Why the Humane Ai Pin Failed: $230 Million and 10,000 Units

The Humane Ai Pin failed because it could not beat the smartphone at any job its buyers did every day. It was slow, often wrong, ran hot and had poor battery life, and it cost $699 plus $24 a month for answers that phones were adding for free. Humane raised about $230 million (reported), shipped about 10,000 pins against a goal of 100,000, and sold its software and staff to HP for $116 million in February 2025, ten days before the pins were switched off.

The low point came fast. Humane unveiled the pin on 9 November 2023 to wide attention; Time had already listed it among the best inventions of 2023 before anyone had used it. The first reviews arrived on 11 April 2024 and were among the harshest ever given to a funded consumer device. Marques Brownlee's review video three days later was titled "The Worst Product I've Ever Reviewed... For Now". By May, Bloomberg reported that Humane was seeking a buyer at about $750 million to $1 billion (reported).

What was the Humane Ai Pin?

The Ai Pin was a small square wearable that clipped to clothing, with a camera, microphones, a speaker, a touchpad and a "Laser Ink" projector that displayed text on the user's palm. It made calls, sent messages, answered questions, took photos, translated speech and played music. It ran on CosmOS, Humane's Android-based operating system, which routed each request to cloud models, chiefly OpenAI's (reported). It needed its own phone number and a $24-a-month plan on T-Mobile's network to work at all.

The founders, Imran Chaudhri and Bethany Bongiorno, had spent roughly two decades at Apple, Chaudhri in the human-interface team behind the original iPhone. They founded Humane in San Francisco in 2018 on a thesis that predated ChatGPT by four years: the smartphone had trapped people in screens, and ambient computing driven by AI could replace much of what the phone did. About 40% of its ~200 staff in 2023 were former Apple employees (reported).

How much money did Humane raise?

About $230 million by November 2023 (reported), including a $100 million Series B in September 2021 and a $100 million Series C led by Kindred Ventures in March 2023. Investors included Sam Altman, Marc Benioff, Tiger Global and SoftBank, and strategic investors that were also partners, among them Microsoft, Qualcomm Ventures and LG Technology Ventures. A valuation of about $850 million was widely quoted as its high-water mark (reported; the round that set it is unclear). The money was raised on the founders' reputations and a secret product, not on evidence from users.

Why did the Humane Ai Pin fail?

It lost to the phone in every pocket

Every job the pin attempted, calls, messages, questions, photos and translation, a smartphone already did faster, with a screen, apps and a better battery. Phones then added the same AI models in 2024–25: Apple integrated ChatGPT into Siri in December 2024, and Google and Samsung built Gemini and Galaxy AI into Android. The research's verdict is blunt: Humane is the AI era's clearest case of a device built to replace the phone that could not beat it at any job. Reviewers in April 2024 found the pin slow, taking several seconds per answer, often wrong, prone to overheating, short on battery life and hard to read in daylight.

Cloud AI on every query added cost and delay

Unlike a phone app, the pin ran little AI locally. Almost every request traveled to cloud models over a cellular link, so each query carried a model and network cost that the subscription had to cover, plus a delay reviewers noticed. At ~10,000 units Humane had no scale to negotiate those costs down. The inference economics study explains why every AI answer has a marginal cost that ordinary software does not.

Returns wiped out the revenue

The Verge reported in August 2024 that between May and August more pins were returned than bought, and that returned units could not be refurbished and resold because of how they were tied to T-Mobile's systems (reported). A device business with negative net sales in a quarter cannot cover fixed costs at any price. At about 10,000 units, device revenue was in the single-digit millions of dollars before returns (stylized: 10,000 × $699 ≈ $7 million), for a company that had raised ~$230 million and employed ~200 people.

Secrecy hid the problems until launch

Humane kept the product hidden for five years and showed it first in a TED talk and at Paris Fashion Week rather than to reviewers. Several 2024 reports described a culture in which doubts were unwelcome: a senior engineer was reportedly fired after asking whether the device would be ready, and executives reportedly chilled pins with ice packs before demos to stop them overheating (New York Times, June 2024, reported). Humane's own launch video contained factual errors in the pin's answers. Five years of stealth meant it shipped without the user feedback that might have revealed the problems earlier.

Could a different price have saved it?

The book tests this with a stylized model of AI device economics, not based on any company's figures. A company has $50 million a year of fixed costs. A premium device sells for $699, costs $500, and carries a $24 subscription that costs $14 to serve. A returned unit is refunded and cannot be resold.

The subscription can carry the cloud bill, but the margin is so exposed to returns that a product buyers send back cannot be rescued by price. Even the good case needs about 145,000 units a year. The pin shipped about 10,000. Cutting the price to $499 in October 2024 did not revive sales.

The same model shows why Rabbit, the other famous AI gadget of 2024, had the opposite problem. Its R1 sold for $199 with no subscription and shipped about 130,000 units by June 2024 (Forbes). But every query ran on paid model APIs in Rabbit's cloud, so each active owner was a recurring cost funded by one sale. The cheap device is profitable only if most owners stop using it. Rabbit survived by becoming an agent-software company.

What did HP buy?

On 18 February 2025 HP agreed to pay $116 million for most of Humane's employees, the CosmOS platform and more than 300 patents and applications. The Ai Pin business was excluded. Sales stopped that day, refunds were offered only for units shipped on or after 15 November 2024, and the pins lost cloud service on 28 February. Customer data was deleted and owners were pointed to e-waste recycling. The founders now lead HP IQ, an AI unit inside HP.

HP's price was roughly half the capital Humane had raised (stylized: $116 million ÷ ~$230 million), before preference terms that were not disclosed, and about a seventh of the price Humane had sought nine months earlier. In late 2024 Humane had begun presenting CosmOS, not the pin, as the product, a sign the software was worth more than the hardware.

What should AI hardware companies learn from Humane?

The research draws a short rule: prove the job before building the object. A device must beat the phone at some task its buyers do daily. The AI wearables that worked took a different route. Ray-Ban Meta smart glasses added AI to an object people already wore, priced from about $299 (reported). OpenAI paid about $6.5 billion for Jony Ive's io in May 2025 to try again with its own models and far more capital.

In terms of The Intelligence Equation, Humane failed on the first factor, capability that survives real work, and the others followed. It spent its capital before testing whether the pin's capability survived daily use, so the second factor, unit economics at real usage, never had a chance, and when buyers sent the pins back, the fourth, capital to last the cycle, ran out. A related pattern runs through the AI failures, hype cycles and bubbles study.

The takeaway

The Humane Ai Pin failed because it was a $699 device with a monthly fee that did worse than the phone already in every buyer's pocket, and because five years of secrecy meant nobody outside the company tested that before launch. Cloud inference on every query added cost and delay, returns outran sales, and the company sold its software and team to HP for about half what it had raised. The lesson for AI hardware is to prove a daily job the phone cannot do before building the object. For the wider list of collapses, see the biggest AI startup failures and our guide to how AI companies make money; Chapter 24 of the book, Building an AI Company, sets Humane beside Rabbit, Builder.ai and Midjourney.