Meta Platforms– Reinvention

Meta claims the frontier (again).

• Meta’s retaking of the high ground demonstrates its return to the front lines but also shows again that there is no moat in models, meaning that the winners and losers will be decided by who owns the touchpoint with the user.
• Meta has launched its flagship versions of Muse, which it has called Muse Wonder and Muse Wonder Full, and unsurprisingly they top all of the benchmarks (cherry-picked?).
• They are both closed models, and Muse Wonder is currently only available via the Meta Model API and Muse Code, while Muse Wonder Full is only available to Muse Code paid subscribers.
• This confirms that the hottest use case for AI at the moment remains writing software, which Meta is addressing both to maximise the monetisation of its compute and compete against Anthropic.
• Muse Wonder will quickly become the backbone of all of Meta’s AI services that are offered to users, where the Muse agent is its most popular product and still tops the Apple and Android download charts.
• This is a very important statistic as it implies that Muse is a bit more than a flash in the pan and is more useful and intuitive than the critics have been forecasting.
• This is a good return to form, where the pressure to keep up caused Mr Zuckerberg to make a mess of Llama, and Meta to fall significantly back in the race to superintelligence.
• One very expensive reorganisation later, and the first real fruits of its efforts are beginning to show, assuming that the benchmarks are an accurate representation of reality.
• By far the most important model here is Muse Wonder because this is the one that is going to power the vast majority of Meta’s AI services that are aimed at its more than 3bn global users.
• This is where Meta (and Google) have an advantage over both Anthropic and OpenAI, as they already have long-term relationships with users, giving them a unique opportunity to entice users to use their AI and their agents over anyone else’s.
• Meta has already made its move in this direction, but when Google is going to launch its Gemini-powered agent remains something of a mystery.
• This is the coming battle for the AI ecosystem, and so far the evolution is proceeding exactly as the smartphone ecosystem did before it.
• I expect that there will be many, many agents all vying to run the user’s digital life, but once the AI ecosystem is mature, only three or 4 will remain.
• Failure to be one of those 3 or 4 will mean no differentiation and an inability to earn more than a commoditised return on any AI service offered.
• With a valuation of $1.2tn, this is a must-win for OpenAI, and it is also a must-win for Apple if it is to preserve the differentiation of its hardware in an environment where users all begin to use agents rather than apps.
• The potential for Muse to become one of the most used agents, combined with the new business of selling compute to enterprises, allows me to value Meta at least $770 per share, meaning that there is still some upside left in the valuation.
• However, if Muse continues to gain in popularity, it is possible that the shares could run past this number as sentiment and momentum take over from fundamentals.
• I have a position in Meta that I am sitting on until at least $770.