China implanted a brain chip in a paying patient. The breakthrough was the billing code.

On Monday 13 July, at Huashan Hospital in Shanghai, surgeons lifted a disc of skull from a man who had not closed his hand in ten years and slid a coin-sized titanium device into the space above the membrane that covers his brain. He had broken his neck in a car accident a decade ago. The device is called NEO. He is the first person in the world to receive one as a prescription rather than as a trial participant.

It is a genuinely remarkable thing, and it is worth being precise about what it was.

Here is what was reported from the operating theatre that day: the signals coming off the electrodes were stable and of good quality, and the patient's vital signs were stable afterwards. That is the whole of it. He did not grasp anything. No function was restored on Monday, and none was expected to be. Chinese coverage has been unusually firm on this point — Beijing News went out of its way to push back on the idea that this is technology that works the moment it is installed. Meaningful improvement is reported to require something in the order of six hours of training a day for more than a month before the patient can operate the system independently at home.

So the milestone is real. It is simply not the milestone most of the coverage has described, and the actual one is more interesting.

What the device actually does

NEO's registered name is the least glamorous and most accurate description available: an implantable hand motor function compensation system. It was developed by Neuracle Technology of Shanghai, a company founded in 2011 by two Tsinghua biomedical engineering PhDs, working with a team led by Professor Hong Bo in Tsinghua's biomedical engineering department.

The implant sits in the epidural space. That means it is under the skull but resting on top of the dura mater, the tough outer membrane enclosing the brain. It does not pierce the dura. It does not enter brain tissue. There are eight electrodes, and they read field potentials — the summed electrical chatter of large populations of neurons — rather than the firing of individual cells. There is no battery: power and data pass wirelessly through the skin by inductive coupling to an external coil, held against the scalp by a magnet.

When the patient imagines closing his hand, the electrodes pick up the corresponding pattern over the hand area of the sensorimotor cortex, and a decoder converts it into a command. That command does not go to his arm. It goes to a pneumatic soft robotic glove worn over the hand, which inflates and closes the fingers for him.

This distinction gets lost constantly and it matters enormously. The device does not repair the spinal cord, reconnect the brain to the muscles, or restore native movement. It reads an intention and hands it to a machine that performs the motion on the body's behalf. The approved indication reflects that honestly: adults aged 18 to 60 with tetraplegia from injury between the second and sixth cervical vertebrae who retain some residual upper-arm function but cannot grasp. The arm has to already work. NEO supplies the hand.

There is even a live disagreement about what to call it. China's regulator and state press describe NEO as invasive. A landscape analysis of the Chinese BCI field, published on arXiv by a large group of Chinese researchers five days before this surgery, classifies the epidural approach as semi-invasive. Scientific American describes it as sitting on the surface of the brain's outer membrane. All three are describing the same geometry; the labels carry different implications about risk, and none of them is wrong.

It bought its approval with the electrodes it left out.

Eight electrodes against a thousand

Neuralink's N1 carries 1,024 electrodes on 64 flexible threads driven directly into the cortex. Precision Neuroscience's thin-film array also carries 1,024, laid beneath the dura. NEO has eight, and does not go beneath anything.

The temptation is to read that as China shipping something crude. The more useful reading is that it is a deliberate trade, and that the trade is the reason this device exists commercially and the others do not.

Channel count buys bandwidth, and bandwidth buys dexterity. Paradromics — an American competitor building a penetrating intracortical system, and therefore arguing against its own commercial interest here — makes the point plainly: even high-density surface arrays lack the data bandwidth and resolution of intracortical systems. NEO can support grasp. It cannot support the fine multi-degree-of-freedom robotic limb control that penetrating arrays are reaching for. Nobody serious disputes this, including Neuracle.

What staying above the dura buys instead is a much gentler risk profile: markedly lower risk of haemorrhage, of the glial scarring that forms when the brain walls off a foreign body, and of the slow signal degradation that follows when it does. Those three problems are the central unsolved engineering difficulties of penetrating electrodes, and they are precisely the reasons regulators move slowly on them.

So NEO traded roughly two orders of magnitude of channel count for a surgery that never breaches the membrane — and that safety margin is exactly what made it approvable. It bought regulatory speed with bandwidth. That is not a compromise disguised as a strategy. It is a strategy, and on the evidence so far it worked.

The part that was actually first

Now the thing worth paying attention to, which is not a piece of hardware at all.

On 12 March 2025, China's National Healthcare Security Administration issued a pricing guideline for neurological services that established brain-computer interfaces as an independent billing category, with defined fees for implanting an invasive BCI, removing one, and fitting a non-invasive one.

There was no approved invasive BCI product in China at that time. There would not be one for another year. The administration wrote the billing code first, and said openly that the intent was to stabilise expected market returns for innovative companies. Provinces then filled in guided prices, generally in the range of 6,000 to 6,600 yuan — roughly $880 to $970 — for the implantation procedure. That is the surgical service only; the implant itself bills separately as a consumable.

Shanghai issued its own notice in August 2025 creating the three local service items, and included a detail that is easy to skim past and shouldn't be: BCI services were provisionally left outside the basic medical insurance scheme. The plumbing was built, and then deliberately not connected to the main supply.

It was connected on 1 July 2026, twelve days before the surgery, through a side channel. Shanghai's supplemental commercial scheme — Huimin Insurance, or 沪惠保 — announced on 26 May that its 2026 policy year would add coverage for the out-of-pocket materials cost of inpatient BCI surgery, capped at 150,000 yuan, around $22,000. The premium did not rise: 129 yuan a year, about $19, open to every enrollee in the city's basic medical insurance with no restriction on age, occupation or health status. The municipal government describes it as the first city-level inclusive insurance scheme in China to cover BCI.

Read the sequence as a whole. National billing category in March 2025. Local service items in August 2025, held outside basic insurance. Regulatory approval on 13 March 2026. Insurance price filing completed 28 April. Supplemental coverage live on 1 July. First commercial prescription on 13 July. Roughly sixteen months from billing code to billed patient, with the financial pathway laid before the clinical one.

How fast the reimbursement code itself arrived after approval is genuinely disputed — a Bird & Bird analysis says 48 hours, one industry tracker says 22 March, and Chinese state coverage documents the price filing on 28 April. The discrepancy is worth flagging rather than resolving. What none of the accounts dispute is the order of operations, and the order is the point. In the United States, a cleared neurotechnology can wait years for a reimbursement pathway. China built the pathway before the device.

What has been demonstrated, and what has been published

The approval rests on 36 procedures: a four-participant feasibility study followed by a 32-participant multicentre registration trial across eleven hospitals. The first human implant of the epidural system was in October 2023 at Xuanwu Hospital in Beijing. Reported outcomes include improvement in grasping function across participants, no serious adverse events, and — per one detailed Chinese industry account of the registration endpoints — a complete grasp-response rate at three months with the device assisting, and around 69 per cent of patients showing meaningful improvement in unaided arm function by six months. That endpoint detail rests on a single source and should be held loosely until the prospectus or a journal paper confirms it.

Which brings us to the gap in the record. The main published scientific account of this device is a preprint on medRxiv describing a single patient, first posted in September 2024. There is, so far as I can establish, no peer-reviewed publication of the pivotal trial. The preprint reports a grasp-detection F1 score averaging 0.91 sustained across nine months of use at home, which is the single most valuable number in the entire story, because it speaks to signal stability over time — and I could not retrieve the paper directly to check it.

Nine months is also, as far as the public record goes, the longest follow-up available. The registration trial reports to six. There is no published multi-year durability data on this device, because not enough time has passed for any to exist. That is not a criticism. It is simply where the evidence currently ends, and a commercial approval does not extend it.

On what timescale

The most useful caution about this technology comes from inside the building where the surgery happened. Mao Ying, president of Huashan Hospital and a NEO investigator, has said the technology cannot instantly resolve symptoms, requires extensive training and rehabilitation, and remains in early stages. That is a hospital president describing his own programme, and he is right on all three counts.

The arXiv landscape analysis, written by Chinese researchers and submitted five days before this operation, reaches the same place from the data: long-term implant stability, standardisation of clinical infrastructure and workflows, and the generalisability of decoding algorithms all remain critical barriers to widespread clinical adoption. A decoder trained on one person's cortex does not transfer cleanly to the next. Every implant so far has been, to a meaningful degree, bespoke.

There is a commercial timescale sitting underneath the clinical one, and it deserves the same scrutiny. Neuracle had implanted 32 patients by the end of 2025 and, in the last reported period, earned no revenue whatsoever from the approved device — its income came from non-invasive EEG equipment. In June it filed for a listing on the STAR Market seeking 2.5 billion yuan. A world-first approval that has produced one commercial patient is an extraordinary scientific and regulatory achievement and a company with a product-market fit of n equals one. Both descriptions are accurate. They will diverge or converge over the next two years, and which way they go is not knowable now.

The framing this story keeps getting — that China has beaten Neuralink — is the least useful thing about it. The neuroscientist Nick Ramsey put the objection well: a race implies an endpoint, and it is hard to say where that is for brain-computer interfaces. Neuralink and NEO are not running the same race. One is trying to establish how much signal you can get out of a human cortex; the other has established how little you need to sell a working product through a national health system.

The second question turns out to have been the tractable one. A man in Shanghai will spend the next month training six hours a day to close a robotic glove with his thoughts, and if it works he will have bought roughly $22,000 of that capability through a supplemental insurance policy costing $19 a year. That is not the future of neurotechnology. It is the first sentence of its accounting, which no one had written down before.