Neuralink: A Full Guide to the N1 Implant, R1 Robot, Clinical Trials, Blindsight and Limits

On October 1, 2026, Neuralink said one of its trial participants had moved a computer cursor at 11.32 bits per second, a new record for the company and above the 8 to 10 bits per second that able-bodied people manage with a mouse. Two and a half years earlier, its first patient was still learning to click.

Neuralink is the brain-computer interface company Elon Musk co-founded in 2016. It makes the N1, a coin-sized wireless implant, and the R1, the robot that sews its threads into the brain. It's also the most hyped name in neurotech, which makes it hard to tell what is working from what is a slide.

This full guide covers how the N1 and R1 work, every registered Neuralink trial, the performance data, Blindsight and speech, the money, how it stacks up against Paradromics, Synchron and Precision Neuroscience, and the honest limits, all as of October 25, 2026.

Last updated: October 25, 2026.

Key Takeaways

• Neuralink's N1 implant records from 1,024 electrodes on 64 flexible threads, placed by the R1 surgical robot and charged wirelessly.
• The company reported 21 participants enrolled across its trials by January 2026, with zero serious device-related adverse events.
• A participant set an 11.32 bits per second cursor record in 2026, helped by a model pretrained on more than 50,000 hours of neural data.
• Investors put $650 million into a Series E in June 2025, and secondary trades have valued Neuralink at up to $42 billion.
• Honest limit: the N1 is investigational and not FDA-approved, Blindsight has no confirmed human implant yet, and rivals are close behind.

What Neuralink actually builds

Neuralink builds an implanted brain-computer interface, or BCI. It's a device that reads the electrical activity of neurons and turns it into commands, so a person who can't move their hands can still drive a cursor, type or steer a robotic arm.

The system has three parts. The N1 implant sits flush in the skull. The R1 robot inserts the implant's threads into the motor cortex. An app called Telepathy decodes the signals on a phone or computer.

The first target is people with paralysis. Neuralink points to roughly 18,000 new spinal cord injuries a year in the U.S. and about 302,000 people living with one. Its later work aims at speech loss from ALS and stroke, and at blindness.

Inside Deep Tech covers BCIs as part of the wider physical AI stack, the same way it covers robot foundation models and humanoid robots. A BCI is a sensor that happens to live inside a skull, and the hard part is increasingly the AI that reads it.

Neuralink N1 and R1 specs at a glance

Here's what Neuralink has published about the hardware. Numbers come from the company's own study updates and technology page.

How the N1 implant and R1 robot work

The N1 implant

The N1 replaces a disc of skull bone. From its underside, 64 threads carrying 1,024 electrodes run into the part of the motor cortex that plans movement. Each thread is far thinner than a hair, which is the point: flexible threads should move with the brain instead of scarring it the way rigid needle arrays can.

The chips inside amplify and digitize neural spikes, then send them wirelessly to the Telepathy app. There's no cable through the scalp. That matters for daily use, because older research BCIs such as the Utah array setups needed a connector bolted to the head and a lab cart.

The R1 surgical robot

A human surgeon can't place 64 threads between blood vessels by hand. The R1 can. Neuralink says it uses a five-camera head with optical coherence tomography and a needle thinner than a human hair to target each insertion and dodge visible vessels.

The next version of the procedure aims to insert threads through the dura, the tough membrane over the brain, instead of opening it. That would make surgery shorter and less invasive, and it's one of the quieter upgrades that matters most for scale.

Telepathy and the decoder

The app is where most of the progress since 2024 has come from. A decoder model maps spike patterns to intended movement. Early users had to recalibrate often. In October 2026, Neuralink said its pretrained models cut average calibration to 55 minutes a week, with some users going from 10 minutes a day to 10 minutes a week.

Every registered Neuralink clinical trial

Neuralink's FDA approval to begin human trials came in May 2023, with recruitment opening that September. Since then, the trial map has grown to studies in four countries: the U.S., Canada, the UK and the UAE.

The registry is the cleanest way to see the program, because each study has a public record with a target enrollment and completion date.

Two things stand out. Every study is small, with single-digit or low double-digit enrollment. And they're all early feasibility or first-in-human designs, built to show safety and basic function, not to win approval. The UK leg launched after approval from the MHRA in July 2025, and the robotic-arm study CONVOY launched in November 2024.

By its two-year mark, Neuralink said it had 21 participants enrolled worldwide. The pace was steep: one implant in the first half of 2024, two in the second half, four in the first half of 2025 and 13 in the second half of 2025.

What the performance data shows

Neuralink measures cursor control in bits per second, or BPS, on a grid-clicking task. Higher means faster, more accurate control. It's a good metric because it's comparable across users and studies.

The trajectory is the story. Noland Arbaugh, the first participant, hit 4.6 BPS in his first session and 8.0 BPS soon after. By 2026 the median participant sat around 10 BPS, roughly mouse speed, and six participants set personal bests in the pretraining study.

Usage tells you more than records do. In the first months, Noland used his device for 69 hours in one week, and up to about 10 hours a day on weekends. People don't do that with a toy. They do it with a tool that gives back independence.

The thread retraction setback

It hasn't been smooth. In early 2024, a number of threads in the first implant retracted from the brain, which cut BPS. Neuralink recovered performance with algorithm changes rather than surgery. Later participants got a revised procedure, and the company says 18 of 20 later participants showed higher signal quality.

The safety line so far is clean on paper. Neuralink reports zero serious device-related adverse events across its participants. That's an early-feasibility number from a sponsor, though, not a long-term independent safety record.

The AI layer: pretraining on 50,000 hours of brain data

The October 2026 update is the most technically important thing Neuralink has published. It treats neural data like language data: pretrain a large model on everything, then fine-tune per person.

• Scale: more than 50,000 hours of unlabeled neural recordings, including over 9,000 hours and 22.4 billion spikes from the first participant.
• Architecture: a Mamba2-based encoder, a state-space model suited to long time series, per Neuralink.
• Stability: one participant got usable control from a calibration done 20 months earlier. A previous robotic-arm decoder had become unusable within 6 days.
• Transfer: a model trained on one participant worked for another with 99% of the weights frozen.

This links BCIs to the rest of AI. The same pretrain-then-adapt recipe drives robot foundation models, and it explains why Neuralink can improve users' performance with software updates instead of new surgery.

There's a catch Neuralink states itself: multi-participant pooled models don't yet beat single-participant models online. The general brain model isn't here yet. An independent read from Inside BCI frames the release as a calibration win more than a raw-speed leap, which is fair.

Beyond the cursor: robotic arms, speech and Blindsight

Robotic arm control

The CONVOY study tests whether N1 users can control an assistive robotic arm. Neuralink's two-year update showed a participant, Nick, controlling a robotic arm. Arm control is a much harder decoding problem than a cursor, because it needs many degrees of freedom at once.

Speech and the VOICE trial

In May 2025 the FDA gave Neuralink Breakthrough Device Designation for speech restoration, covering ALS, stroke, spinal cord injury, cerebral palsy and multiple sclerosis. The VOICE trial at UT Southwestern started in October 2025 with a target of six participants.

Neuralink's stated speed goal for VOICE is 140 words per minute, close to natural conversation. The urgency is real: the company notes that up to 95% of people with ALS decline ventilation, often because losing speech is so isolating.

Blindsight

Blindsight is Neuralink's visual prosthesis. It would stimulate the visual cortex directly to produce images, bypassing damaged eyes and optic nerves. The FDA granted it Breakthrough Device Designation in September 2024.

As of late October 2026, Neuralink's updates page lists no Blindsight human implant. Treat Blindsight as a designated, not demonstrated, product until a trial record and a first-patient update appear.

Funding and valuation

Neuralink is private, so the numbers come from its own announcements and press reports.

• Series C: $205 million in July 2021.
• Series D: $280 million in August 2023.
• Series E: $650 million in June 2025, from investors including ARK, Founders Fund, Lightspeed, Sequoia, Thrive and Valor.
• Secondary market: Business Insider reported in July 2026 that shares traded at valuations between $29 billion and $42 billion, with the last priced round near $9 billion.

For context, at the time of its Series E, Neuralink had five participants and trials in three countries. Investors are paying for the platform and the data, not current revenue. There isn't any product revenue yet.

Neuralink vs Paradromics, Synchron and Precision Neuroscience

Neuralink isn't alone, and 2026 is the year the field got crowded. Each rival makes a different trade between signal quality and surgical risk.

Paradromics is the closest like-for-like rival. It got an FDA Investigational Device Exemption in November 2025, implanted its first participant at the University of Michigan in June 2026, and in September 2026 reported real-time speech decoding with its Connect-One participant. Its Connect-One registry entry is small, but speech is exactly where Neuralink wants to win.

Synchron avoids open-brain surgery by threading its Stentrode through the jugular vein into a vessel near the motor cortex. Signal quality is lower, but surgery is far simpler. Its INTENT study began in May 2026 across sites including Mayo Clinic and Mount Sinai.

Precision Neuroscience took a regulatory shortcut. Its Layer 7-T surface array won FDA 510(k) clearance for temporary use of up to 30 days. It's a foothold in hospitals, not a permanent implant.

Inside Deep Tech's take: Neuralink leads on participant count, data and decoding software. Paradromics may beat it to a convincing speech result. Synchron could win on ease of surgery. The field won't be winner-take-all in the 2020s.

Honest limits: where the Neuralink story can break

1. It's still an experiment

Every Neuralink study is an early feasibility design, and the device is investigational and not FDA-approved. A pivotal trial, the kind that supports approval, hasn't been registered. Approval before the late 2020s would be fast for any implanted neural device.

2. Long-term durability is unknown

The first implant is under three years old. Thread retraction showed that the brain-device interface can change over time. Nobody knows yet how N1 signals hold up over 10 years, or how a battery-powered implant gets replaced.

3. Most data comes from Neuralink

Nearly all the performance numbers in this guide come from company blog posts, not peer-reviewed papers. The ClinicalTrials.gov records give structure, but outcomes there are sparse. Independent replication is the missing piece.

4. Small samples, narrow patients

Twenty-odd participants with similar conditions don't tell you how the device works across ages, injury types or brain anatomies. Multi-country trials help, but the sample is still tiny.

5. Reimbursement and access

Even an approved BCI needs surgeons, follow-up care and payers willing to cover it. There's no reimbursement path for implanted BCIs today, and that may slow adoption more than the technology does.

What Neuralink means for you

• If you're a patient or caregiver: Neuralink trials run in the U.S., Canada, the UK and the UAE. Check the PRIME registry entry for eligibility.
• If you invest in deep tech: watch for a pivotal trial registration. That, not a new record, is the signal that approval is in sight.
• If you build AI or robotics: Neuralink's pretraining work is a live case study in foundation models for biosignals. It sits next to neuromorphic chips as a place where brains and silicon meet.

For the wider market picture, see the Physical AI Market Report: October 2026 and Inside Deep Tech's robot foundation models guide.

Humanoid Robots 2026: A Full Guide

Every major humanoid robot in 2026, from Figure and Tesla Optimus to Unitree, with specs, prices, deployments and honest limits.

Inside Deep Tech

FAQ

What to watch next

Three signals will show whether Neuralink turns its lead into a product. First, the jump to about 3,000 electrodes and through-the-dura surgery. Second, the first VOICE speech results against Paradromics. Third, a pivotal trial registration, which would put approval on a real calendar.

If you follow AI, keep Neuralink on your list. The next big foundation model might be trained on spikes, not tokens.