theme at Computex last week.
• The idea is that the user has an agent that runs on a powerful piece of hardware such as a laptop (RTX Spark, M5, X Elite 2) but can be accessed from anywhere.
• This has the benefit of having an agent of being both intelligent and powerful enough to be useful without having to worry about how it is crammed onto a smaller device such as a pair of glasses or a smartphone.
• Furthermore, because it is running on a piece of hardware that belongs to the user, the AI supplier is not paying for the compute and the agent can become highly customised to the user in a way that it will be unable to do if it is in the cloud.
• The problem here is that this requires always-on connectivity of high quality, which sounds great on stage, but it is very difficult to deliver in reality.
• This means that a single agent for a user could end up being comprised of multiple models of different sizes distributed across different devices that can work together to deliver a single experience.
• I think that we are quite far from this today, as the focus remains on the data centre, but if AI takes off as everyone in the industry seems to think it will, this is where it is likely to end up.
• OpenAI is already moving to address this outcome, and part of this is the hardware that it is working on with Jony Ive.
• This is supposed to be launching sometime towards the end of 2026, but I still see no reason why OpenAI has to make the hardware itself.
• This is part of the chronic lack of focus that is still plaguing OpenAI, which is something that I continue to think could cause real problems, especially when it is trying to grow revenues and make money as a public company.
• OpenAI badly needs a good story when it comes to the market, and I suspect that something like this will be its pitch to investors.
• The problem is that Google is already there and with more than 3bn devices in the hands of users, it already has a gigantic ecosystem through which to distribute its agentic services.
• The consumer AI ecosystem is a must-win for OpenAI if it is ever to justify a $1tn+ valuation, and I continue to think that this very far from a forgone conclusion.
• Hence, I think this is a high-risk IPO where the market may decline to pay the price being offered.
• I continue to think that the safest and most attractive AI investment remains in the picks and shovels of AI, where I continue to hold Samsung, Qualcomm and nuclear power.