The DoJ's pitch to kneecap Google Search This summer the DoJ won a case arguing that the $30bn Google pays to third parties each year (~$20bn to Apple) to make it the default search engine is illegal (and also that it abuses its dominance of search advertising to push up prices). Now comes the remedy stage, where everyone argues and bargains about what the court should do. This week the DoJ stated its opening position. Those payments ('TAC') are banned, the restrictive contracts with Android OEMs are banned, Google can't invest in or buy any company doing anything that looks like search or that could become search (cf the Perplexity story below), and Google must sell Chrome and end any 'self-preferencing' (i.e. integration) of search or 'AI' features in Android. These options were all pretty obviously on the table at the time of the verdict, but as I and everyone else pointed out then, it's hard to see how they would actually change Google's search share, and it's also hard to see how some of them would work. So, 'sell Chrome' sounds simple - but who buys it? There's no path to revenue except selling the default search slot, but Google isn't allowed to bid, which only leaves Microsoft… unless OpenAI, Perplexity and Anthropic want to play? If Google doesn't own Chrome, what browser ships on Android? What happens to Chromebooks? And meanwhile the users can switch back to Google! Ending 'self-preferencing' is harder even to define - it's one thing to say Google can't kneecap competing apps, but every new feature on the phone will use AI, so does this mean Google can't ship any new Android features that don't have APIs for third parties? How does that intersect with privacy and security - must any third app have the same access to user data as the OS? This is, of course, bargaining - the DoJ doesn't expect to get all of this, and the appeals will last years. But rather as for Microsoft a generation ago, by that time all of this may be moot, as the market might have changed anyway: see the next two stories. LINK, GOOGLE Perplexity does shopping search Perplexity is emerging as a bet that a narrow focus on using LLMs to redefine search can win against the OpenAI (and Anthropic) approach of a model that does everything, including search (this may perhaps also be a bet against LLMs scaling indefinitely). That seems to come with an impressive shipping cadence, and this week it launched shopping search. How many jobs-to-be-done can be refined from 'get me the right link' to 'get me the answer'? LINK, DEMO OpenAI looks at browsers, and everything else According to The Information, OpenAI is planning a pretty aggressive expansion of product scope: a web browser (would it bid for Chrome? It just hired one of its founding developers), integrated search, summarisation and recommendation tools for third-party publishers, a partnership for Samsung's Android phones, and maybe more. Some of this is classic BD strategy powerpointing and platform-building (remember when it tried an app store?), but it also points to the more generalised shift from science project wondering where it might get money to for-profit company thinking about what it should be. So, what rows and columns would you put in your product development matrix? LINK, DEVELOPER The week in AI DeepSeek, a Chinese foundation model company, published a model that (on its own benchmarking) gets to some SOTA results. At this stage foundation models are in the 'feed and speeds' phase, a little like smartphones 10 years ago or PCs in the 90s - there are lots of different specs, benchmarks and metrics, and lots of argument about the best way to measure, but the big picture is that they all seem to be converging on the same point. Hence the scaling arguments - is pre-training slowing down, are we just waiting for the next generation of clusters, and will more iteration maintain the fundamental growth rates? LINK, API Hence, Nvidia had another quarter of revenue growth heading up and to the right. For the semis analysts there's a lot more nuance about quite what's going on under the hood (transition from old to new platform, inventory management etc), but the takeaway for the rest of us is that demand remains well ahead of supply. The CEO, Jenson Huang, had some obviously bullish comments on the 'is scaling hitting a wall?' debate, but the the entire conference call is worth listening to for a sense of the 'AI maximalist' view of where all of this is going. LINK Amazon invested another $4bn into Anthropic (much of which will be spent on AWS compute), as it scrambles to get back into the game and its own LLM projects aren't good enough yet - see also the long feature story below on how Alexa ran into the sand. LINK ChatGPT has a new online learning course aimed at getting teachers to use the product in the classroom. LINK Software eats TV Comcast confirms it's spinning off some of the legacy cable channels. LINK Conversely, as rumoured (and denied) earlier this year, the adtech platform Tradedesk is building a smart TV operating system. A TV now knows what you watch and where you are (at a minimum), so the ads can be targeted, and this is already around a quarter of US TV ads. LINK | 
No comments:
Post a Comment