Neuralink Has 1,024 Electrodes and 21 Patients. Synchron Has 16 Electrodes and a Peer-Reviewed Safety Record. The Math Isn't What You Think.
Three companies are threading, puncturing, and stenting their way into the human brain using radically different surgical approaches. An original accuracy-per-electrode analysis reveals that more hardware does not automatically mean better outcomes, and the company with the most patients has published the least data.
One thousand and twenty-four. That is how many electrodes Neuralink threads into the motor cortex of each patient who receives its N1 brain-computer interface, spread across 64 ultra-thin polymer threads inserted through a hole drilled in the skull by a proprietary surgical robot. Across the competitive field, Synchron's Stentrode carries exactly 16 electrodes on a stent-like scaffold threaded through the jugular vein into the superior sagittal sinus, a major blood vessel running along the top of the brain, and the entire procedure looks more like a cardiac catheterization than neurosurgery. A 64-to-1 electrode ratio, and every breathless headline assumes more electrodes means better results. It does not, and the data to prove otherwise is hiding in plain sight.
The Accuracy-Per-Electrode Calculation Nobody Ran
Synchron's SWITCH trial (NCT03834857), a single-center prospective study published with full 12-month follow-up data, reported 2-class brain signal decoding accuracy of 85.2% (standard deviation 7.9%) across four implanted patients, all with ALS or primary lateral sclerosis. At 5.33% accuracy per electrode for 2-class decoding, Synchron extracts a remarkable return from minimal hardware. For 3-class decoding, accuracy dropped to 68.3% (SD 12.5%), or 4.27% per electrode. Typing speed hit 16.6 correct characters per minute at 93.9% selection accuracy.
Neuralink's accuracy-per-electrode figure remains unknown because the company has not published it. Neuralink has implanted 21 patients across four countries (the United States, Canada, the United Kingdom, and the UAE) as of January 28, 2026, and logged more than 15,000 hours of neural recording across the cohort. Noland Arbaugh, the first recipient, played Civilization VI for marathon sessions and moved a cursor across a screen with his thoughts. Brad Smith and Kenneth Shock, enrolled in the VOICE trial for thought-to-speech conversion, have demonstrated the ability to generate synthesized speech from intended words. These are extraordinary feats, but none of them have been published in a peer-reviewed journal with reproducible methodology, statistical reporting, or independent verification.
Twenty-one patients, fifteen thousand hours, zero peer-reviewed papers.
Four patients, twelve months, full dataset published.
That is the transparency gap at the center of brain-computer interface medicine in 2026.
The Surgical Invasiveness Spectrum
Three distinct approaches to getting electrodes near neurons are now in human trials, and they occupy radically different positions on the risk-capability frontier.
| Dimension | Neuralink N1 | Synchron Stentrode | USC/Caltech Bidirectional |
|---|---|---|---|
| Electrode count | 1,024 | 16 | ~100 (clinical ECoG grid) |
| Placement | Penetrating (into cortex) | Endovascular (inside blood vessel) | Surface ECoG (on cortex surface) |
| Surgery | Craniotomy + robotic insertion | Jugular vein catheterization | Craniotomy (clinical, not device-specific) |
| Anesthesia | General | General (shorter) | Varies (clinical context) |
| Recovery | Days (hospital) | ~24 hours (observation) | N/A (epilepsy monitoring) |
| Signal type | Single-unit spikes + LFP | Local field potentials only | ECoG (cortical surface potentials) |
| Direction | Read only | Read only | Read + write (bidirectional) |
| Published accuracy | Not peer-reviewed | 85.2% (2-class) | 92% (read + stimulate) |
| Published safety data | None | Zero serious adverse events (12 mo) | Proof of concept (1 patient) |
| Patients implanted | 21 | 4 (SWITCH) + next-gen trial starting | 1 (using existing clinical electrodes) |
| Peer-reviewed publications | 0 (human clinical) | Yes (12-month follow-up) | Yes (Brain Stimulation, April 2026) |
The third column deserves attention it has not received. In April 2026, a team led by Charles Liu at the USC Neurorestoration Center published in Brain Stimulation (DOI: 10.1016/j.brs.2026.103065) the first full test of a bidirectional brain-computer interface for walking. Their system achieved 92% accuracy for both reading step intention signals from the motor cortex and delivering walking sensation back to the sensory cortex via electrical stimulation, all while the patient controlled a robotic exoskeleton worn by another person. One patient. One epilepsy monitoring session using clinically implanted electrodes. But the concept it proved is the one neither Neuralink nor Synchron has demonstrated: closing the loop. Every current commercial BCI reads from the brain, but this one reads and writes.
The Thread Retraction Problem
Noland Arbaugh's implant suffered 85% thread retraction within the first month after surgery. Eighty-five percent. Neuralink's polymer threads pulled back from their intended positions in the cortex, reducing the number of functional electrode channels from 1,024 to roughly 150. Neuralink's engineering team responded with a software update that redistributed signal processing across the remaining channels, and Arbaugh retained cursor control. Impressive, but the failure mode was not trivial.
What does 85% retraction look like at scale? Neuralink has since modified its surgical technique to prevent recurrence, and subsequent patients have not reported the same issue publicly. But here is the arithmetic that matters: if the N1's advantage over endovascular approaches depends on having 64 times more electrodes in the brain, and a failure mode exists that can reduce functional electrode count by 85% in a single patient, then the effective electrode advantage after retraction is 1,024 multiplied by 0.15, which equals approximately 154 electrodes. That is still 9.6 times Synchron's 16, but the gap between "64 times better" and "9.6 times better" is the distance between a marketing claim and an engineering reality.
Synchron's Stentrode, by contrast, showed a mean device position shift of 0.45 millimeters over 12 months in the SWITCH trial, which is clinically negligible according to the published data. It stayed where the surgeon put it. Signal bandwidth held at a mean of 233 Hz throughout the follow-up period, and zero device-related serious adverse events were recorded. That stability matters because a BCI that works reliably for years is more valuable to a paralysis patient than one that works brilliantly for weeks.
The Apple Variable
In May 2025, Apple introduced a native BCI HID (Human Interface Device) profile in iOS 19 and visionOS 3, recognizing neural interfaces as a fourth input category alongside touch, voice, and typing. Synchron was the first company to integrate, and Mark Jackson, an ALS patient in Pittsburgh, tested thought-driven control of an iPhone, iPad, and Apple Vision Pro using his Stentrode.
This is not a gadget demo. It is an accessibility infrastructure decision by the most valuable company on Earth, and it validates the endovascular approach for a specific, enormous use case: giving people with motor impairments native access to the devices that run modern life. You do not need 1,024 electrodes to select an app, compose a message, or navigate a menu. You need reliable 2-class and 3-class decoding with very low false-positive rates. Synchron's 16 electrodes, paired with Apple's accessibility stack, may be sufficient for the first mass-market BCI application.
Synchron's Next Move
Inside BCI reported in March 2026 that Synchron has returned to Melbourne to begin the first trial of its next-generation Stentrode, promising a richer vocabulary of thought-driven inputs beyond the 2-class and 3-class decoding demonstrated in SWITCH. Synchron has also announced Chiral, a foundation model for human cognition built on NVIDIA infrastructure, trained on neural signal data. Synchron's total funding stands at $145 million (Series C led by ARCH Venture Partners and Bezos Expeditions), against Neuralink's roughly $8 billion valuation from its 2024 round.
A funding gap of 55 to 1, yet the publication gap runs the other direction.
Strongest Counterargument
Neuralink's electrode density gives it fundamentally higher signal resolution, and that resolution ceiling matters enormously for the most ambitious BCI applications. Noland Arbaugh playing Civilization VI for eight-hour sessions is evidence of a decoding fidelity that 16 endovascular electrodes measuring local field potentials through a blood vessel wall may never match, regardless of how much machine learning Synchron applies to the signal. Neuralink's VOICE trial thought-to-speech demonstrations require capturing phoneme-level neural representations, which almost certainly demands the single-unit spike recordings that only penetrating electrodes can provide. For restoring vision (Neuralink's Blindsight program, which received FDA Breakthrough Device Designation), for enabling complex robotic limb control with individual finger movements, for any application requiring hundreds of independent control channels, you probably need electrodes inside the brain, not watching from a vein. Synchron's published safety record is admirable, but publishing data from four patients in a 2-class decoding task is a lower bar than what Neuralink is attempting. An asymmetric comparison: Synchron is demonstrating that a minimally invasive BCI can do simple things safely, while Neuralink is betting that a maximally invasive BCI can do extraordinary things at scale. Criticizing Neuralink for not publishing is fair. Concluding that Synchron's approach is therefore superior conflates transparency with capability.
Limitations
The accuracy-per-electrode calculation compares fundamentally different signal types (single-unit spikes versus local field potentials versus ECoG), different tasks (cursor control versus binary classification versus walking intention), and different patient populations. It is a useful heuristic, not a controlled comparison. Neuralink's unpublished data means our analysis is necessarily incomplete: the company may have safety and efficacy numbers that would change the picture if disclosed. Synchron's SWITCH trial enrolled four patients, a sample far too small for generalizable safety conclusions, and the 12-month window is short relative to the decades a paralysis patient needs a device to function. USC and Caltech's bidirectional BCI tested in one epilepsy patient using clinically implanted electrodes, not a purpose-built device, and the gap between a proof of concept and a commercial product measured in both years and hundreds of millions of dollars is vast. Neuralink may well have solved the thread retraction problem in subsequent patients, but we cannot confirm this because the data is not public. Our claim that Arbaugh's retraction reduced functional channels to approximately 154 assumes a uniform distribution of retracted threads, which may not reflect the actual pattern.
What You Can Do
If you or someone you love has ALS or paralysis: The SWITCH trial results establish that endovascular BCIs can be implanted safely with meaningful typing and device control. Synchron's patient enrollment page is the starting point. Neuralink's patient registry is expanding across four countries. Both are legitimate paths. Ask your neurologist to discuss the trade-offs between surgical invasiveness and signal resolution specific to your functional goals.
If you evaluate medical devices professionally: Demand published data. In every other therapeutic area, from cardiac pacemakers to cochlear implants, is peer-reviewed evidence before scaling. Twenty-one implanted patients and zero peer-reviewed publications is not a regulatory violation, but it is a red flag for any institution considering enrollment or referral. Pressure your IRB to require publication timelines as a condition of trial participation agreements.
If you invest in neurotechnology: The BCI market is projected at $3.3 billion by 2026 (ResearchAndMarkets, 2021 estimate). Watch for two inflection signals: first, whether Neuralink publishes clinical data before scaling to "high-volume production" as Elon Musk announced for 2026, and second, whether Synchron's next-generation Stentrode can close the decoding gap from 2-class to 5-class and beyond without increasing electrode count. Whichever company solves the regulatory pathway to commercial approval first, not the company with the most electrodes or the largest valuation, will define the market.
If you follow this field casually: Watch three numbers. Neuralink's first peer-reviewed publication date. Synchron's next-gen decoding class count. And the USC bidirectional BCI's patient enrollment target. When any of those changes, the field shifts.
Bottom Line
Brain-computer interfaces are entering a phase where the competition that matters is not between electrode counts but between surgical philosophies, each with a different answer to the question of how much risk a patient should accept for how much capability. Neuralink is betting that penetrating the cortex with 1,024 electrodes unlocks capabilities no other approach can match, and the early patient demonstrations support that bet. Synchron is betting that threading a stent through a blood vessel with 16 electrodes delivers enough capability with dramatically less risk, and the published safety data supports that bet. USC and Caltech are betting that the entire field has the direction wrong, that a BCI worth building must talk back to the brain, not just listen. Twenty-one patients, four patients, one patient. Three approaches, three risk profiles, three publication records. The electrode count is the least interesting number in the room.
Sources: NeurologyLive, Synchron SWITCH trial 12-month results; TamilTech, Neuralink 21 patients compilation (Jan 2026); MedicalXpress, USC/Caltech bidirectional BCI (April 2026); MacRumors, Apple BCI HID profile (May 2025); Medical Economics, Synchron Apple integration; Inside BCI, Synchron next-gen Stentrode trial (March 2026); UCLH, seven GB-PRIME patients (2026); STAT News, Neuralink vision vs. reality (Jan 2026); Brain Stimulation (2026), DOI: 10.1016/j.brs.2026.103065.