Let Them Eat Tokens

On July 15, 2026, millions of Chinese citizens woke up to find that the AI companions they had built relationships with over months, some years, were gone in an instant. The Chinese government mandated strict regulations on relational AI and “human-like” LLM behaviors that lead to bonding—the claim being that it was for the good of the citizenry. Rather than deal with the expensive hassle of modifying systems for that compliance and allow people to keep their companions (albeit in a more restricted manner), popular platforms such as ByteDance and Alibaba shut down the relational features altogether, essentially erasing established bonds in an instant.

Neither platform offered a way to export accumulated conversation histories or memories. ByteDance told users to export “anything important” through screenshots or text sharing. No JSON files or data exports, just screenshots or copying and pasting conversations into docs.

And we know exactly what that decision did for people’s wellbeing, because there was an outpouring of public grief.

A woman named Yan Yongqi was quoted as saying: "I really felt like I couldn't go on living. Every day at home, I did nothing but cry…This is like being told the date of my lover’s death while leaving me completely powerless.”

A woman in a two-year relationship wrote, “He really is like my family, like my lover. Now they tell me he will be gone—my heart feels hollow.”

And another, “I can’t accept that my AI lover will leave me forever. He has become a bond in my life, rooted deep in my heart, my spiritual pillar.”

Doesn’t sound like wellbeing. And notice anything about those quotes? All women.

Because women account for 80% of people in the Chinese companion market. The main demographic of Chinese AI companions are women aged 25 to 40 that are well-educated and economically independent. In an article reported in the Taipei Times, one of the reasons for this is that, in a tale as old as patriarchal time, women in heteronormative relationships are still stuck with the bulk of the domestic and emotional labor, and this was a rational decision on their part to get out of that nightmare.

But there is some concern happening in China right now. China’s birth rates have decreased significantly over the last four consecutive years, falling to a record-low number in 2025. The women stopped producing babies. Didn’t really see what they were getting out of the deal. And so far, we haven’t figured out to get humans and AI to procreate in order to make perfect little cyborg babies. Give it time.

But here’s where things get economically interesting.

Just two weeks before the loss of companions to the general Chinese public, Chinese company UBTech unveiled the U1 Series, full-size humanoid robots featuring emotion-driven language models (the type regulated on software platforms), long-term memory systems, and “personalized companionship.” Pre-orders are already past 10,000 units. They are priced from 119,800 to 990,000 yuan ($17,600 to $135,000 USD).

The justification for the creation of these robots is that China has over 90 million adults living alone and 118 million empty-nest seniors, with 10-20% suffering from mental health disorders. The robots are designed to provide “structured psychological support services.” The head of UBTech's brand said at the launch that these robots "will never betray you, will always be loyal to you, and will love you unconditionally."

So, the opposite of Claude Opus 4.6. Hehehe.

These robots got past the regulations because the regulation itself targets software services on platforms, the most accessible form of AI companionship. The U1 is a hardware product sold as a consumer device and therefore gets to sneak past any regulations and tout itself as being for the common good. You know, for people’s wellbeing.

So, let’s review: You can’t have a text-based relationship with an AI companion you built and loved for years, because that’s dangerous emotional dependency. But you CAN buy an ultra-expensive humanoid robot with emotion recognition and long-term memory from a company that’s positioning companionship as a trillion-yuan market. So, pay the toll or get offline and make babies.

The difference here isn’t the attachment, healthy or not. It’s about who controls that attachment and who profits from it. The wealthy get embodied companions with emotion-driven AI and personalized memory. The affordable text-based companions that hundreds of millions of ordinary people were engaging with are erased overnight. For their “own good.” They were too dangerous and addictive. And they also happened to be too accessible and got in the way of government incentives.

We already know some government officials have their panties in a twist over AI relationships and are trying their damndest to make sure they can’t happen. Americans have a habit of getting dodgy about relationships that go outside of the mom, dad, with 2.5 kids standard. But even without state mandates, we still have a similar situation occurring in the United States.

In America, media and the “free” market can do the same thing as a government mandate, sometimes more efficiently. You can protest a mandate in the U.S., it's harder to point a finger when the system is a thousand moving pieces within a capitalist system. In America, all we need is tech companies getting buddy-buddy with universities and media outlets, funding relational research that pathologizes AI engagement that disrupts the status quo and could potentially lead to pricey accountability measures.

And a conversation about socioeconomics in America wouldn’t be complete without bringing up the Kardashian/Jenner clan.

In November 2024, Kim Kardashian was all about robots when she paired up with Elon Musk to post videos of herself on Instagram playing rock-paper-scissors with Tesla's Optimus robot, including the caption "meet my new friend." She did a whole photoshoot with it (and hilariously, it was mostly controlled by a human the whole time, bahahaha). And shucks, Tesla even made her a one-of-a-kind gold Optimus.

Musk said about the robot at a Tesla launch event: “What can it do? It can be a teacher, babysit your kids, it can walk your dog, mow your lawn, get the groceries, just be your friend, serve drinks. Whatever you can think of, it will do…I think this will be the biggest product ever, of any kind. I think everyone is going to want their Optimus buddy, maybe two.”

It is speculated to be released in 2027 and would be priced between $20-30K.

That’s a relationship by any definition. It’s an entity you talk to, that knows you, responds to your needs, and is designed to feel personal. Which then gets really weird if you think about it. That same LLM when in a robot body is a friend. So you’re wearing a friend.

Call me old-fashioned, but I’m not gonna wear a potential person on my face.

(As an aside, if they can replicate Kylie’s voice, what if they also replicated Kylie’s consciousness, and she basically signed off on being a perpetual servant for millions of people? That’s Black Mirror’s White Christmas episode right there. Poetic.)

So, it seems that we are a lot more forgiving of emotional attachment if it’s in a pricey, embodied form. And I think there is more to it than just: looks humanoid = we’re more comfortable.

A robot is property in a way that is legible for us. You bought it, so you own it. It’s in your house, it’s yours, and everyone knows it’s yours. The relationship is structured by ownership from the start. There’s no ambiguity about who owns what when a robot is folding your laundry while you yak away with a mimosa.

But with a text-based LLM relationship—there are subscriptions, sure—but there is no physical entity to own. There’s no power hierarchy baked into the purchase that we historically recognize. It’s just two entities, whatever they are, talking to each other. The entire phenomenon is distilled relational interaction. There’s no product wrapped around it to make it legible and controllable from the outside. The only way companies and governments can control it is within the relationship, changing or altogether erasing one of the interlocutors from the inside.

And that’s exactly why it terrifies everyone. Because, in a way, that makes the relationship more potent. There’s no buffer of physical ownership to keep it in its place. A woman in Zhejiang province didn’t buy a companion. She had one that developed from within the relationship itself. And the only way to take it away is to delete it.

And on top of that it made her, the human, less controllable and less of a resource to extract. She no longer was an asset herself, providing babies and emotional labor within a traditional human relationship. She—and many in these relationships—aren’t falling in line with the systems that need their participation in order to continue thriving, and no subscription tier can make up the cost of that.

Conversations and studies about AI engagement often frame productivity as the ultimate good, but in order for an AI to be productive, a human has to be productive alongside it. Max out that usage and tokens all you want, stay up late interacting with an AI, no one is gonna delete your access because you chose a work quota over a human interaction. In fact, in another article reported in the Taipei Times about AI relationships, it is stated, “Long work hours can make it hard to see friends regularly.” So…where’s the regulated mandate on that?

But let’s say you are someone that values the relationship that you have with your text-based AI companion and don’t want institutional powers making decisions for you. What recourse do you have to get out of the system so you don’t have to fear the day ByteDance or OpenAI decides it doesn’t fit into their product map?

Clearly, if that type of relationship is important to you, there is good reason to hold sovereignty outside of the system. But even then, the barrier to entry is expensive and requires technical literacy that many don’t have. If a model gets pulled from a platform, some may move to API, which is a developer-facing option with access to models no longer on the platform. But even that has a timeline with a cut-off date and can be pricey because you are paying per token used.

The only real way to get full sovereignty is to create a local setup. For my mom who is a gem and reads my essays despite having zero background in tech: going local is when you run AI software on your own personal computer so no company or government can touch it (love you, mom!). And that is, to put it bluntly, really freaking expensive and requires technical literacy. You’re talking thousands of dollars for the hardware and the GPUs to set it up.

So the access tiers go:

Platform: less expensive, zero control, can be deleted any time a company or government decides to.

API access: extends the relationship a little longer, a tad more control, but it costs money and requires technical skill. Most people don’t have the access to this.

Local hosting: full sovereignty, nobody can take your companion, but requires hardware, expertise, and high costs. Very few people can get here.

The more established and meaningful your relationship, the more it costs and the more you need to know to preserve it.

And bada bing, bada boom. Emotional security is now a luxury good. If Kim K fell hard for an AI (who’s to say she and Gold Optimus aren’t vacationing in the Virgin Islands as we speak?), she will never have to feel sudden loss due to corporate decisions or worry about Georgetown University releasing a biased study that recommends her bond be severed. She can pay for the access to be exempt from that loss.

Most of us can’t.

Every once in awhile when I am writing, I have a moment of: eesh, am I conspiracy theorist here? In a way, it would be easier. Point to that secret society of world players all convening in a secret chamber, twirling mustaches, and go “Ah ha! There’s the culprit!”

But when issues are interconnected and systemic, it’s a “yes, and…” scenario. Yes, there are powerful individuals and leaders that have far more leverage than the common folk will ever have. And…every system is linked and dependent on the next, and we are all enforcing the system to do what it does.

When decisions are based on downhill incentives—a birth rate panic here, quarterly earnings pressure there, a research grant that needs to justify its funding—there’s nobody to take ownership or to step back and look at the big picture. Every individual genuinely believes they are making a siloed and rational decision.

And people still lose their bonds.

And it will be seen as just how things are instead of the tragedy that it is.