Inside Neuralink, Elon Musk’s mysterious brain chip startup: A culture of blame, impossible deadlines, and a missing CEO
Never miss a story: Follow your favorite topics and authors to get a personalized email with the journalism that matters most to you.
At first glance, the video is impressive: A monkey plays the classic Atari game Pong, using only its mind, its mental commands wirelessly transmitted to a computer. Uploaded to YouTube on April 8, 2021, it has been viewed 5.7 million times and received some 120,000 likes.
Among those who really like the clip is Elon Musk. That’s not just because monkey MindPong is the kind of sci-fi stuff that gets Musk jazzed. It’s because that monkey—it’s named Pager, the video’s narrator tells us—is Musk’s monkey. Or rather, Pager belongs to Neuralink, a company Musk founded in 2016. Neuralink is dedicated to developing a device that, once implanted in the human brain, would allow a computer to translate a person’s thoughts into action—eventually allowing the individual to do anything we do today through typing, pressing buttons, or manipulating a mouse or joystick—by simply thinking about the desired result instead. It would also, Musk says, allow information to be beamed from a computer back into the brain. Musk has said that in the future we’ll all need such brain-computer interfaces, or BCIs—that the only way to keep up with rapidly advancing artificial intelligence will be to merge with the machines, cyborg-style.
If that sounds bonkers, well, it is. But this is Elon Musk we’re talking about. The guy who has done more than anyone else to make electric cars a reality. The guy who regularly blasts rockets into space carrying both satellites and astronauts, and then safely lands the booster stages back on Earth—vertically, and sometimes on floating autonomous barges—so they can be reused.
Musk has a reputation for accomplishing engineering feats that others believe are technically improbable and turning them into revenue-generating businesses. But Musk also has a reputation for overhyping his technologies and missing promised timelines (see Full Self-Driving) and production targets (see Tesla, Model 3).
Which brings us back to Pager. Many neuroscientists were decidedly less impressed with Pager’s mind control than a layperson might be—or, to judge from his tweets, Musk was. It turns out that science has had the ability to implant electrodes in a monkey’s brain and teach it to play a Pong-like game, sans hands, since at least 2002. In fact, BCIs have allowed monkeys to manipulate robotic arms remotely, performing far more sophisticated actions than the simple joystick movements Pager needed for Pong. You can find this video on the internet too. But it has not gotten anything close to 5.7 million views and 120,000 likes. It was not Elon Musk’s monkey.
Neuralink’s team are far from the only researchers working on BCIs. For many of these groups, the purpose of such devices is medical—to restore some agency and capability to people with spinal cord injuries and to treat conditions such as Parkinson’s or multiple sclerosis or schizophrenia. Musk also wants to help such people. But he sees that as a first step on the road to widespread adoption of a device that can bestow superpowers on the already able-bodied and -brained.
There have been signs, meanwhile, that all is not right at Neuralink. Over the past two years, most of its original founding team has departed—and former employees interviewed by Fortune describe an enterprise marked by internal tensions and erratic management. Last May, less than a month after the Pager video dropped, Max Hodak, the biomedical engineer who had served as Neuralink’s president since its founding, in charge of day-to-day management, announced via Twitter that he had left the company. He later implied that it was not a voluntary departure. Musk remains the company’s de facto CEO, but Hodak has not been replaced. The company has yet to receive Food and Drug Administration approval to implant its device in a person, a milestone that a rival startup achieved last year. (Musk and Neuralink did not respond to multiple requests for comment. Hodak declined to provide a statement.)
This year, the wider world could learn more about whether Neuralink is living up to Musk’s billing. In December, Musk told a conference audience that “we hope to have this in our first humans—which will be people that have severe spinal cord injuries like tetraplegics, quadriplegics—next year, pending FDA approval.” In January, the company posted a job listing for a clinical trial director, an indication that it may be on track to meet Musk’s suggested timeline.
To many who work in neuroscience and rehabilitative medicine, much of what Musk says about Neuralink rings alarm bells that he is exaggerating what the technology can do. In 2019, Musk said that he saw Neuralink’s goal as not just treating but curing brain diseases and disorders. James Wu, a neuroscientist and technologist, recently tweeted that he was dismayed at the attention lavished on the company (he used an asterisk, presumably, to avoid tangling with Musk fans searching for the name). “Neur*link is pouring funding only into implants that realistically will never be used by the average patient with disabilities in the absence of structural changes to our health care systems,” he wrote.
Can Neuralink deliver on Musk’s grandiose promises? Although the least scrutinized of Musk’s companies, it could be among the most impactful—helping to unlock the mysteries of the brain and perhaps even changing the definition of what it means to be human. Neuralink may also be the most challenging endeavor Musk has yet taken on. “Unlike Tesla or SpaceX, we are not talking about technological problems or infrastructure problems,” says Christof Koch, a researcher at the Allen Institute for Brain Science in Seattle. “These are fundamental science problems.” In some ways, Musk’s company has already made a mark: The billionaire’s enthusiasm for BCI technology has turned a formerly niche academic field into a fast-growing commercial industry, similar to the way Tesla helped turbocharge the global electric-vehicle sector. The problem with this “Elon Musk effect,” however, is this: If the technology fails to live up to Musk’s hype, will the inevitable disillusionment set back progress? And will hyperbole give false hope to those who most need a brain-tech breakthrough?
The Utah Array is a tiny silicon square with 100 needles sticking out of it. Those needles are designed to be pushed directly into the brain—and they’ve made the array the gold standard for relaying electrical signals from a brain to a computer. Developed at the University of Utah in 1992, it was, until last year, the only FDA-approved BCI. Each of its needles can, in theory, record the electrical activity of a single neuron. The data is streamed along a wire to a boxy processing unit that sits atop someone’s skull, with a fat cable piping power in and data out. The system has been used to help paraplegics “type” by moving a cursor to select letters on a computer screen and to pick up objects with robotic limbs.
But the array—based on 1990s technology and with coverage for only 100 of the brain’s roughly 86 billion neurons—was never going to allow humanity to merge with artificial intelligence. In 2016, Musk began quietly assembling a dream team to build a brain-computer interface that would significantly advance the technology. He gathered eight scientist “cofounders” with expertise in BCIs and specialties including implantable microsensors, brain-inspired computer chips, robotics, and neurosurgery.
Musk reserved for himself most of the key responsibilities of a chief executive. (In the company’s registration forms and in other paperwork, Jared Birchall, a wealth manager who runs Musk’s family office, has been listed as the company’s chief executive; former employees say Birchall has no direct role at Neuralink.) But leading operations was Hodak, a prodigy who had worked at the leading BCI lab run by Miguel Nicolelis at Duke University (the first person to perfect monkey MindPong). Though only 28, Hodak had already founded and helped lead multiple startups, including a company that allowed researchers to run experiments at remote, robotically operated labs.
For four years, the Neuralink team labored mostly in stealth mode, except for a rare public recruiting event in 2019. The world got its first real look at what Neuralink had built in August 2020—in a livestreamed event in which Musk shared the spotlight with some pigs that had recently received Neuralink BCIs. Musk showed the audience how data depicting the firing of the pigs’ neurons could be streamed wirelessly to a computer as the pigs nosed around their pens. He also showed a video of a pig walking on a treadmill, to demonstrate how that neural-firing data could be used to predict the position of the pig’s joints. But the real star of the livestream was Neuralink’s hardware. Musk held in his hand the implant itself—which the company calls the Link.
The Link is a disk-shaped device, about the diameter of a U.S. quarter coin—elegantly compact compared with the Utah Array. Numerous tiny metal threads dangle from it, each much thinner than a human hair. The coin sits flush inside a hole drilled in the skull; the threads are what get embedded in the brain. Because they are flexible, they can move as the brain moves, hopefully averting damage to surrounding tissue. They’re also coated in a conductive polymer engineered to withstand the brain’s corrosive environment and to minimize inflammation and scarring, which can wreak havoc with a BCI’s ability to pick up signals and can even cause cognitive impairment.
Branching off each thread are the electrodes themselves. According to an interview that a senior Neuralink manager gave to a scientific publication last summer, each Link has 64 threads, with 16 electrodes on each thread. That means that each Neuralink BCI can monitor 1,024 channels of information—more than 10 times the capacity of the Utah Array. The device processes and filters those signals on the embedded computer chips themselves, a big innovation over prior BCIs that did their processing “off-device.” And Links can transmit information wirelessly, at what Neuralink has said is an unspecified “megabit rate” (conventional Bluetooth transmits at three megabits per second). Completely sealed under the scalp—“I could have a Neuralink right now and you wouldn’t know. Maybe I do,” Musk said mischievously during the presentation—the device’s battery can be recharged wirelessly, a key selling point for a device aimed at mass consumer adoption.
To implant the BCI, Neuralink developed a surgical robot that acts a bit like a sewing machine, using a very sharp, very hard needle made of a tungsten-rhenium alloy to position the tiny threads. The robot drills a hole in the skull, “sews” in the threadlike electrodes—imaging the brain through powerful lenses to avoid piercing veins and arteries—and closes the wound with surgical glue. “We eventually want this robot to be able to do the entire surgery,” Musk said. He added that Neuralink was confident it could implant a Link in under an hour, without general anesthesia.
The Link wowed scientists and laypeople alike with its integration of complex technologies. But the process leading up to its 2020 debut had been anything but harmonious, according to a half-dozen former Neuralink employees. Most of these former employees requested anonymity, concerned about violating nondisclosure agreements and the possibility of drawing Musk’s ire.
From the beginning, there was tension between the competing goals of Neuralink’s founding scientists. “Each of the cofounders wanted something different,” says Tim Hanson, a senior scientist at Janelia Research Campus, part of the Howard Hughes Medical Institute in Ashburn, Va., who was part of the founding team and, at the time, specialized in surgical robotics for BCIs used in animal studies. Hanson and another cofounder leaned more toward basic science; others were interested in using BCIs to treat disease. Hodak, too, saw treatment as important but wanted a device with broad appeal. Musk saw medical uses as an aspect of Neuralink but not its ultimate goal, which lies more in creating a device that might empower the able-bodied.
The pressure to deliver results was described by former employees as intense. “There was this top-down dissatisfaction with the pace of progress even though we were moving at unprecedented speeds,” says one former member of Neuralink’s technical staff, who worked at the company in 2019. “Still Elon was not satisfied.” Multiple staffers say company policy, dictated by Musk, forbade employees from faulting outside suppliers or vendors for a delay; the person who managed that relationship had to take responsibility for missed deadlines, even those outside their control. This culture of blame and fear, former employees said, contributed to a high rate of turnover.
The pressure could be problematic because of the multiple challenges Neuralink was tackling. “There’s this mismatch,” Hanson says, between the speed at which engineering obstacles can be solved and the more deliberative pace of fundamental science. “Basic science is basically slow,” says Hanson, who left the company in 2018. Engineers sometimes had to make decisions about issues such as electrode design before relevant data was available from scientific teams working on animal research. Animal research can take months and years; the engineers were under pressure to act in days and weeks. There were also delays caused by Neuralink’s efforts to fabricate custom-designed computer chips, one former employee said. Musk wanted to move into human implantation as fast as possible.
Despite being de facto CEO, Musk, primarily occupied with Tesla and SpaceX, spent little time at Neuralink’s offices—initially dropping by once a week, and then later visiting the offices only about twice a quarter, often for no more than a few hours, according to former employees. On some occasions, he could be spotted roaming the halls, usually accompanied by Hodak; Shivon Zilis, a former venture capitalist who is now director of operations and special projects at Neuralink; and a bodyguard. Often, following these visits, former employees say, Hodak instructed them to shift priorities, sometimes jarringly. When Musk wasn’t there, senior managers struggled to get the billionaire’s attention even though his sign-off was required on major decisions, one former employee recalls. The former employee believes that this remains a problem. Adding to the dysfunction, Musk encouraged junior employees to email issues and complaints to him directly, bypassing normal management channels, the employee says.
In some cases, Musk intervened to nix ideas teams had spent weeks on—such as when he told one group not to use silver in the Link’s electrodes because of its potential toxicity, even though the team had planned to sheathe the silver in a nontoxic polymer. There was also consternation among scientists over Musk’s views on publishing research: He thought his engineers should be producing products, not papers. But when other leaders convinced Musk the company should publish a white paper describing some of its innovations in 2019, Musk wound up as its sole named author, rankling researchers.
At his companies, Musk has a policy against hiring outside contractors and consultants. At Neuralink, this led to decisions that struck some former employees as bizarre. For example, Hodak initially put Neuralink’s software leads in charge of managing construction of the company’s new offices in Fremont, Calif. That massive project required liaising with specialized contractors and dealing with regulators on complex environmental permitting. Hodak said, according to former employees, that this decision was intended “to shake things up.” After many months of delays, the company put two managers with mechanical engineering backgrounds in charge of supervising the project.
Former employees said Hodak and Musk had an increasingly strained relationship, especially as the company fell further behind on Musk’s ambitious timelines. Finally, on May 1 last year, Hodak tweeted, “Some personal news: I am no longer at Neuralink (as of a few weeks ago). I learned a ton there and remain a huge cheerleader for the company!” When someone tweeted in reply that “I am not a fan of leadership leaving a company w/o having a single product on the market. Seems too early,” Hodak tweeted back simply, “Same.” The clear implication was that Hodak’s departure was not his choosing.
Neuralink has yet to announce a new president. And of Musk’s eight cofounders, only two, Dongjin Seo and Paul Merolla, remain at the company.
As former Neuralink microfabrication specialist Felix Deku recalls, Musk would sometimes tell staff, “Guys, we need to do this, people need this device.” This outreach, another former employee said, was often inspired by emails from despairing patients who believed Neuralink’s technology could be life-changing—aggravating some employees’ concerns that Musk was raising expectations far too high.
Numerous brain-computing experts say that as Neuralink has catapulted their discipline into the public eye, they’ve received more messages like these. Many are also seeing an increase in something else: interest from investors. Tech giants like Facebook parent Meta and heavy-hitting VCs like Peter Thiel have jumped on the brain-computing bandwagon. Investment firm Zoic Capital noted in a recent report that “Musk has created a fear of missing out” among investors, calling it—of course—“the Elon Musk effect.” In fact, Musk’s interest has helped spawn a generation of BCI startups, developing products for both medical and commercial uses. And many credit Musk with taking a field that was years away from having commercial impact and making it instantly investable. Investors plunked $531 million into brain-computing startups in 2021, nearly four times the amount in 2020, according to PitchBook data.
“From a fundraising standpoint, it’s been 100% a positive force,”said Matt Angle, the CEO of BCI startup Paradromics, about investor enthusiasm brought on by Musk. Paradromics has raised $49.4 million since 2017, including a $20 million seed round last July. “I could not have funded the company the way that I’m funding it right now if Elon hadn’t normalized this field.”
“Nobody presented it better than [Musk] has,” says Marcus Gerhardt, CEO of BCI company Blackrock Neurotech. “He got it out to millions of people. You can sit back and be annoyed by that, but the great thing is…the awareness has risen. So, you know, kudos where kudos are due.”
Several departed members of Neuralink’s founding team are now working on other BCIs. Hodak has cofounded a company called Science Corporation that is exploring BCIs that, while still implanted, are less invasive than Neuralink’s. Benjamin Rapoport, who initially led Neuralink’s surgical team, is now cofounder of Precision Neuroscience, a startup exploring what it calls “minimally invasive BCIs.” Philip Sabes, a pioneering BCI researcher, now leads a BCI company called Starfish. Meta hired Vanessa Tolosa, who developed the materials for Neuralink’s implant, to work on hardware for its Reality Labs unit, which focuses on devices for navigating the metaverse. “When somebody leaves [Neuralink], it’s like, ‘Here’s money,’ ” says Deku, who was not part of the founding team but who left Neuralink last March to join brain-computing startup Braingrade.
A number of investors told Fortune that they are playing the long game, believing that it will take years, possibly decades, for the technology to go mainstream. But many are going long on Musk. In July, Neuralink announced it had raised $205 million from prominent VC funds, including GV (formerly Google Ventures), Vy Capital, DFJ Growth, Craft Ventures, Founders Fund, and Gigafund. The company has also been backed by several individual friends of Musk’s, including fellow members of the so-called PayPal mafia and Sam Altman, who cofounded the A.I. company OpenAI with Musk. In total, Neuralink has raised $363 million in venture funding.
“We’re at this moment where you can make some leaps” in BCI technology, says Krishna Yeshwant, a partner at GV. He declined to comment specifically about Neuralink but said of the broader field, “I feel pretty confident that we are going to be able to do some good things for patients.”
Even neuroscientists who have been underwhelmed by what Neuralink has shown it can do so far are excited about the hardware, which they say represents a genuine leap forward. “It is damn impressive,” Koch of the Allen Institute says of both the Link device and Neuralink’s surgical robot. On YouTube, you can find a breakdown of the monkey MindPong video created by Paul Nuyujukian, a neurosurgeon and bioengineer who heads the brain-computer interface lab at Stanford University. Nuyujukian dwells on all the innovative features of Neuralink’s system: the fact that the two Link devices implanted in Pager’s brain can gather so much data—2,048 channels, more than 20 times what could be done before; that Neuralink’s sophisticated algorithm can handle not just directional movement but also a mouse click or “select” function; that it runs data processing right on the Link’s chip; and that it can stream data over a conventional Bluetooth connection to a smartphone app. Reinhold Scherer, codirector of a group that works on BCIs at the University of Essex, in England, says Neuralink’s hardware innovations and its ability to gather data from so many neurons over a long period of time are an example for the entire field. “Neuralink is doing what we all need to do,” he says.
When somebody leaves Neuralink, it’s like, ‘Here’s money.’Felix Deku, former Neuralink employee
At the same time, many worry that Musk is raising hopes too high. “Elon is a very colorful, eclectic individual and he has brought a lot of attention to the space,” says Tan Le, the founder and CEO of Emotiv, a company that makes a variety of EEG headsets and software. “But this is a technology that can improve and transform people’s lives and we need to make sure we don’t scare people away prematurely.” Le is concerned that some of what Musk talks about—merging with A.I., for example—certainly sounds scary.
Fears over inflated expectations extend to Musk’s comments about nearer-term goals too. Musk’s vision of a full BCI mind meld—seamless, high-bandwidth two-way exchanges between brain and computer—faces even more daunting obstacles. Some are bureaucratic. Koch says that the FDA is unlikely to ever approve the implantation of BCIs in healthy individuals. Proving that any enhancement in cognitive abilities would exceed the risks of implantation would be very hard for any company, he notes.
Tim Harris, a senior fellow at Janelia Research Campus and the lead designer of a BCI for animal research called Neuropixels, notes that the interface Musk seeks would require what neuroscientists call “full brain coverage”—which in turn could require implantation of hundreds of Links in each user. “It is a major worry from my perspective,” Harris says. “How many holes can you insert and still feel?”
Solve for all that, and Neuralink would face an even bigger conundrum—how to send data back into a person’s brain. Neurosurgeons have found that stimulating small areas of the brain can produce complex thoughts and images—emotions such as intense guilt and vivid visual hallucinations of shapes and colors. But move the electrical probe just a millimeter and suddenly the patient feels nothing, or something completely different, Koch says. No one understands why. And no one knows how to reliably deliver a particular piece of information directly into the brain through electrical stimulation. “I’m 65,” Koch says. “In my lifetime we won’t get anything close to science fiction.”
While some worry about the implications of failure, others wonder, What happens if Musk succeeds? If you’re uncomfortable with what Google or Facebook learn from your online behavior, imagine what Neuralink might be able to know about you someday. “The brain is the most sacred place—the most important place for privacy and the creation of self,” says Nita Farahany, a law professor at Duke University who specializes in the legal and ethical quandaries posed by advances in neuroscience. “Once that becomes accessible to companies and governments and everyone else, that is a concerning if not terrifying thought.”
Musk has emphasized in some public statements that his vision won’t come to fruition rapidly. “This will be a slow process,” he said in a 2019 presentation. But he also underscored the urgency of his mission, saying that “even in a benign A.I. scenario” humanity would increasingly “be left behind.” And then, as only Musk can, he made preparing for a doomsday scenario by having holes drilled in your skull sound, well, almost fun. “With a high-bandwidth brain-machine interface, I think we can actually go along for the ride.”
The “Elon Musk effect” is real: The attention that Neuralink has attracted has helped foster a flock of startups specializing in wireless brain-computing interfaces. here are four of the leaders.—Jonathan Vanian
2022 will be a big year for this Peter Thiel–backed company, as it works with the Federal Drug Administration to approve its newest BCI device, which itself builds upon an older FDA-approved device that has been used in several clinical trials. Neuralink’s device can record more brain activity than Blackrock’s, but Blackrock has operated for over a decade, and it has implanted its technology into 31 people worldwide. The tech is based on the 1990s-era Utah Array devices and has already helped paralyzed patients regain some body movements—grasping a cup of coffee, for instance. The next goal: helping immobile patients accomplish feats like moving a wheelchair, just by thinking.
Synchron’s BCI device doesn’t get implanted into the brain. Instead, physicians insert the device—a Stentrode, or an electrode array mounted on a stent—through the jugular vein in the neck so that it can be snaked along into blood vessels in the brain, where it can record electrical signals. The device can help patients who have lost the ability to use their arms move digital cursors on their laptops. Synchron, which has previously implanted its Stentrode in patients in Australia, has received FDA approval to move ahead with a clinical trial and is looking for its first U.S. patient.
Precision Neuroscience, cofounded by Neuralink founding team member Benjamin Rapoport, is developing a BCI device that it describes as a thin film that contains thousands of electrodes; the film can be inserted into a small slit in the skull so it can rest on top of the brain. (The startup’s leaders describe the process as analogous to placing an old-school floppy disc into a disc drive.) Precision hopes to use it to treat epilepsy and other disorders involving abnormal electrical signals in the brain. Its founders plan to submit for FDA approval in early 2023.
The Harvard University–backed startup has developed a noninvasive BCI device that looks like a head visor from Star Trek. The headband can record electrical activity through electrodes that press against the temples from outside the brain. Although the data isn’t as detailed or high-quality as what can be captured by a device within the brain itself, the startup’s executives say the information can still be helpful as a device to provide neurofeedback, enabling people to train themselves to be more focused or calm. BrainCo is also seeking FDA approval for its BrainRobotics prosthetic hand, which analyzes electrical signals sent to the brain via the arm’s muscles to provide greater motor control to amputees.
The problem’s in your head
Brain-implant technology has been around for decades. but the biology of the brain itself has been a major obstacle. Here’s how Neuralink and others are overcoming that hurdle.—Jeremy Kahn
Brains and implants = oil and water
The brain is an inhospitable environment for the microelectronics that make BCIs work. It’s a salty sea of chemicals, most of which are corrosive to metal. If that metal breaks down, it can release toxins. The brain also doesn’t like having things poked into it, so it forms scar tissue around foreign objects. That tissue acts as an insulator, diminishing an electrode’s ability to pick up signals. It can also displace electrodes, separating them from the neurons they’re monitoring.
The brain moves around a lot, shifting inside the skull and pulsating as blood circulates. If an implant’s electrodes are stiff, as they are in most BCIs, this movement can damage neurons and blood vessels, as brain tissue bumps against them like jellyfish against a pier. The larger the electrode, the more chance that it will cause bleeding, especially during implantation—a risk that has plagued the relatively large electrodes used for deep brain stimulation in Parkinson’s patients. There is also the risk of chronic inflammation. In some laboratory rats, this has led to serious cognitive decline.
Betting on flexibility
The solution that Neuralink and other BCI startups are exploring involves using much smaller, more flexible probes to do the brain-poking. In the Link, each of the 64 threads on which electrodes are mounted is much thinner than a human hair. Because they are flexible, they can move as the brain moves, rather than colliding continually with surrounding tissue. And while the threads are metal, they’re coated in a polymer engineered to be less vulnerable to corrosion, reducing the risk of harm.
This article has been updated to include more information about Neuralink’s corporate structure.
A version of this article appears in the February/March issue of Fortune with the headline, “Neuralink wants to get inside your head.”