Neuralink
| Neuralink Corp. | |
|---|---|
| Type | Private |
| Industry | Neurotechnology Brain–computer interfaces Medical devices |
| Founded | 2016 |
| Founder | Elon Musk, Max Hodak, Benjamin Rapoport, Dongjin Seo, Paul Merolla, Philip Sabes, Tim Gardner, Tim Hanson, Vanessa Tolosa |
| Headquarters | Fremont, California, United States |
| Key people | Elon Musk (Majority owner) Jared Birchall (President) DJ Seo (Chief operating officer) |
| Products | Telepathy (N1 implant) Blindsight R1 surgical robot |
| Employees | Several hundred |
| Website | neuralink.com |
Neuralink Corp. is an American neurotechnology company developing implantable brain–computer interfaces. It was founded in 2016 by Elon Musk and eight scientists and engineers, first became publicly known in March 2017, and is headquartered in Fremont, California, with a large facility planned at Del Valle, Texas, about ten miles east of Tesla's Gigafactory.
The company's implant, marketed as Telepathy, consists of a coin-sized sealed device placed in a recess cut into the skull, from which flexible polymer threads carrying electrodes are inserted into the motor cortex by a purpose-built surgical robot. The device reads the electrical activity of groups of neurons, transmits it wirelessly, and allows a person to control a computer by intention alone. Its first human recipient, Noland Arbaugh, was implanted on 28 January 2024, and by September 2025 Musk reported twelve trial participants with more than 15,000 cumulative hours of use.
Neuralink's stated short-term purpose is to restore function to people with paralysis, blindness and other neurological conditions. Its stated long-term purpose, which Musk has described consistently since 2017, is human enhancement and what he calls symbiosis with artificial intelligence, on the argument that the bandwidth between a human brain and a machine is the binding constraint on the relationship between the two, and that a species unable to keep pace with machine intelligence is at existential risk.
The company has been criticised on several grounds: for the number of primates that died or were euthanised during its animal programme and for the conditions documented in veterinary records obtained by campaigners; for secrecy about its human trials; for timelines that have repeatedly slipped by years; for claiming, falsely, the status of a small disadvantaged business in a 2025 federal filing; and for overstating the novelty of engineering that in several respects extends work done in academic laboratories since the early 2000s.
History
Founding and early years
Neuralink was incorporated in 2016 by Elon Musk together with Max Hodak, Benjamin Rapoport, Dongjin Seo, Paul Merolla, Philip Sabes, Tim Gardner, Tim Hanson and Vanessa Tolosa. The founding group was drawn from neuroscience, materials science, electrical engineering and robotics, several of them from the laboratories at the University of California, San Francisco and Berkeley whose work on flexible electrodes and high channel-count recording underlies the company's approach.
In January 2017 representatives acting for Musk approached Pedram Mohseni and Randolph Nudo, academics who held the rights to the name NeuraLink, and acquired it.
The company announced its direction publicly in April 2017: devices to treat serious brain disease in the short term, human enhancement in the longer term. Musk has attributed the idea to the "neural lace" of Iain M. Banks's Culture novels, a mesh that grows through the brain and connects a person to a wider machine intelligence. He initially described the lace as something that might be introduced through a vein or artery rather than by open surgery, an approach the company did not pursue and which is instead the basis of the competing firm Synchron.
Musk has described the objective in terms that oscillate between the medical and the speculative, including the suggestion that the device might eventually allow a person to save and restore a mental state "analogous to a video game, like a saved game situation", alongside the more immediately defensible claim that it could "address brain injuries or spinal injuries and make up for whatever lost capacity somebody has with a chip".
Jared Birchall, who runs Musk's family office and holds similar roles at several of his ventures, was listed as chief executive, chief financial officer and president in 2018. Musk has been the majority owner without holding an executive title.
Departures and internal conflict
The founding team did not stay. Benjamin Rapoport, a neurosurgeon, left in 2018 citing safety concerns about penetrating electrodes, and founded Precision Neuroscience, which places electrodes on the surface of the cortex rather than through it, accepting lower signal quality in exchange for avoiding tissue damage. By August 2020, according to reporting by Stat News, only three of the eight founding scientists remained, and the publication described years of internal conflict in which schedules imposed from above collided with the pace at which the underlying science could actually move. Max Hodak, the co-founder and president, announced in May 2021 that he had left; he later founded Science Corporation, which works on a retinal implant. By January 2022 two of the eight founders remained.
The company was based in San Francisco's Mission District until 2022, sharing the Pioneer Building with OpenAI, another Musk co-founded organisation, before moving to Fremont. In February 2024 Musk moved Neuralink's state of incorporation from Delaware to Nevada, days after the Delaware Court of Chancery under Chancellor Kathaleen St. J. McCormick voided his US$55 billion Tesla compensation package.
By 2019 Neuralink had raised US$158 million, of which US$100 million came from Musk, and employed about 90 people. Subsequent rounds, including a reported US$650 million in 2025 at a valuation of around US$9 billion, have brought in outside investors including Founders Fund, Thrive Capital, ARK Invest, Sequoia Capital and Vy Capital.
In April 2025 Neuralink described itself as a small disadvantaged business in a filing with the United States Small Business Administration, a designation reserved for firms owned by socially and economically disadvantaged individuals. The characterisation was false and the company did not explain it.
Technology
Neuralink remained highly secretive until July 2019, when it published a white paper and gave a presentation at the California Academy of Sciences describing three components: flexible probes, a robot to insert them, and electronics to read them.
Probes
The probes are threads of polyimide, a biocompatible polymer, carrying conductors of gold or platinum. They are between four and six micrometres wide, which is thinner than a human hair and comparable to the scale of the tissue they enter, and their flexibility is the central design claim: a rigid electrode moves relative to the brain, which pulses with each heartbeat and shifts with posture, and the resulting micromotion produces scarring that degrades the signal over months or years. A flexible thread moves with the tissue.
Each thread carries 32 electrodes. A full implant uses 64 threads and so provides 1,024 recording sites, and the published architecture allows for arrangements of 48 or 96 threads giving up to 3,072 electrodes.
The trade-off is that a thread flexible enough to move with the brain is too flexible to be pushed through the meninges, which is why the robot exists.
The surgical robot
The R1 robot inserts threads individually. Its insertion head carries a needle of tungsten-rhenium alloy, 25 micrometres in diameter, which engages a loop on each thread, drives it through the dura and into the cortex, and withdraws, leaving the thread behind. A linear motor provides the rapid retraction that allows the needle to leave without dragging the thread back out, and insertion speed is variable. A pincher, made from a 50 micrometre tungsten wire bent at the tip and driven both axially and rotationally, manipulates the threads.
The head includes an imaging stack for guidance, allowing the robot to see the surface vasculature and place each thread to avoid blood vessels, which is the main cause of damage in this kind of surgery and which a human surgeon cannot do at the required speed. The published rate is up to six threads, or 192 electrodes, a minute.
Electronics
The implant contains a custom application-specific integrated circuit providing, in the published design, 1,536 recording channels through 256 individually programmable amplifiers with analogue-to-digital conversion and serialisation on the chip. The device is sealed, powered by an internal lithium-ion cell charged inductively through the skin, and communicates over Bluetooth to a nearby computer or phone.
A limitation the company acknowledges is that its electrodes are not small enough to isolate individual neurons reliably and instead record the summed firing of small groups. Neuralink's position is that this can be addressed algorithmically, at computational cost and without exact results.
The compression problem
On 29 May 2024 Musk published a public challenge seeking a method of compressing the implant's data stream by a factor of more than 200, losslessly, in real time and at very low power, the constraint being that transmitting raw neural data at the required rate would exceed the device's thermal and battery budget.
The framing drew scepticism from information theorists and engineers. The software consultant Roy van Rijn described 200-fold lossless compression of noisy neural recordings as "just outlandish", the objection being that lossless compression is bounded by the entropy of the data and that neural recordings contain substantial genuine noise. The practical answer, which the field has used for years, is to extract features such as spike times on the device and transmit those, which is lossy with respect to the raw signal and lossless with respect to the information that matters. The episode was read by critics as an illustration of the company presenting a solved engineering trade-off as an unsolved fundamental problem.
Blindsight
Announced as a second product, Blindsight is intended to restore a degree of vision to people who have lost both eyes or their optic nerves but whose visual cortex remains intact, by stimulating the cortex directly rather than recording from it. Musk said in March 2024 that it was working in monkeys and that resolution was initially very low, comparing it to early video game graphics, and would improve.
The United States Food and Drug Administration granted the project breakthrough device designation in September 2024, which accelerates regulatory review without constituting approval or any finding about efficacy.
Cortical visual prostheses have a long history, including the work of William Dobelle from the 1970s and the more recent Orion device from Second Sight, and the difficulties are well documented: phosphenes produced by cortical stimulation do not map simply onto the visual field, stimulation thresholds drift, and the resolution achievable with any plausible electrode count is far below normal sight.
Animal research
Neuralink has tested its devices in rodents, pigs, sheep and macaque monkeys, and this programme has been the company's most sustained source of controversy.
Work on monkeys was conducted between 2017 and 2020 in partnership with the University of California, Davis, at the California National Primate Research Center. At the end of the partnership UC Davis transferred seven monkeys to Neuralink's own facility.
In 2022 the Physicians Committee for Responsible Medicine, an animal welfare organisation, obtained veterinary records through public records litigation and alleged that the experiments had caused chronic infections, partial paralysis, brain swelling, loss of coordination, bloody diarrhoea and psychological distress. It stated that at least 23 monkeys had been used and that 15 had died or been euthanised as a consequence. It also alleged that UC Davis had withheld photographic and video evidence.
Neuralink confirmed in February 2022 that macaques had been euthanised after experimentation and denied that any abuse had occurred, stating that the animals had been terminally ill or that euthanasia was planned as part of the study design. In December 2022 it was reported that the company was under investigation by the United States Department of Agriculture for animal welfare violations, and a Reuters investigation quoted employees saying that pressure for speed from Musk had caused rushed surgeries and avoidable suffering. A September 2023 investigation by Wired, based on records and on interviews with a former employee and a primate centre researcher, set out the specific complications, and in October 2023 the same publication reported that the company had worked to keep details of animal deaths from becoming public.
A USDA investigation completed in July 2023 found no breaches other than one incident the company had itself reported in 2019. The Physicians Committee disputed the finding. In November 2023 members of Congress asked the Securities and Exchange Commission to examine whether Neuralink had misled investors by omitting information about animal deaths, and in December 2024 Musk posted a letter from his lawyer stating that the Commission had reopened an investigation.
Earlier, in 2022, after the FDA rejected its first application for human trials, Neuralink conducted further work in pigs, some of which developed granulomas, masses of inflammatory tissue, in the brain. The company said it could not determine the cause and that the implant and threads were not responsible.
Musk said in March 2024 that no monkey had died or been seriously injured as a result of a Neuralink device, a statement inconsistent with the documented record and with the company's own earlier acknowledgements, on the interpretation that the animals' deaths were attributed to other causes.
In January 2025, in the first week of the second Trump administration, seventeen inspectors general were dismissed, including Phyllis Fong, whose office was responsible for the Agriculture Department investigation. The status of both federal investigations has been unclear since.
The best-known animal demonstrations were presentational rather than scientific. In August 2020 Neuralink livestreamed a pig, Gertrude, whose implant registered signals from neurons connected to her snout as she explored her pen; two other pigs appeared, one of which had had an implant removed to show that removal was possible. In April 2021 the company published video of a macaque, Pager, playing the game Pong with a wireless implant. Researchers noted that a monkey moving a cursor by neural signal had been demonstrated in 2002 and that the advance lay in the wireless operation and the electrode count rather than in the capability itself.
Human trials
Regulatory path
Neuralink applied to the Food and Drug Administration in 2022 for permission to begin human trials and was refused. Reuters reported that the agency raised major safety concerns on three points: the lithium battery, the possibility that threads might migrate within the brain, and whether the device could be removed without damaging tissue.
Approval followed in May 2023, and in September 2023 the company opened recruitment for the PRIME study, standing for Precise Robotically Implanted Brain-Computer Interface, under an investigational device exemption. The study sought people with tetraplegia from cervical spinal cord injury or amyotrophic lateral sclerosis. Because it is an early feasibility study of a first-in-human device, Neuralink is not obliged to publish results or to demonstrate safety or efficacy at this stage, and it has not registered the trial on ClinicalTrials.gov, which academic researchers have criticised as falling short of the norms of the field.
Surgeries have been performed at the Barrow Neurological Institute in Phoenix, Arizona, and subsequently at other sites.
Noland Arbaugh
Musk announced on 29 January 2024 that a device had been implanted the previous day and that the patient was recovering. On 20 March 2024 the company introduced him in a livestream on X: Noland Arbaugh, then 29, who had been tetraplegic since a diving accident dislocated his C4 and C5 vertebrae.
Arbaugh demonstrated moving a cursor by intention, controlling music and playing chess and Civilization VI. He said the device had given his life back, while acknowledging that it was not perfect. The demonstration was widely covered and was the first time most people had seen a wireless high channel-count implant in use.
In May 2024 Arbaugh disclosed that approximately 85 per cent of the implant's threads had retracted from the brain in the weeks after surgery, because his brain had moved roughly three times as much within the skull as Neuralink had anticipated. Performance fell sharply and was partly recovered through changes to the signal processing, using the remaining electrodes and a different decoding approach. Neuralink initially considered removing the device. The episode is the clearest demonstration of the specific engineering risk in the design, and the company's decision to disclose it, after reporting by The Wall Street Journal indicated it had not initially done so, was criticised on timing.
Arbaugh has remained a participant and said in a Fortune interview in August 2025 that he had regained much of the autonomy he had lost.
Subsequent participants
The FDA accepted Neuralink's proposed remedies, which involved reducing brain motion during surgery and placing the threads closer to the cortical surface so that less thread is exposed to relative movement.
A second participant, identified only as Alex, was implanted in August 2024. He has reported using CAD software, designing and printing a custom mount for the charger, and playing first-person shooter games at a level above what he had managed with other assistive technology. No thread retraction has been reported in his case.
Neuralink received approval from Health Canada in November 2024 for its first trial outside the United States, and in August and September 2025 the University Health Network in Toronto performed Canada's first two implant surgeries, on patients with cervical spinal cord injuries.
On 28 October 2025 University College London announced that Paul, the first recipient in the United Kingdom, had controlled a computer by thought within hours of surgery at a UCL hospital.
By September 2025 Musk reported twelve participants with more than 2,000 cumulative patient-days and about 15,000 hours of use, against seven reported by Barrow in June of that year.
The field
Neuralink is the best known company in its field and is not the only one, nor the first.
Academic brain-computer interface research runs back to the 1970s, and the BrainGate consortium, working with the Utah array, a rigid silicon electrode bed developed at the University of Utah in the 1990s, has had people with paralysis controlling cursors since 2004 and robotic arms since 2012. The Utah array's limitations, a few hundred channels and signal degradation over years as scar tissue forms, are precisely the problems Neuralink's flexible threads are meant to address.
Synchron takes the approach Musk initially described and abandoned: its Stentrode is delivered through the jugular vein into a vessel adjacent to the motor cortex, requiring no open skull surgery. It began human trials before Neuralink, in Australia in 2019 and the United States in 2021, and has the support of investors including Bill Gates and Jeff Bezos. Its signal quality is lower and its safety profile is considerably better understood.
Precision Neuroscience, founded by Neuralink co-founder Benjamin Rapoport, places a thin film of electrodes on the cortical surface through a slot in the skull, again trading resolution for the avoidance of penetration. Paradromics uses fine penetrating microwires and is aimed at high-bandwidth speech decoding. Blackrock Neurotech, which makes the Utah array, has the longest clinical record of any company in the field. Science Corporation, founded by Max Hodak, works principally on retinal prosthesis. Chinese efforts, including NeuCyber and the state-backed Beinao programme, have advanced quickly.
Separately, academic groups at Stanford and the University of California, San Francisco have demonstrated speech decoding from neural activity at rates approaching conversational speed in people who have lost the ability to speak, which is arguably the most consequential recent result in the field and was achieved with electrode technology Neuralink does not use.
Reception and criticism
Assessments of the engineering
Neuroscientists have generally assessed Neuralink's contribution as strong engineering rather than scientific novelty. The high channel count, the fully wireless sealed implant, the inductive charging and above all the surgical robot are real advances in integration and manufacturability, and the robot in particular addresses a genuine bottleneck, since the number of people who can implant these devices by hand is a limit on any clinical deployment.
The corresponding criticism is that the company's public presentation attributes to itself capabilities the field established earlier. A monkey controlling a cursor, a person controlling a cursor, and wireless recording had each been demonstrated before Neuralink existed, and the framing of demonstrations as firsts has irritated researchers whose work they extend.
The deeper scientific objections concern the long-term claims. Reading motor intention from the motor cortex is comparatively tractable, because the cortex is organised in a way that maps reasonably onto movement, the signals are strong, and there is a clear error signal for training a decoder. Reading or writing thought, memory or language is a different problem, and the claim that a high-bandwidth interface will allow symbiosis with artificial intelligence assumes a level of understanding of neural coding that does not exist. Critics including the neuroscientist Anne Churchland and others have argued that no quantity of electrodes substitutes for that understanding.
Ethics, transparency and hype
The ethical commentary concentrates on several distinct questions: informed consent by participants whose condition gives them a strong incentive to accept risk; what happens to a recipient if the company ceases to support the device, a problem already demonstrated when Second Sight abandoned support for its retinal implants leaving users with inert hardware in their bodies; the security and privacy of neural data, which is not covered adequately by existing frameworks; and the prospect of enhancement, which if it works would be available first to those who can pay.
Transparency has been the most concrete criticism. Neuralink has not registered its trials on public registries, has published its results principally through livestreams and posts on X rather than in peer-reviewed literature, and has disclosed adverse events late or after press reporting. Bioethicists including Arthur Caplan have argued that a company conducting first-in-human brain surgery has obligations of disclosure beyond the legal minimum.
Timelines have slipped consistently. Human trials were said in 2019 to be starting in 2020; they began in 2023. Musk has at various points suggested that devices would be widely available within a few years, and has spoken about applications including curing depression, restoring memory and enabling telepathic communication, none of which is close.
Animal welfare organisations, principally PETA and the Physicians Committee for Responsible Medicine, maintain active campaigns, and the Physicians Committee instigated both federal investigations.
Assessment
Neuralink has done something genuinely difficult and has done it in public. A sealed wireless implant with a thousand electrodes, placed by a robot in a procedure lasting a few hours, is an engineering achievement that the academic field had not produced in twenty years of work on the same problem, largely because academic funding does not support that kind of integration. Noland Arbaugh's use of a computer, and the subsequent participants' results, are real and matter to the people involved.
The company's difficulty is that the same qualities that produced the device, speed, secrecy and a founder who describes intermediate results in ultimate terms, have also produced the animal welfare record, the retraction of Arbaugh's threads, the repeated missed timelines and a reputation among researchers that is considerably worse than the technology deserves. The thread retraction in the first patient is the most instructive single fact: it was a foreseeable consequence of an aggressive design choice, it was disclosed late, and the response was competent.
Whether brain-computer interfaces become a routine treatment for paralysis within a decade is now a reasonable question rather than a speculative one, and Neuralink has contributed materially to making it so. Whether they become anything resembling what Musk has described, a bandwidth solution to the problem of artificial intelligence, depends on scientific understanding that no amount of engineering currently substitutes for.