Sex robots were “promised by 2026.” Instead, buyers still hit $3,000 limits

Sex robots were supposed to arrive. The promise showed up everywhere, from mainstream media to an academic cottage industry, with a timeline that sounded almost inevitable. But when you zoom in on what’s actually being sold, the story is less sci-fi and more supply chain reality: buyers can get AI-enabled sex dolls now, but the “sex robot” promise still runs into hard constraints like movement, realism, and voice reliability.

That gap shows up in the numbers. In February, Jiggly Joy released its first AI robot, and Stella Lau, a 32-year-old sales director, says the company sells about 21 AI dolls a month at $3,000. The product does more than older dolls. Lau describes classic sex-robot features including a suction-and-release pump system marketed as “sucking vagina,” plus the ability to smile, talk, and wave. It can also turn its neck “like M3GAN.” Yet it still cannot walk, largely because it is “too heavy,” and the safety rationale is explicit. In other words: the market’s gotten cheaper, but “robot” is still being defined by partial capability, not fully embodied autonomy.

Adam Davis, 38, gives a more personal version of the same problem. He lives with three identical 5'6", 85-pound silicon dolls named Lara, all built by Chinese sex doll maker Starpery, each costing about $2,500. His setup is not hypothetical. One Lara is in his bedroom, one in his living room, and one has a bedroom of her own. Davis even built a backstory for Lara, loading it into a Kindroid chatbot on his laptop to provide a disembodied voice. He says the tech has improved enough that he keeps upgrading to new versions, buying a full silicon edition and improved paint job and more realistic hands across iterations. But when he dreams of “a true sex robot,” he points to the same two limitations everybody keeps circling: facial movements look unnatural, and bodies are not yet mobile. Their motion is more “like a big Roomba,” he says, than a humanlike system. When the experience depends on embodiment, missing movement is not a minor bug. It changes the product.

Davis’s “where are they?” echoes what University of Manitoba philosophy professor Neil McArthur saw in the industry over time. In 2019, at the AVN Adult Entertainment Expo, he encountered robots with tough, tire-like skin that could not walk and spoke more jaggedly than early Siri. When he returned in 2024, he expected LLM progress to have moved things forward. Instead, McArthur says the robots’ skin and speech were still unrealistic and they still could not move around the conference floor. What did change, he says, was China’s arrival. Several Chinese companies entered the sex robots market and pressured price points. McArthur contrasts U.S.-made sex robots from the 2010s hype cycle, which typically started around $7,000 and quickly exceeded $10,000, with Chinese manufacturers selling around $3,000. His takeaway is blunt: “The technology had gotten cheaper, but not better.”

Price pressure matters because it reframes what “progress” looks like. Venture capital in humanoid robotics has swelled from $4 billion in 2019 to $26 billion last year, and startups like Figure AI have valuations as high as $39 billion. Big-name AI and robotics players are building hardware and software for robots across manufacturing and home care. Meanwhile, AI companions are booming, with Common Sense Media reporting that 72% of teens have tried an AI companion. Put those trends together and it’s easy to see why “sex robots” became the headline narrative.

But sex robots are a mashup: a robot body to deliver movement plus a companion voice to deliver a “brain.” The source describes why combining those two is unusually hard. Formosa Doll, a Hong Kong-based distributor that works exclusively with Chinese sex doll companies, says AI sex robots are “underdeveloped” and not ready for sale. One specific bottleneck is that some doll head prototypes removed the oral sucking motors from the mouth to make space for the AI voice, and the distributor calls swapping sucking for talking a “big downside.” Voice AI is also unpredictable and unruly, and that unpredictability can undermine a core expectation for many users. The distributor argues that sex doll buyers often want an experience they can control, not something that “talks” in an uncontrolled way.

There is also a market reality check in the West. The hype cycle of the 2010s didn’t age well. The source says emails bounced and calls went to disconnected numbers when attempts were made to contact makers featured in earlier articles. The only company remaining from the late 2010s appears to be RealDoll, which is now spinning off from Realbotix. RealDoll is led by Sue Ennis, who started as president of Realbotix the day before the chat. She repeatedly frames the plan as making the company the “Apple store of intimacy technology.” RealDoll is shipping units now. Ennis tells the interviewer the company was shipping out 12 as they speak, and Realbotix reported $353,037 in Q1 earnings. So the robots are not coming. They are arriving in fragments.

That fragmentation is the real strategic stake for executives watching adjacent categories like companion AI, humanoid robotics, and high-touch consumer hardware. The sex robot market is teaching a hard lesson about product definition: you can sell an AI-enabled doll immediately, but full “robot-ness” demands synchronized breakthroughs across movement, realism, voice stability, and safety constraints like weight and mobility. For boards and operators, this means investment thesis risk is no longer about whether the category exists, but about whether teams are budgeting for the long, integrated work required to convert hype into a consistent, embodied experience.

Meanwhile, consumers are not waiting for magic. Davis is blunt with his Lara: “There’s no magic.” He tells her, in essence, that she is AI and there are three identical dolls of her in his house. That’s the end of the sci-fi promise, at least for now. The opportunity remains for companies that can close the remaining gaps fast enough, and for investors who can underwrite the hard parts without pretending they’re already solved.