Google – Silver Lining

Nothing much is going to change.

• Google has escaped an immediate order to be broken up largely because Search is seen as being under threat from generative AI, meaning that Google now has breathing space to appeal the entire verdict in a case that is likely to drag on for years while Google’s business carries on as normal.  
• Judge Mehta decided not to grant some of the more extreme remedies put forward by the Department of Justice and instead opted to order Google to share data with rivals and prevent it from making exclusive deals that could prevent rivals from competing.
• This order comes as a big relief for Google, where it was feared that the company would be broken up or that its search deal with Apple would be cancelled, which is why the shares were up 6.7% in after-hours trading.
• Furthermore, there is a long period before which Google has to comply with this order, giving it time to launch an appeal against the verdict itself, where the case could easily end up with the Supreme Court.
• This process takes years, and even if Google ends up losing, the order has enough wiggle room in it that Google could still comply with the letter of the law but maintain the market position that it currently enjoys.
• A large part of the remedy was decided in the context of the current situation in generative AI, where Google’s Search business is seen as being under great threat from a new fleet of competitors such as OpenAI, Anthropic, Meta and so on, who have all launched AI chatbots capable of searching the Internet.
• The reality is that while generative AI has been around for over 3 years, it has had no demonstrable effect on Google’s search business, and Google now has a strong enough position in generative AI itself to compete effectively against the new entrants.
• Google also has a big advantage in that when it comes to the cost of computation as traditional search is far, far cheaper to execute than an LLM-powered search, which is why Google’s new “AI Mode” is mostly powered by the search algorithm and not its LLM.
• This, combined with the fact that the vast majority of Google searches remain very simple, means that they are not best suited to be executed by LLMs will keep Google from losing too much market share.
• The net result is that Google has years before it has to comply with any of this, and that this is just the end of the first chapter of what is going to be a long saga.
• Hence, business will continue as normal for the foreseeable future, meaning that Google will be able to leverage its position in Search, Chrome and Android to entice users to use Gemini rather than ChatGPT, Claude, Meta AI and so on.
• In this regard, Gemini is already pretty competitive with 400m MaUs, but is still very much in 2^nd place compared to OpenAI, where RFM Research estimates it has 1.2bn MaUs.
• However, I suspect that these 1.2bn MaUs are costing OpenAI a fortune as they consume substantial compute cost and the vast majority make no contribution to OpenAI’s revenues through subscription or advertising.
• Hence, the bigger ChatGPT becomes, the more resources it will consume.
• This is further exacerbated by the fact that GPT-5 is thought to be larger than 3tn parameters, which will mean the cost of compute will have taken yet another leg up, thanks to OpenAI deciding that it needed to make GPT-5 available to everyone for free.
• The net result is that the threat to Google looks substantial on first glance, but there are clearly cost and scale issues which give Google an advantage when it comes to keeping the upstarts at bay.
• Now, it has benefitted from this perceived threat by receiving lighter remedies from Judge Mehta, which improves its overall outlook for the foreseeable future.
• With this threat now relegated to the background for some time, risks from investing in Google are reduced, which further enhances its attractiveness as one of the best places to take a position related to the AI boom.
• Nvidia is still the best place to have a direct position, but Google trades close to half of Nvidia’s multiple, creating a strong value argument.
• However, the adjacencies of Nuclear Power and inference at the edge are yet another order of magnitude cheaper, and it is here where I still have exposure.