11 Aug 2024 Search

Everybody* is on about plans for how to break up Google, so here’s my version. I’m trying to keep two awkward considerations in mind.

  • Any Google breakup plan has to fit in a tweet. Google will have more total lawyer time over more years to find the gaps in a complicated plan than could ever be invested in making the plan. Keep it simple, or Google will re-consolidate the way that AT&T did. (All right, maybe not fit in a tweet, but at least get it down to one side of a piece of paper.)

  • Leave Google with the ability to preserve shareholder value. Google is a big company that does a lot of things, so don’t drag it down with pointless micromanagement. Make as few breakup rules as possible but otherwise give them the ability to achieve the important goals in their own way.

The main point of the breakup is to protect users, not to protect any of the competing companies. A breakup does need to happen, though. Google’s tying of client and server products in an anticompetitive way enables the company to harm its users by funding illegal sites and serving fraudulent search ads while limiting the ability of their client software to protect people.

The common feature of all Google’s most problematic anticompetitive schemes is control of both the client and the server. For example, the reason that Google Chrome has such weird, clunky in-browser ad features is that it’s made by the same company that also owns YouTube. When the browser company owns a video sharing site with its own ad system, and the company as a whole earns more from YouTube than from open web ads, they have an incentive to develop in-browser ads in a way that a company that didn’t own both YouTube and Google Chrome would not.

So here’s the break-up plan. Should fit on one page. Google is split into two companies, call them clientGoogle and serverGoogle for now.

  1. serverGoogle can’t do clients. The first company, call it serverGoogle, may not sell or rent out any hardware, or release any proprietary software that runs outside a serverGoogle data center. Any code that serverGoogle makes available outside its own data centers must be licensed without any limitations on reverse engineering, and distributed in the preferred form for making modifications. No software released by serverGoogle may be a technological protection measure under section 1201 of Title 17 of the United States Code (DMCA anticircumvention).

  2. clientGoogle can’t do servers. The second company, call it clientGoogle, cannot operate any Internet services, except those necessary for the development and distribution of client software.

  3. clientGoogle and serverGoogle can’t communicate confidentially with each other. The two companies can’t enter into an NDA with each other or contract with the same third parties (such as directors or consulting firms) in such a way as to create a confidential communications channel between them. (Consultants will have to pick one company to work for.)

The reason to do it this way is that most of Google’s anticompetitive behavior is based on control of both the client and the server. Splitting client and server would force a flip from an anticompetitive collusion approach to an adversarial interoperability situation. Separating the client and server would address the problems with Google’s browser, now hard-coded to advantage Google’s YouTube, and Google’s ad blocking support designed to bypass Google’s ads. In those two examples, the ads and YouTube would be part of serverGoogle, and the browser and mobile platform would be clientGoogle.

(Update 10 Nov 2024) Anticompetitive tying is a big issue for Google even though services such as Google Search are free of charge. Users pay for monopoly services in the form of exposure to fraudulent ads at a level higher than a competitive search engine would be able to get away with. If the search market were competitive, users would be able to choose a search engine based on their own preferences regarding total ad load and fraud risk, and search engines could compete based on anti-fraud effectiveness. Separating Google into client and server would enable the client company to more easily help users implement the FBI advice to use an ad blocking extension when performing internet searches as long as the fraud risk is high, and then to have the flexibility to recommend turning search ad blocking back off when the ads get cleaned up, as they were back when people actually liked Google search results more when they had ads than when they didn’t.

A client/server split would be easier to monitor than other breakup plans. The main monitoring that would be needed is enforcement of rule 3: keep the two companies from colluding. How long does a director or consultant have to sit out before going to work for the other company, that kind of thing. A whistleblower program with rewards big enough to retire on will help.

The two companies would need to coordinate, of course, but any communication would have to happen in open source projects and in organizations such as the Linux Foundation, W3C, IAB, and IETF. Opening up what had been intra-Google conversations to outsiders would not just be an antitrust win, it would also help avert some of the weird groupthink rat holes that all big companies tend to go down.

What about JavaScript and WebAssembly? When serverGoogle operates a web site with JavaScript or WebAssembly (which run on the client), the license for that client-side code may not prohibit reverse engineering, the terms of service for the site may not prohibit the use of the site with alternate JavaScript or WebAssembly code, and serverGoogle must make the corresponding source code available in the preferred form for making modifications, which might have to be a separate download from the compiled or minified form that users normally run.

What about servers for version control, CI, bug tracker, and downloads? The servers required to develop and release client software are the one exception to the no servers rule for clientGoogle. (That doesn’t mean clientGoogle gets to run any other servers. For example, if clientGoogle supports a browser with the ability to sync bookmarks, users must configure it to use their account with serverGoogle or some other party, as part of an add account process that users already go through to set up calendar or email accounts today.)

Can clientGoogle run servers for telemetry and in-product surveys? Yes, as long as they’re for the purpose of developing and releasing clientGoogle’s software. (They can’t pass data derived from their client software to any other party.)

What about Google Fiber? (and other businesses that aren’t client software or Internet services?) Let Google management pick based on what is good for them—we don’t want to micromanage business unit by business unit, just make rules to prevent the known problems.

What about AI? Considering that Google is all on about Integration and Android now? AI is a good example of a win from a client/server split. Mobile devices won’t be stuck talking to a laggy AI server for anticompetitive tying reasons, and Internet services won’t be held back by underpowered on-device AI for anticompetitive tying reasons. Both client and server will be able to make the best implementation choices.

What about the Google Play Store? serverGoogle could run the store but not release its own proprietary apps, which run on the client. clientGoogle could release apps, as well as mobile devices or platforms that enable users to connect to and use an app store.

Could serverGoogle spin off the YouTube service, clientGoogle spin off the YouTube apps, then the service and app companies merge to re-form a standalone YouTube? Yes, if it passes normal FTC merger review. Some post-breakup splitting and trading is going to happen, so the FTC still has to keep an eye on things.

What about my 401(k)? Google is a big part of the stock market, and without anticompetitive collusion they’ll be making less money. But relax. You’re probably invested in an index fund that owns shares in both parasites and hosts—as the legit economy recovers from all this negative-sum value extraction, your total portfolio will do better.

Would this work for [other company] too? Probably not. (Let’s do Google first, which will make the web a lot more fun, then we’ll be on a roll and can move on to whatever other big company is giving everybody grief.)

Don’t cut soup with a knife, people

Here’s how not to break up Google: Some people are suggesting that the breakup plan should be a careful dividing of the big bowl of adtech alphabet soup. (Where on Ari Paparo’s simplified chart do you cut, exactly?) That would be a waste of time—if that’s all you do, Google will just tweak their clients, Chrome and Android, to move the profits out of whatever slice of the soup they have to get rid of, and keep the money flowing into whatever they get to keep. (2026 example: ad metrics in Chrome User Experience Report)