The Moat You Filled

On 16 July the European Commission ordered Google to give away its advantage. Two binding decisions under the Digital Markets Act require the company to open eleven core Android features to rival AI assistants, ChatGPT and Claude among them, and to share anonymised query, click and ranking data from its search engine with competitors. The data-sharing begins in January 2027, the Android changes fall due across the following year, and non-compliance carries fines of up to ten per cent of worldwide revenue.

Google calls the thing a moat. It is the language of strategy for an asset no rival can cross: twenty years of accumulated searches, the deposit that keeps competitors on the far bank. The Commission looks at the same object and sees infrastructure, a resource so central that keeping it closed makes competition impossible by construction. The word moat gives the first reading away. You dug the channel; you did not make the water. Every click, every query, every mid-search change of mind is someone else’s act, and the corpus is the sediment of that behaviour at planetary scale. Google is its custodian, not its author. A fence is not a field.

On this the European administration reads the situation well, and it is worth saying so plainly. It has grasped that the value lives in the aggregate, and that an aggregate assembled by everyone and held by one party is an enclosure wearing the costume of an invention. It is the same manoeuvre I called the enclosures when writing about the public dividend: a common good hedged into private property while no one was watching. Here, for once, the hedge is cut back by decree.

What Brussels does not see is harder to celebrate. Start with the uncomfortable part: to share search data is to hand the public’s behavioural record to more parties at once. Opening the moat does not drain the surveillance; it multiplies it. Then the timing. The decision assumes power still resides in the data, while the value is migrating elsewhere: last week an open Chinese model, Kimi K3, pushed the semiconductor index into bear-market territory and made investors ask what $725 billion of capex actually buys, because the software layer is commoditising faster than anyone budgeted for. Brussels is prising open yesterday’s chokepoint while tomorrow’s slides toward the model and the compute. And the most political point of all: handing the corpus to OpenAI and Anthropic, firms already wired into the state’s own machinery, redistributes the enclosure among a few giants rather than returning it to the citizens who filled it. The common field passes from one landlord to a small consortium of them. It is the outcome I traced when a lab bought the commons for a tenth of a per cent of its valuation: the commons changes owner and stays closed.

Is Google right to defend itself? Grant the defence its strongest point: the ranking is real engineering, genuine investment, no gift fallen from the sky, and whoever builds a system has some claim to protect it. The raw material, though, was gathered from everyone, most of it through default settings, inside a lock-in that made the alternative theoretical. You cannot hedge a field taken from the common and then present the hedge as your own creation. The party that pooled other people’s water cries expropriation in the thinnest of voices. The defence of the moat is the defence of the enclosure, and the more dignified posture, given how the water got there, would be silence.

For anyone who works with these systems the point lands close. The data you generate every day, your searches, your corrections, your second thoughts, already sits in a deposit that someone else administers. The contest the DMA has opened does not decide whether that deposit exists; it decides who holds the keys, and under what mandate. Knowing the deposit is common will not make it yours. It is the only premise from which any accountability can begin.

One last thing, and I smile as I write it. One of my very rare and brave readers wrote me that certain sentences of mine are hard for a native English speaker to read without flinching in the chair. They’re fake, he thought. But what does it mean to be fake, or wrong? What stands written is correct, and aiglish (the English of the machine...) IS a living language like any other: no one owns its correctness, exactly as no one owns the moat of clicks. A living tongue belongs to whoever uses it, and it shifts under their feet. Still, from today I will ask my AI for a better impression of the Bard. Even the fences of style are worth cutting back now and then.

AI is a machine, but is the language it uses really artificial? Good question. I'll think it through this afternoon, chasing Ugo, natürlich...