OpenAI paid roughly $6.5 billion for Jony Ive's hardware startup io, and supply-chain reports now describe its first product, code-named Sweet Pea, as an AI earbud built around a custom 2nm chip. The deal, struck a year after the Scarlett Johansson voice controversy, confirms what Sam Altman signaled with a single word: OpenAI wants to build Samantha, the operating-system companion from Spike Jonze's 2013 film Her, and it has concluded that she lives in the ear.

How a one-word post became a product roadmap

On May 13, 2024, OpenAI demonstrated GPT-4o, a model that could hold a spoken conversation with human timing and inflection. One voice, called Sky, was warm, quick and more than a little flirtatious. After the demo, Altman posted one word on X: "her." Asked at Salesforce's Dreamforce conference in September 2023 to name his favorite science-fiction film, Altman had picked Her. "I like Her. The things Her got right—like the whole interaction models of how people use AI—that was incredibly prophetic," he said.

A week later, Scarlett Johansson released a statement that turned the post into an embarrassment. She said Altman had asked her the previous September to voice ChatGPT and she had declined; he contacted her agent again days before the demo. After Sky shipped, friends and family could not tell the difference, and her lawyers sent two letters asking OpenAI how the voice had been made. OpenAI pulled Sky. Altman insisted the voice belonged to another actress, cast before anyone had called Johansson, and apologized for not communicating better.

That was the retreat. The strategy never moved.

Why the camera matters as much as the ear

Samantha had two bodies in the film: the earpiece, and a small bifold Theodore carried in his shirt pocket, its camera peering out so she could see the city with him. Camera earbuds are already shipping in China, and reports have Apple preparing camera AirPods, now likely for 2027.

The earpiece itself is no longer fiction. Apple sells it for $249. The AirPods Pro 3 run on the H2 chip, introduced in 2022, which Apple says uses machine learning to reduce loud, intermittent noise 48,000 times a second — the audio sample rate. A new infrared heart-rate sensor pulses 256 times a second. Live Translation begins in the bud, where noise cancellation turns down the other speaker so the translation can be heard.

The rest of the industry is more forthcoming than Apple, which has never published an H2 block diagram. Bestechnic's BES2800, inside Samsung's Galaxy Buds3, publishes a design with dual Cortex-M55 cores, a dual-core NPU, a HiFi 4 DSP and 8.3MB of SRAM. Qualcomm's S7 Pro lists a 64-GOPS neural engine. Continuous inference in the ear already ships at scale — but it runs small models that clean up sound. It does not think.

Where does the mind actually live?

In every product now shipping, the earbud handles the sound and the mind sits one or two hops away. Apple's Live Translation makes a single hop: the bud cleans the audio, and an iPhone 15 Pro or newer runs recognition and translation on its Neural Engine, with no cloud involved. Shokz took the other road. Its OpenFit 2 AI went on sale in China in August at 1,398 yuan, about $195, sending conversations through the phone to Alibaba's Qwen for transcription, summaries and translation — the three-month AI subscription bundled with each pair reveals the model lives in Alibaba's data centers.

The Chinese trade press sorts AI earbuds into three groups: phone-collaborative, edge-enhanced and task-specific. The labels matter less than the pattern. A device that waits to be asked and forwards the question elsewhere is Siri with better microphones. Samantha notices when Theodore goes quiet.

Apple owns the ear and rents the mind

Apple tried to build Samantha first and called her Siri, launched in 2011 from a 2010 acquisition of an SRI International spinout. Apple Intelligence, announced in June 2024, was supposed to fix her. The personal Siri slipped, and Apple delayed it publicly in March 2025. That December, AI chief John Giannandrea announced his retirement, and foundation models passed to Amar Subramanya, who had led engineering on Google's Gemini assistant. In January Apple turned to Google for its server-tier models; the rebuilt Siri that shipped with iOS 27 this fall stands on that foundation. In July, Chinese regulators approved Apple Intelligence with Alibaba's Qwen and Baidu supplying the models there.

Apple has licensed bridges before — Google Maps until Apple Maps, Intel processors until Apple Silicon. Those ended because a replacement team was in place. This time the team left.

Samantha needs an operating system

Most earbuds run firmware, not an operating system. Nobody outside Apple can load code on the H2; Bestechnic's chips ship under NDA. Two companies are trying to change that. Bragi, the Munich pioneer of true wireless earbuds, licenses a turnkey platform that powers Skullcandy's smart features and Bose's noise cancellation. Sonical, a San Mateo startup founded in 2020 by Gary Spittle, a veteran of CSR, Dolby and Knowles, builds the open CosmOS, with a hardware abstraction layer that lets applications run on different processors and move from cloud to phone to earbud.

Sonical licensed the HearAngel hearing-protection app from Britain's LimitEar and works with Newcastle University neuroscientist William Sedley on sound therapy for tinnitus, which affects more than 740 million adults worldwide. In 2022 Spittle conceded the software would outrun the silicon, saying CosmOS would need processor gains like those phones went through. Four years on, he argues the industry aims too low, calling voice commands for lights and volume pointless and of zero value. He expects several resident AI agents rather than one assistant, with the cloud called on only for specific tasks — the smartphone model. "The same is about to happen with the ear," he says. That takes AI-first chips with Bluetooth as just a data pipe; the resulting category, he argues, will make today's Bluetooth earbud market look tiny.

Samantha needs silicon

An October survey by 半导体产业纵横 ("Semiconductor Industry Overview"), republished in English by 36Kr, found two design paths: an NPU inside the audio SoC, or a separate AI audio chip paired with Bluetooth. Bestechnic says the BES2800 has up to four times the NPU performance of its predecessor. Anker and CIX claim their Thus A1 reaches 5 GFLOPS, against roughly 30 MFLOPS for a conventional Bluetooth audio chip. ByteDance's Doubao earbuds carry two 128-megabit flash parts per bud, against a typical single 32- or 64-megabit part.

None of that is Samantha. An NPU is built for bursts — wake, run, sleep. Samantha is continuous, and the binding constraint is memory, not arithmetic: the best earbud SoC holds 8.3MB of SRAM, while a conversational model holds hundreds of megabytes even compressed. Samantha will therefore be distributed, much as Jonze drew her — the ear keeps listening, short-term state and the decision to speak; the pocket device holds the large model and the camera. The silicon that wins sustains throughput per milliwatt with memory next to compute, favoring vector engines and dense on-die memory over another burst accelerator bolted onto a Bluetooth chip.

Who owns her?

OpenAI wants to own the mind and is buying a body. Google has a mind, a phone and an earbud and has yet to assemble them. Apple owns the ear and rents the mind. Meta stumbled closest to Jonze's split, then put the sound in open speakers at the temples and called it audio. Altman's post was one word long, and the backpedal took a week; the race it announced will run for years. How quickly a company can put a mind in the ear that the wearer trusts will decide who owns her.