The brutal verdict of 2026
According to several converging analyses, the failure rate of AI hardware devices launched in 2024-2025 stands at around 85%. In our article on the great AI failures, we mentioned the Humane AI Pin and the Rabbit R1. But those two cases are only the tip of the iceberg. Friend Pendant, Looki L1, AI Pins from various Chinese manufacturers, Razer Project Motoko, and many others: all have disappeared or are in serious difficulty.
Conversely, only one product has genuinely taken off in this category: the Ray-Ban Meta, which has passed 300,000 pairs sold and continues to grow. The difference between the two camps is not a question of technical quality. It is a simple economic rule that many engineers still refuse to see.
The graveyard of wearable AI
Let us go through the dead in order of how spectacular they were.
Humane AI Pin (April 2024, dead February 2025). $230 million raised from OpenAI, Microsoft, Salesforce and Marc Benioff personally. Sold for $699 plus a $24 monthly subscription. Every Pin sold was remotely deactivated on 28 February 2025 after the assets were resold to HP for $116 million. Fewer than 10,000 units in total.
Rabbit R1 (January 2024). 100,000 units sold on the basis of a CES demo that turned out to be largely faked. The promised "Large Action Model" does not work. The company is struggling to pay its salaries in 2026.
Friend AI Pendant (2024). Launched as an "AI emotional companion" worn around the neck. Several million dollars raised. Business model deemed unviable by specialist media. Virtually gone from public conversation.
Looki L1. An AI pendant shaped like the Nickelodeon splat (yes, really), equipped with a 1080p camera that films your life in order to offer you commentary. A niche of a niche, with limited distribution.
Razer Project Motoko. A gaming headset with two AI cameras meant to comment on your game live. A disastrous demo, released to indifference.
In 2024, the American tech press had ranked the Rabbit R1 and the Humane AI Pin among "the three biggest hardware failures of the year". The New York Times had spoken of an "AI hardware bubble bursting". In 2026, the phrase has been borne out. No standalone product dedicated to AI has found its audience.
Meanwhile, the Ray-Ban Meta are flying off the shelves
The Ray-Ban Meta, launched in 2023 and improved in 2024-2025, have passed 300,000 pairs sold according to the figures reported. They are integrated with Meta AI, take photos, translate live, and answer visual questions. The more advanced Ray-Ban Display model is announced at around $800. And everyone in the industry looks at this product as the benchmark to imitate.
Apple is preparing its own smart glasses for 2027, with production that could begin as early as December 2026. Google is also returning to the game after the historic failure of Google Glass. And several Chinese manufacturers (Oppo Air Glass 3, Xreal Air 2) are trying their own version.
>>Why do the Ray-Bans work where Humane failed? The answer comes down to four words: they look like Ray-Bans.
The rule of invisible accessories
Here is the rule that the designers at Humane and Rabbit ignored: a consumer electronic device only succeeds if it looks like a familiar object. The Ray-Ban Meta are first and foremost Wayfarer glasses, made by Luxottica (the Italian giant that also owns Oakley and Persol). The cameras are hidden in the hinges. The microphones are invisible. The only visible indication is a small LED, added under pressure from European regulators.
The user is not wearing an "AI device". They are wearing glasses. It is this continuity with an object already socially legitimised that makes the product acceptable. No one looks at them strangely in the street. No one asks them "what's that thing clipped to your t-shirt?" as with the Humane Pin.
Conversely, Humane's AI Pin proposed to replace the smartphone with a new object that existed in no cultural frame of reference. A strange light pinned to a jacket lapel. A laser projector that drew text in your palm. These gestures belonged to no existing social ritual. The product forced the user to constantly explain what they were wearing and why. No consumer technology has ever succeeded with that kind of social friction.
Remember: in 2013, Google Glass was killed by that same social friction. Wearers were nicknamed "glassholes" and banned from certain bars and restaurants. The technology was technically 10 years ahead, but it forced the user to become publicly "the one who films everything". Humane and Rabbit repeated the same mistake in 2024, thinking that AI would solve the social problem. It did not.
The other trap: replacing rather than augmenting
The second problem common to AI hardware failures is their ambition to replace the smartphone, rather than integrate with it. Humane's AI Pin had its own cellular connection, its own subscription, its own projector. It tried to be an alternative phone. But it did almost every task worse than a smartphone, while asking the user to relearn everything.
Apple appears to have learned the lesson. According to Bloomberg, Apple is developing an AI pendant the size of an AirTag, which would clip to a shirt or a necklace. But with one crucial difference: it would connect to the smartphone rather than replace it. It is the same logic as the Ray-Ban Meta. An accessory that extends, not one that replaces.
The Apple AirPods with built-in cameras, also reportedly in development according to leaks, follow the same philosophy. People already wear AirPods everywhere. Adding small cameras to give the AI context is a small conceptual step, not a leap. It is that small step that makes the difference between a product that works and a product that dies.
The special case of productivity devices
All of this analysis concerns consumer AI hardware. There is one category that is doing well but flying under the radar: AI tools for professionals. Otter connected speakers for meeting transcription, Meeting Owl for video conferencing, XReal and Vuzix augmented reality headsets for industrial training. These products are not trying to revolutionise everyday consumer life. They solve a specific problem for a specific job.
The economic logic is exactly the same as for the software startups that outperform: a targeted audience, a measurable productivity gain, a coherent business model. Professional AI hardware works because it fits into existing workflows. Consumer AI hardware fails because it tries to create new ones.
The next test: Apple in 2027
All eyes are now on Apple. The company is preparing three AI hardware products in parallel: smart glasses for 2027, an AirTag-like pendant, and AirPods with cameras. Tim Cook told employees in February 2026 that the company was "extremely excited" about these new product categories.
If Apple succeeds where Humane failed, it will be because the company has applied exactly the rules we have just described: continuity with familiar objects, integration with the smartphone rather than replacement, measurable productivity gain. If Apple fails despite these advantages, it will be a sign that consumer AI hardware is fundamentally a bad idea, whatever the manufacturer. The next 18 months will give us the answer.
In the meantime, if you are tempted to buy an AI gadget worn around your neck or clipped to your jacket, ask yourself a simple question: do you really want to spend all day explaining what you have there? If the answer is no, the product is probably not for you. And it is probably not for anyone.