BrainCo, a Chinese start-up that claims to be able to read brain activity without any operation at all, has just raised around 280 million dollars and is preparing its stock market debut in Hong Kong. The promise: to take on Neuralink, Elon Musk’s company, without drilling a single skull. An impressive technology, a spectacular fundraising round — and, in the background, an inquiry demanded by the US Congress into the company’s links with the Chinese government and its collection of brain data from leading athletes.

An article doing the rounds presents this fundraising as a straightforward Chinese technological win over Neuralink. That is true, but very incomplete. Le Recul has checked the figures, traced the company’s history, and dug into what the original article does not say.

The technology: reading the brain without opening the skull

BrainCo was founded in 2015 in a Boston garage by Bicheng Han, then a doctoral student at Harvard, with the support of the Harvard Innovation Labs. The company moved to Hangzhou in 2018. Where Neuralink surgically implants electrodes directly into the cerebral cortex, BrainCo bets on dry-electrode sensors placed on the scalp, which capture the brain’s electrical activity from outside. The signal collected is necessarily “noisier” and less precise than an implant’s, but specifically trained artificial intelligence algorithms decode it to extract usable information — with no surgery required.

Three product families make up the company’s offering: consumer EEG headsets and headbands — including the FocusCalm model — which measure attention or relaxation; a bionic hand driven by an amputee’s residual neuromuscular signals, already approved by the American FDA; and a sleep aid device that uses low-intensity electrical pulses to stimulate neurochemicals associated with relaxation. The bionic hand remains, to date, the company’s most accomplished medical product: unlike the consumer headbands, it had to go through a genuine American regulatory pathway before going on sale, which clearly separates the purely commercial uses (concentration, sleep) from medical ones, subject to far stricter controls.

In practice, a BrainCo headband works like this: dry electrodes placed on the scalp capture the micro-variations in electricity produced by the activity of millions of neurons (electroencephalography, or EEG, a technique almost a century old). The signal passes through the skull, which heavily attenuates and blurs it — a little like listening to a conversation through a wall. An AI algorithm, trained on large quantities of similar signals, then extracts usable patterns from that noise: level of alertness, concentration or relaxation. These decoding models are nonetheless very different in nature from the consumer generative AI that Le Recul compares in its AI ranking: they generate neither text, nor images, nor code, and are the subject of no comparable public benchmark — making it impossible, to date, to say objectively whether BrainCo’s algorithm “decodes better” than a competitor’s. Neuralink does the opposite: a surgical robot inserts an array of ultra-fine electrode threads directly into the motor cortex, as close as possible to the neurons themselves, which captures a far cleaner signal but requires an operation, a surgical risk and a device implanted for life.

What the technology can really do — and what it cannot (yet)

The difference between the two approaches is not only a question of comfort: it is measured in bitrate, the amount of information the brain can transmit to a machine per second. The best current non-invasive EEG systems, with onboard AI decoding, top out at around 3 to 4 bits per second in real usage conditions; invasive systems routinely reach 50 to 100 bits per second, with some research prototypes exceeding 200 bits per second. At Neuralink, the 21 participants in the Telepathy trials — against 3 in 2024 and 12 in September 2025 — now reach between 8 and 10 bits per second, the equivalent of a sighted, able-bodied person’s computer mouse use, with typing speeds that can reach 40 words per minute by simply thinking about finger movements. A new implant, announced for the end of 2026, would target three times that throughput.

Neuralink is also developing Blindsight, an implant aiming to restore a form of vision in blind people, including those blind from birth: the company announced in late January 2026 that it was ready for a first human trial, pending final regulatory authorisation. Elon Musk himself warns that the vision obtained will at first be very low resolution — he compares it to Atari video game graphics — before improving with the hardware and the brain’s own learning. The stated long-term ambition: to combine movement, speech and vision in the same person, with implants of more than 10,000 capture channels, against a few hundred today.

None of this allows, at Neuralink or at BrainCo, “thought-reading” in the sense often implied: no current system decodes complex inner speech, a memory or a precise emotion. What is decoded are intentions to move, states of alertness or attention, or residual motor signals — much narrower categories of information than the word “telepathy” suggests, even though their medical usefulness is very real.

The uses, today and tomorrow: for whom, why, how

The first field of application, and by far the most legitimate, remains medical. For a paralysed person, a brain-computer interface can restore the ability to move a cursor, write, or operate a robotic arm. For a blind person, it could one day restore a rudimentary perception of light and shapes. For an amputee, as with BrainCo’s bionic hand, it translates a residual intention to move into a real gesture. China itself is advancing on this ground by a different route from BrainCo: in March 2026, the Shanghai company Neuracle Technology, linked to Tsinghua University, obtained the world’s first marketing authorisation for an invasive brain implant, named NEO — a coin-sized device placed on the dura mater (the brain’s protective membrane, so shallower than Neuralink’s electrodes, which penetrate the cortex directly), intended for patients aged 18 to 60 paralysed by a spinal cord injury but retaining some arm mobility. After eleven months of rehabilitation at two and a half hours a day, one patient, Dong Hui, regained enough control to rewrite his own name. Authorised within months and immediately added to the Chinese health insurance system, NEO illustrates an approval pace bearing no comparison with the American regulatory path, which generally takes several years — a difference reflecting less a technical superiority than a political trade-off between speed of deployment and health caution.

The second field, blurrier, is wellbeing and individual performance: concentration at work or school, stress management, sleep, athletic training — exactly the space occupied by BrainCo’s consumer products and by the uses made of them by Shiffrin, Sinner or Leclerc. It is also the least regulated: unlike a medical device, a wellbeing headband does not have to prove its clinical effectiveness before going on sale.

The third field is the one China openly embraces at national level. A joint strategic plan from seven ministries, approved by the State Council, elevated brain-computer interfaces to the rank of a strategic future industry, alongside quantum computing, embodied AI, 6G and nuclear fusion. The stated objectives: major technological breakthroughs — on invasive as well as non-invasive approaches — by 2027, and the emergence of two or three world-class Chinese companies by 2030, with a gradual extension into industry, healthcare and emergency management. It is in that context that the investment in BrainCo has to be read: less an isolated bet than one piece of an openly stated national industrial policy.

That context is also what feeds the legitimate concern of the US Congress described above: a technology that captures fine neurological and muscular signals could, in theory, interest uses going well beyond wellbeing — assessing the fatigue or attention of personnel in sensitive settings, for example. No public, verified evidence demonstrates to date a direct military use of the data collected by BrainCo; at this stage, this is a risk zone identified by American lawmakers and investigative journalists, not an established fact. It is precisely that uncertainty — between documented real use and plausible unproven risk — that justifies an inquiry rather than a presumption of guilt, in either direction.

BrainCo raised around 2 billion yuan (about 280 million dollars) in a funding round co-led by IDG Capital and Walden International — the latter founded by Lip-Bu Tan, Intel’s current chief executive. The company had already been valued at 1.3 billion dollars in a pre-IPO round in August 2025. It has since filed, in January 2026, a confidential application for a Hong Kong flotation, with CICC and UBS as advising banks, targeting several hundred million dollars more and a post-listing capitalisation estimated at between 1.5 and 2 billion dollars. That is, to date, the second largest fundraising ever carried out in the brain-computer interface sector, just behind Neuralink.

Neuralink, precisely, posts even more dizzying figures: around 1.85 billion dollars raised since its creation in 2016, including 650 million dollars in a series E round in April 2026 alone. Its official valuation reached 9.6 billion dollars in May 2025, but secondary market transactions reportedly now suggest an implied value close to 40 billion dollars — a figure not officially confirmed. By the end of 2025, the company had implanted its device in a dozen patients.

It would be unfair to point only at Chinese grey areas: Neuralink has its difficulties too. In Noland Arbaugh, its very first implanted patient, the electrode threads partly retracted from the brain tissue in the months after the operation — up to around 85% of the threads affected, degrading signal quality. The company compensated for the problem with software adjustments, but the episode illustrates a real long-term biocompatibility challenge, inherent to any invasive approach. The FDA had in fact initially rejected Neuralink’s clinical trial application in 2022 on safety grounds, before granting a limited authorisation in 2023.

The record that worries: from Chinese schoolchildren to Olympic athletes

This is where BrainCo’s story takes a distinctly darker turn — and where the article doing the rounds says nothing. Back in 2019, BrainCo Focus headbands used to measure in real time the concentration levels of primary school pupils in Jinhua, in Zhejiang province, triggered a major public controversy in China itself. In an online poll of more than 3,300 internet users, 88% of respondents judged the practice pointless, even unacceptable. The local education bureau eventually suspended use of the device on 31 October 2019.

Seven years later, the company has changed targets but not really methods. A joint investigation by the outlet Hunterbrook and the podcast Pablo Torre Finds Out, published on 16 September 2025, revealed that BrainCo FocusCalm headsets had been used by top-level athletes — the American Olympic gold medal skier Mikaela Shiffrin, the world number two in tennis Jannik Sinner, and the Formula 1 driver Charles Leclerc. Asked about it, a BrainCo spokesperson said brain data was “purged from the app at the end of each use” and never retained. The problem: according to the investigation, BrainCo’s own privacy policy states in black and white that the company collects its users’ brainwaves, and that aggregated data may be transmitted to the group’s offices worldwide, or even to third parties — a direct contradiction between what the company tells the press and what it writes in its own documents.

The inquiry Washington still has not closed

Nine days after that investigation was published, on 25 September 2025, the chair of the House Energy and Commerce Committee, Brett Guthrie, together with subcommittee chairs Gus Bilirakis and John Joyce, sent an official letter to Attorney General Pam Bondi, Federal Trade Commission chair Andrew Ferguson and Commerce Secretary Howard Lutnick, demanding an inquiry into the national security risks posed by BrainCo.

The file documented by Hunterbrook is dense: BrainCo has reportedly been quietly funded for nearly a decade by entities linked to the Chinese government, including a military supplier under American sanctions. BrainCo belongs to an informal group nicknamed Hangzhou’s “Six Little Dragons” — six technology start-ups from the same Chinese city, among which also feature DeepSeek in generative AI and Unitree in robotics, presented by Beijing as the spearheads of its national technological ambition. The company has also built its technology into humanoid robots made by Unitree, which the US Congress has already described as a tool serving the Chinese People’s Liberation Army. BrainCo reportedly also cooperated with at least three of the “Seven Sons of National Defence”, those Chinese universities whose postgraduate students were barred from certain American university programmes from 2020 onwards because of their close ties to the military apparatus. Earlier this year, the company also joined the Shijingshan government centre dedicated to training humanoid robots, a Beijing facility meant to produce more than a million multimodal data points a year. Asked about it, BrainCo denied any collaboration with the Chinese military, but declined to answer questions about its ties with defence contractors or the Chinese Communist Party. To date, no public decision from the Department of Justice, the FTC or Commerce has been announced on this file.

That mix of opaque funding and collection of sensitive data recalls a mechanism Le Recul has already documented elsewhere: a company stating one thing to the press while its own documents say the opposite, as observed during the controversial launch of Meta’s Muse Image. The sector changes, the underlying logic — say the minimum, move fast, correct only under pressure — repeats.

And meanwhile, French research is advancing differently

It is hard to read this file without setting it against what is being done in France. The Clinatec laboratory, in Grenoble, has been conducting research on invasive brain-computer interfaces since 2008, funded by the CEA (France’s atomic and alternative energies commission), Grenoble University Hospital and Grenoble Alpes University — public money, then, not nine-figure fundraising rounds. Its teams notably developed the WIMAGINE implant, and published in 2023 in the journal Nature, in collaboration with Switzerland’s EPFL and CHUV, the restoration of walking in a paraplegic person thanks to a wireless digital bridge between the brain and the spinal cord. The financial scale bears no comparison with the BrainCo-Neuralink race, but the scientific recognition is very real — a reminder that public research remains a legitimate player in this sector, even without raising hundreds of millions of dollars.

The BrainCo case raises a question very few legal systems in the world currently cover clearly: who has the right to use the data produced by your brain? Chile was the pioneer here: as early as 2021, its Senate wrote a right to neural privacy into the Constitution, with very real legal force — the Chilean Supreme Court subsequently condemned the neurotechnology company Emotiv for collecting and retaining a customer’s brain data without sufficient consent, and ordered it to delete them. In the United States, Colorado and California adopted in 2024 the first state laws specifically protecting neural data, and half a dozen other states are considering similar texts. In Europe, the GDPR does not explicitly mention brain data, but it would very probably fall under the category of biometric or health data, subject to reinforced explicit consent — an interpretation shared by several law firms specialising in digital law, in the absence, to date, of a dedicated European text.

What to take away

BrainCo, a Chinese start-up founded in 2015 out of Harvard, has raised around 280 million dollars and is aiming for a Hong Kong flotation, for a post-IPO capitalisation estimated at between 1.5 and 2 billion dollars — the second largest raise in the brain-computer interface sector after Neuralink.

Its technology rests on electrodes placed on the scalp, decoded by AI, with no surgery — against surgically implanted electrodes at Neuralink, which has itself run into reliability problems (retraction of electrode threads in its first patient).

The technical gap between the two approaches is measured in bitrate: 3 to 4 bits per second for the best non-invasive systems, against 8 to 10 bits per second already reached by the 21 participants in Neuralink’s Telepathy trials, and up to 200 bits per second for the most advanced invasive systems. China is also advancing on the invasive route, with Neuracle Technology’s NEO implant, the first in the world authorised for sale (March 2026).

In 2019, BrainCo headbands intended for Chinese schoolchildren were suspended after a major public controversy (88% unfavourable opinions in a local poll).

An investigation published in September 2025 revealed that top-level athletes (Shiffrin, Sinner, Leclerc) were using BrainCo headsets, and that the company’s privacy policy contradicts its spokesperson’s public statements on data retention.

On 25 September 2025, American lawmakers requested a federal inquiry into BrainCo’s ties with Chinese government and military entities — an inquiry that remains without public conclusion to date, while the company pursues its fundraising and its flotation plan.

The figure to remember

88%.

That is the share of Chinese internet users polled in 2019 who judged the use of BrainCo headbands on schoolchildren unacceptable or pointless — a rejection that came from China itself, not from a Western anti-Chinese reflex. Seven years later, the same company is preparing to go public with Olympic athletes’ brain data on its servers, and a US Congress inquiry still open into its funding.