The prescription is written on an ordinary afternoon in July. A clinician at Huashan Hospital, Fudan University, records a device, an indication and a follow-up schedule, and the hospital's systems accept a procedure that had no payment code when the device was cleared in March.
The patient has tetraplegia. The implant is a coin-sized array that rests on the outside of the dura, the membrane covering the brain, without entering tissue.
What made that afternoon possible was not a better electrode. It was a pathway.
Key takeaways
- NMPA approved Neuracle's NEO on 13 March 2026 — the first implantable brain–computer interface (脑机接口) device cleared for marketing anywhere.
- The trial ran 36 cases across 11 hospitals: all showed a grasp response, and patients could sit up three days after surgery and went home within about a week.
- Commercial use began at Huashan on 13 July 2026; by August, more than 36 patients had implants and cumulative implant days passed 10,000 (People's Daily).
- In that commercial follow-up, 22 of 32 tracked patients (68.8%) showed clinically meaningful upper-limb improvement — a different measure from the trial's 100%.
- H1 2026 brought 64 BCI funding rounds worth RMB 7.25 billion (≈ US$1.0bn / HK$8bn), more than all of 2025.
What was actually approved
The device is an eight-electrode array, placed epidurally and sized about like a coin (NMPA). It is indicated for C2–C6 tetraplegia in patients aged 18 to 60. The route is semi-invasive by clinical usage; regulators still classify it as an invasive Class III implant.
That description contains the regulatory achievement. An epidural array does not penetrate brain tissue, so its surgical risk profile belongs to a class of procedure neurosurgeons already perform, not to an experimental implant. A regulator can measure it against expectations it already holds for electrode safety and removal.
The pathway did the rest. Thirty-six cases across 11 hospitals gave NMPA a multicentre dataset, and a Class III route already existed for implants that alter bodily function.
None of this required a decoding breakthrough. It required a device designed to fit a regulatory door that was already open.
The numbers, read carefully
Two figures are now circulating, and they do not measure the same thing.
In the trial — four feasibility cases plus 32 confirmatory cases — all 36 showed a grasp response and, at three months, clinically significant hand-function improvement; patients could sit up three days after surgery and generally went home within a week, and were using the system at home within a month (NMPA). In commercial use, among 32 patients tracked, 22 showed clinically meaningful upper-limb improvement (company disclosure).
The first counts a response on a trial grasp measure; the second counts a threshold of clinical meaningfulness in routine patients. A 36-patient trial, recruited to narrow criteria and treated by specialist teams, will usually outperform a commercial cohort. Reading 100% as a success rate and 69% as a decline misreads both.
What the approval did not establish:
- That function improves beyond a measured grasp score, in tasks patients name as important.
- That performance holds over years; implant days accumulate across a cohort, not within one patient.
- That outcomes transfer to patients outside the trial's age band or injury level.
- That the intervention is cost-effective against alternatives, including surgery plus conventional rehabilitation.
The payment question
The national medical-insurance administration assigned a procedure code two days after approval, on 15 March 2026, and 27 provinces have since listed the device at a procurement price of RMB 475,000 (≈ US$66,900 / HK$523,000) per set — without admitting it to basic insurance. A code is the precondition for a bill: without one, no surgeon has a reason to schedule.
A code is not a price. It does not fix a national tariff, guarantee provincial coverage, or settle who pays for electrodes, therapy time and replacement. Those decisions are where device categories usually stall.
The addressable population is large. China has about 3.7 million people living with spinal cord injury (a clinician estimate cited by CCTV). The eligible subset — C2–C6, aged 18 to 60, medically suitable, near a centre that can implant and follow up — is much smaller, and no disclosed figure sizes it. A device can be approved, coded and still reach a thin slice of the patients it was designed for.
Durability, data and the next approval
Two things now decide whether this becomes an industry or a single celebrated product.
The first is durability. Electrodes sit against tissue that moves, scars and changes with time. No published follow-up extends beyond the current cohort, and the trial's endpoint is function, not longevity. A device that degrades in five years is a different proposition for a 40-year-old patient, and for whoever pays.
The second is data. Two national standards took effect on 1 August 2026 — a reference architecture drafted by 23 organisations, and a multimodal data-format standard drafted by 25. A format for neural signals is also a decision about who may hold them, combine them and train on them. China has chosen to write the format before the market exists.
According to Li An, chief scientist at BrainNet (脑机网), the approval matters less as a scientific first than as proof that a novel device class can be given a predictable route from trial to prescription — and a reusable route is worth more than one device.
The money has already accepted that reading. In the first half of 2026, BCI companies raised RMB 7.25 billion (≈ US$1.0bn / HK$8bn) across 64 rounds, more than the whole of 2025 (industry data). Opinions issued by seven ministries in August 2025 target two to three industry clusters by 2027 and two to three globally influential companies by 2030.
Honest limitations
Trial figures come from the regulator and the company, and no control arm is described in these disclosures. The commercial follow-up — more than 36 patients, 22 of 32 improved, cumulative implant days past 10,000 — rests on company and clinical disclosures rather than an independent audit. "First cleared for marketing" is the regulator's and manufacturer's claim; no independent body has verified the precedence. Nothing here establishes how the device compares with rehabilitation alone.
What readers can do now
- Operators: model the whole episode cost, including therapy, maintenance and replacement, not the implant alone.
- Investors: discount approvals that arrive without a published durability curve; the reimbursement question is the next filter.
- Researchers: the field needs controlled comparisons and five-year outcomes more than it needs another demonstration.