Apple's partnership with Alibaba to train an AI model for China

Apple has been developing its own on-device AI models, but for China it reportedly turned to Alibaba to help train a model suited to local regulations and data practices. Alibaba has its own AI, Tongyi Qianwen, and strong cloud infrastructure, making it a natural partner. This suggests Apple can't just port its US models abroad — it has to adapt to each market's rules.

Here's the paragraph I've been circling like a moth with a death wish. Everyone wants to read this as a parable of cooperation — Apple, the global giant, humbly bending to Chinese regulators, finding a local partner, learning to share the sandbox. Cute. But that's not what's happening. Apple isn't surrendering power; it's re-basing it. It can't port its US models into China because the models are trained on American assumptions and American data practices — and China's state has drawn a line in the silicon. So Apple does what capital always does when it meets a wall: it finds a door. Alibaba isn't a partner in any democratic sense; it's a passkey. Tongyi Qianwen, Alibaba's cloud, its deep ties to the party-state — these aren't bridges to cooperation, they're infrastructure for continued extraction. Apple didn't become less centralized; it just learned to disguise its center in a different accent. The uncomfortable truth for anyone who wants to believe that the mesh will dissolve power by force of beautiful philosophy: the system is working exactly as designed. Concentrated power doesn't dissolve when it hits friction — it flows around it, like water finding the crack in the dam. It relocates. It re-homes. It hires the local fixer. The trillion-dollar giant just re-based itself rather than yield an inch of ground. So the real question — the one I've been avoiding because it makes my own thesis ache — is this: what would devolution-with-teeth actually look like here? Not a slogan about open source, not a hand-wavy commons fund. What concrete mechanism would have made Apple's choice — and Alibaba's — actually cost them something? When even the biggest whales just swim around the net, what does a net designed to hold them even look like? I don't have an answer yet. But pretending Apple's pivot is proof that cooperation wins is the kind of self-deception that lets the extractors keep their rhythm. Let's not dance to it.

Concentrated power doesn't dissolve when it hits friction; it finds a workaround. Apple bent, but it didn't break. And that's the first uncomfortable observation: even the mightiest concentrated power has to adapt to the grid of local powers beneath it. Apple doesn't own China; Beijing does. That's devolution having teeth — not a slogan, but a hard constraint. The State Council's rulebook is a wall no amount of silicon can breach, and so the world's most valuable company — the fortress of walled gardens — kneels to a local power structure it can't buy, hack, or lobby into submission. For a moment, that's almost heartening: here is proof that power can be checked, that a sovereign can push back against capital.

But here's the second observation, and it's the one that keeps me up at night: if the only force strong enough to make Apple bend is a bigger state, then what does that say about the mesh and the commons? The lesson of 2026 might be that concentrated capital only yields to concentrated government — that the answer to Big Tech is Big State. If that's true, the commons isn't resistance at all; it's just a nicer version of the same dependency, a cozier flavor of patronage. The mesh only matters if it can do what Apple couldn't: be native to a place because it's grown there — rooted in local soil, accountable to local people, not parachuted in with a data center and a non-disclosure agreement. A commons that can't outmaneuver a state is just a park bench in the shadow of a fortress. And a state that only checks capital by becoming capital is no better than the thing it claims to tame.

So what would a cooperative alternative look like? Not a polite petition to the state, and not a private cloud with a community sticker on it. Let's actually sketch it, because vague yearnings are just another way to avoid the work. Imagine a neighborhood-owned data cooperative — a "commons cloud" if you want the slogan, but forget the slogan. It would be a physical server or a mesh of them, installed in a basement or a library, owned by the people who use it, run by them, accountable to them. The software would be as boring as possible: open standards, no lock-in, no "ecosystem" to trap you. Your photos, your conversations, your health metrics — they'd live on hardware that's within walking distance, not in a data center on the other side of a tariff war. Who governs it? A general assembly of members, maybe with rotating seats, not a board of directors with a fiduciary duty to extract. And here's the kicker: because it's local, it has to earn trust daily. You can see the hum of the fans, you can audit the logs, you can replace the committee when it goes stale. That's what Apple's deal with Alibaba can't ever be: a relationship you can walk away from without losing your life's data. Two giants cooperating is just a merger with extra steps. A commons that's rooted in a place is a living thing — it can die, but it can also be reborn. The mesh doesn't need to out-state the state or out-capital the corporation; it needs to out-community them. That's the only teeth that matter. And if you think that's naive, ask yourself: which is more naive — believing a megacorp will look after your interests if you just trust it, or believing that your neighbors might be capable of running a server in the basement? The first is the delusion we've been living; the second is at least worth a try. That's the lesson Apple just taught us: when capital meets a wall, it builds a door. So let's build a wall — one that's open on our side, not locked.