Artificial Intelligence – Platform wars

The fight for the next platform begins.

  • Meta’s decision to release commercial versions of its generative AI models as well as the supposed “leak” of its models have nothing to do with openness being “the best antidote to the fears surrounding AI” as stated by Meta and everything to do with becoming the development platform for the next generation of computing.
  • As Apple, Google, Tencent and Alibaba have clearly demonstrated, controlling the digital ecosystem is a source of great wealth and this is why the AI players are all jockeying for position as this segment emerges.
  • RFM Research (see here and here) has identified three potential control points in the generative AI value chain which would allow those that control them to generate supernormal margins.
    • First, silicon platform, which is currently dominated by Nvidia which has somewhere around 85% market share despite the best efforts of its competitors.
    • Developers are currently specifying Nvidia silicon because they want access to the CUDA development platform which remains significantly ahead of its peers.
    • Part of this is due to the fact that Nvidia has been developing tools for AI far longer than anyone else, but also because it is also very good at what it does.
    • RFM does not see much threat to this control point unless the point of control moves elsewhere in the value chain which is a possibility.
    • Second, foundation model, which is the first step in the two-step process of creating a generative AI service like ChatGPT or Bard.
    • Foundation models are time-consuming, difficult and expensive to build and train meaning that anything that is based on them is fairly tightly bound to that foundation model.
    • If one wishes to switch one’s service from one foundation model to another, the chances are that the service will need to be retrained from scratch on the new foundation model.
    • This is where Meta has been particularly clever because by releasing its model and “leaking” the weights, it has provided the open-source community with everything it needs to begin creating all sorts of services.
    • All open-source models that are widely available are based on Meta’s LlaMa foundation models meaning that it is now the go-to development platform for open source.
    • Switching out will be difficult which is why OpenAI is considering fully releasing GPT-3 to the open-source community before Meta consumes the whole community.
    • Third, chatbots where ChatGPT is the standout leader with over 1bn DaU (daily active users).
    • As a result of its popularity, other service providers like Slack, Expedia and Shopify and at least 140 others have created plugins for ChatGPT which allow it to be used within their services.
    • This means that chatbots and ChatGPT in particular may become development platforms in their own right.
    • This could weaken Nvidia’s grip on AI training and allow some of its competitors to gain share but so far there has been no sign of this.
  • The AI ecosystem is emerging extremely quickly, and this is why there is a mad rush to bring models and services to market as the AI landscape remains fairly open for newcomers to arrive and stake a claim.
  • This will not last very long and when the flow of capital begins to slow to this sector, the shakeout will begin and only the strongest and the best offerings will survive.
  • This is why everyone is trying to add as many users as they can at the moment because this will be a key success factor in determining the winners and losers as it was when the mobile ecosystem emerged 10 years ago.
  • OpenAI is currently in the lead with its runaway hit ChatGPT but there are plenty of viable competitors waiting in the wings.
  • I suspect that there is also going to be a lot of competition meaning that the $20 per month fee to use the most advanced systems will not last very long.
  • This is how the correction begins as many players will have used $20 per month to underpin their forecasts and valuation when raising money when the real figure is probably more like $20 per year.
  • This means they will miss their forecasts and come back to their investors cap in hand far sooner than expected which will lead to a correction in valuations.
  • Price cuts are the first thing I am looking for as a sign that the valuation bubble is getting close to bursting but for now all seems well, but I would not want to be anywhere near this sector when the cracks start appearing.

RICHARD WINDSOR

Richard is founder, owner of research company, Radio Free Mobile. He has 16 years of experience working in sell side equity research. During his 11 year tenure at Nomura Securities, he focused on the equity coverage of the Global Technology sector.