Elon Musk is preparing for the most consequential launch of his career. But this one isn’t rocket science—it’s brain surgery. Musk’s company Neuralink Corp. is seeking a volunteer for its first clinical trial, meaning it’s looking for someone willing to have a chunk of their skull removed by a surgeon so a large robot can insert a series of electrodes and superthin wires into their brain. When the robot finishes, the missing piece of skull will have been replaced with a computer the size of a quarter that’s meant to stay there for years. Its job will be to read and analyze the person’s brain activity, then relay that information wirelessly to a nearby laptop or tablet.
For the purposes of the trial, an ideal candidate would be an adult under age 40 whose four limbs are paralyzed. Such a patient would likely have Neuralink’s implant inserted into what’s known as the hand knob area of their premotor cortex, which governs the hands, wrists and forearms. The goal is to show that the device can safely collect useful data from that part of the patient’s brain, a key step in Neuralink’s efforts to convert a person’s thoughts into a range of commands a computer can understand.
Several companies and research teams have already created implants that can help patients perform basic tasks with their thoughts, such as clicking objects on a screen with a cursor. Neuralink, in familiar Musk fashion, has issued far wilder promises. For the past four years, starting with the company’s first public demonstration, he has made it sound as if there would soon be ubiquitous clinics where anyone could go in for a 15-minute robosurgery and come out a human-machine hybrid. These cyborgs would be able to download knowledge the way Keanu Reeves does in The Matrix or upload their thoughts into storage, even to other brains. “This is going to sound pretty weird, but ultimately we will achieve symbiosis with artificial intelligence,” Musk said at that first demo in 2019, when the company said human trials could begin in 2020.
Unrealistic timetables are one of Musk’s favorite management techniques. To his credit, he’s made several improbable dreams come true—eventually. But while rockets and cars are serious business, neural implants require perfection on a whole other level. One does not rush a brain implant to market and hope for the best.
Two other companies, Synchron and Onward, have more than a year’s head start on human trials with brain implants and related technology. Neuralink has, however, gotten dramatically more attention than the decades of incremental, largely academic research that preceded it, and not always to its credit. Some neuroscientists have said Neuralink is hyping the technology. Animal-rights groups have accused it of cruelty to the monkeys, pigs and other mammals it’s tested implants on so far. The through-line is Musk, whose increasingly manic and reactionary online persona has done little to suggest he stands as the ideal candidate to mass-produce mind-control devices.
All these concerns are valid. Yet Neuralink’s trial is exciting, too. The company appears to have turbocharged progress in this slow-and-steady field, and it’s now built the world’s most powerful and elegant brain implant. If the product works as intended, later iterations could improve, in miraculous ways, the life of millions of people suffering from paralysis, stroke, Lou Gehrig’s disease, and hearing and vision loss. In the meantime, its high profile already has investors hunting for the next Neuralink. Once again, Musk has reshaped an entire industry, and this one could be the most transformative of all.
In the past three years, I’ve made 10 trips to Neuralink’s facilities in Silicon Valley and its growing operations in Austin. In tandem with Musk’s impatient demands, I’ve seen his team profoundly advance their technology and ambitions. As they prepare for the trial, the pressure to succeed is something even Musk hasn’t seen before. Tesla Inc., after all, took many years to mass-produce its cars, and SpaceX’s first three rockets exploded. When it comes to brains, “We can’t blow up the first three,” says Shivon Zilis, Neuralink’s director of special projects. “That’s not an option here.”
The modern history of brain implants began with the technological advances of the 1990s. The science, simplifying tremendously: Thoughts cause neurons to fire in particular patterns, and these patterns have some degree of consistency across brains. In fact, roughly the same neurons fire when a person thinks about moving their arms and fingers, whether they can physically move them or not. Brains light up in similar, consistent ways when people want to move a mouse cursor to click somewhere on a computer screen, too. The same is true for speech: If you can think of saying a letter or word, you’re making the same neurons fire as you would by physically speaking it. Even if you can’t speak then, a well-trained computer should be able to discern your intent and, theoretically, speak for you.
The challenge is figuring out each entry in the neurons-to-English dictionary, which requires gathering and studying tons of data about the patterns of many people’s brain signals. To get the clearest signals, you want to place sensors as close to the neurons as possible. Some researchers have tried to avoid surgery by keeping their devices outside a person’s skull, but the added distance and interference have yielded muted results. The most precise data usually comes from electrodes sitting right beside brain cells.
“The long-term goal is to have this available for billions of people and unlock human potential and go beyond our biological capabilities”
For most of the past 20 years, the Utah array has been the implant to beat. It’s a tiny, flat square of silicon that could fit atop a child’s fingernail. Wires protrude from its edge, and on one face of the chip is a bed of about 100 rigid spikes. To implant the Utah array, a surgeon must perform a craniotomy, cutting a large hole in the patient’s skull, then gently hammer the tiny spikes into the brain. The wires are positioned to connect to a metal port that visibly pokes out of the scalp after the opening is sutured shut. Post-op, to use the device, an ice-cube-size computer is attached to the patient’s head.
Researchers have made major advances with Utah arrays. They use them to read and translate the brain activity of people with paralysis and other conditions. Software created with this information allows patients to communicate with caregivers and loved ones, or to manipulate robotic arms to pick up objects. The catch is the hardware’s clunky design, which has remained largely unchanged over 20 years. The arrays also need a raft of computers and other equipment operated by trained personnel and lots of medical care, which has kept them confined mostly to research labs.
Musk co-founded Neuralink in 2016 with seven scientists and $100 million of his own money. The splashiness of his investment, and his grand promises about the underpinnings of the technology, proved irresistible to venture capitalists. Neuralink has since raised more than $500 million, including $280 million this year, and the attention has helped draw investors to other brain-computer interface efforts, including long-standing university projects as well as newer startups. Last year, 37 such companies raised more than $560 million, according to the research company PitchBook.
Most of these enterprises have the same primary objective: Build a brain-scanning device that can leave the lab behind. The ideal implant will have plenty of computing oomph to record and input lots of data and also to transmit the data via strong wireless signals. This must all be done while using as little battery power as possible and without running too hot, which could irritate or injure a patient. Beyond the hardware, the brain-computer interface companies also need serious machine-learning software skills and to perform thousands upon thousands of tests.
Neuralink’s implant sits invisibly beneath the scalp, flush with the skull. It’s also packed with enough computing horsepower to handle jobs well beyond think-and-click. In the nearish future, the idea is to enable high-speed typing as well as seamless use of a cursor. Neuralink has also been working on a complementary spinal implant intended to restore movement and sensation in paralyzed people. “The short-term goal of the company is to build a generalized brain interface and restore autonomy to those with debilitating neurological conditions and unmet medical needs,” says DJ Seo, a Neuralink co-founder and vice president for engineering. “Then, really, the long-term goal is to have this available for billions of people and unlock human potential and go beyond our biological capabilities.”
Although some competitors have beaten Neuralink to human trials, the company’s raw technology is closest to being a general-purpose computer in the brain. The implant has more than 1,000 electrodes for gathering brain data, compared with 16 or so in rival devices. The Neuralink hardware is a nesting doll of processing, communications and charging systems, including a battery and signal amplification. Competitors, meanwhile, must still connect their implants via wires to bulky pacemaker-size battery and amplifier units that are often surgically implanted in a patient’s chest. Neuralink’s battery lasts a few hours and can be recharged wirelessly in a couple of hours via a custom baseball cap.
Another favorite Musk move is bringing key manufacturing in-house, which adds financial risk, obviously, but saves time. Neuralink even makes its own semiconductors, an exceedingly rare step in the medical-device business. It tailors them specifically for its low-power, low-heat needs. In Austin it’s turned a former ax-throwing bar into a 12,000-square-foot implant manufacturing line and testing center. Along with the usual mills, lathes and laser cutters, the shop includes more outlandish equipment, such as a refrigerator-size cabinet—filled with a type of synthetic brain fluid—that heats, cools and jostles implants to simulate years of wear and tear.
The priority during the surgery is to avoid creating any bleeding or scar tissue in a patient’s brain. To that end, Neuralink also built its own surgical robot. It’s a hulking, 7-foot-tall white machine with a steady, cubed base supporting a tower of electronics and equipment.
Once a human surgeon cuts a hole in a patient’s skull, the robot performs the ultra-delicate task of placing the electrode-laced wires, which Neuralink calls threads, into the brain. The robot has cameras, sensors and a tiny laser-milled needle that it hooks into a loop at the end of each thread. One by one, the needle pushes the 64 threads, each lined with 16 electrodes, into the brain, all while carefully dodging blood vessels. No human would be allowed to try this given that each thread is 5 microns thick, or about 1/14 the diameter of a strand of human hair. To further avoid damaging tissue, the threads have been engineered to be a mix of thin, pliable and sturdy and have been coated with a special polymer to keep them from deteriorating over years.
Neuralink’s dozen or so robots performed 155 of these surgeries on sheep, pigs and monkeys in 2021 and 294 last year. With human subjects, the surgical prep and craniectomy are expected to take a couple of hours, followed by about 25 minutes for the actual implantation. “The last two years have been all about focus on building a human-ready product,” Seo says. “It’s time to help an actual human being.”
During one of my visits, Musk pushed Seo and the rest of the engineers to do more, and faster. He wanted the robot to perform the surgery in less time and ideally without the help of a human surgeon. He wanted semiconductor experts to forget what they learned in school and try out simpler manufacturing techniques. He wanted the implant to look sleeker and last longer and, well, maybe everyone needed to rethink just about everything. I saw scientists wince as they considered the distance between their boss’s demands and the physical capabilities of their hardware. But I also saw Musk perform a type of pattern-matching for which he was especially suited, thinking ahead to how a range of design tweaks would affect mass production down the line. His been-there, done-that attitude seemed to give the staff confidence that he was right—that he had some master plan he knew they could pull off.
Musk’s management style has its upside. It’s yielded the world’s most successful rocket and its most valuable car company. Of course, as anyone who’s Googled “Cybertruck window” can tell, his first drafts can be a mess. And, as he’s torn X, né Twitter, down around himself, he’s shown that the engineering shoe-banging doesn’t always lead to rational product choices.
“We need to get there before the AI takes over. We want to get there with a maniacal sense of urgency. Maniacal”
Seo and Musk are the only Neuralink founders still with the company. (The other six left of their own volition or because of disagreements about the company’s direction; a number of them have since started similar companies.) Seo runs development of both the implant and the surgical robot. Jeremy Barenholtz, a computer scientist only two years out of Stanford University, has arisen as one of the lead executives and has managed the arduous US Food and Drug Administration approval process. Musk tends to pop in and out, but he looms as the ultimate authority and chief executive officer. A July 2022 visit to the company’s headquarters in Fremont, California, showed how that works in practice.
Musk summoned top executives and engineers to catch him up on the latest progress toward human trials. The team gathered at a rectangular, Ikea-style table in the office’s main work area. It’s a large, open-plan space full of people at computers, robot prototypes and testing equipment. Musk, in a black suit and holding a Red Bull, stood at the head of the table and started asking about competitors. As employees gave updates, he peppered them with technical questions. He was particularly focused on Synchron, which had already received regulatory approval to begin human trials.
Synchron’s pitch is technology that doesn’t require cutting into the skull. It makes a small, stentlike product called the stentrode that can be slid into a brain’s blood vessel through a patient’s vascular system. The hardware can’t sit right against neurons for the best signals, but patients with paralysis were already using it to navigate web pages and send WhatsApp messages. “We should be far exceeding the stentrode,” Musk said. “And they are currently kicking our ass. I want to be in dozens of people next year.”
His lieutenants grimaced in sync. Barenholtz had the unenviable task of explaining to Musk that the FDA wanted to wait at least a year after the company’s first surgery took place to try and implant more individuals.
“Unacceptable,” Musk said. “Generally, the way regulators work is that they’ll say one thing and that they really just kind of want to see what happens. If the patient is an advocate for the device and says it’s incredible and that we see no complications, they will approve the next one very quickly.” It was like SpaceX getting federal approval for more rocket tests, he said: “If the thing is working well, and you get a letter-writing campaign to the FDA, I guarantee they’ll move.”
At another “update Elon” meeting a couple months later, Musk trained his sights more on Onward, which makes implants that nestle against the spine. Its device sends electrical pulses that help reanimate muscles and have enabled people with paralysis to walk again. Their gait tends to be somewhat herky-jerky, but for people who’ve been paralyzed—and their loved ones—the act of standing and walking again can be similar to a religious experience.
Musk was also there to prepare for an important demo. Neuralink was planning to announce that it had started work on its own spinal implant, to be paired with its brain implant. This time, Musk seemed more agitated, and he pushed harder.
“In general, the company needs to make faster progress,” he said to Joseph O’Doherty, one of the assembled engineers. “The pace of progress is too slow.”
“OK,” O’Doherty said.
“Step it up.”
“OK.”
“Yeah.”
“OK.”
“I mean it.”
“Understood.”
“Step it up. Let’s go.”
None of this was playful. O’Doherty took it in stride and rolled right into a lengthy presentation on Neuralink’s progress with its early spinal technology, showing videos of an implant being used to stimulate the legs of pigs to make them walk on a treadmill. The conversation covered different parts of the brain, parts of the spine, spike grids, joint angles, machine-learning models and more. Musk, who has no formal medical training, kept up with it all. He made suggestions on how the implant could be tweaked to perhaps produce a less jerky motion in the animal’s gait.
His ideas were sometimes far afield of practicality, but then also often spot on. Following one visit, the semiconductor engineers retooled their process for bonding the threads to the company’s chips on Musk’s advice, and their manufacturing speeds went up 50% while defects went down, according to Zack Tedoff, the chip division’s head of brain interfaces. The team working on the spinal implant went back to the drawing board to try to get their pigs to walk in a more true-to-life manner, and Barenholtz more or less started living at the office to address Musk’s every demand.
Musk turned out to be right about the FDA. Neuralink has received an outpouring of interest from thousands of prospective patients, and the agency recently gave it the green light to perform additional implant trials in 2024 without a yearlong evaluation period. The company estimates that each implant surgery will run it about $10,500, including exams, parts and labor, and that it will charge insurers about $40,000. It forecasts annual revenue as high as $100 million within five years. Neuralink says it plans to perform 11 surgeries in 2024, 27 in 2025 and 79 in 2026. Then things really ramp up, going from 499 surgeries in 2027 to 22,204 by 2030, according to documents provided to investors.
Before that September 2022 meeting ended, Musk stressed speed on an entirely different level, “like the world is coming to an end.” The reason the staff needed to work dramatically faster, he said, was to make sure hybrid human-implant brains stayed competitive with a theoretical artificial superintelligence that might otherwise wipe out humanity. “We need to get there before the AI takes over,” he said. “We want to get there with a maniacal sense of urgency. Maniacal.”
There’s nothing pleasant about testing medical devices on animals. The practice is, on some level, a form of animal sacrifice on the altar of science, increasing their pain in the hope of decreasing it in humans. Yet Neuralink has come in for special scrutiny regarding its treatment of animal test subjects, especially this year. Reports in Wired, Reuters and other outlets have detailed surgical complications, behavioral side effects and prolonged suffering, particularly with Neuralink’s primate subjects. Some implanted monkeys, the reports said, scratched and yanked at their heads until they drew blood, or acted despondent or agonized until they were euthanized. Wired’s recent stories, which describe the monkeys’ deaths as “gruesome” and “grisly,” were based on public records it obtained and interviews with researchers.
Neuralink acknowledges that it’s made mistakes during exploratory surgeries, though it attributed them to human error rather than issues with its equipment. It stresses that the most troubling reports are drawn from its early years, before it built its own testing facility in Fremont, and that it has gone to great lengths to provide better living conditions there. “I will always find a way to protect the animals in front of me,” says Autumn Sorrells, who manages Neuralink’s nonhuman test subjects and previously oversaw lab-animal welfare at the University of California at San Francisco. “We get called ‘killers’ and ‘animal abusers’ and then have to come in to work and snuggle a sheep and make sure they have a great day. That’s f---ing hard.” She says Neuralink’s animals have larger cages, more food and entertainment options, and much more socialization than she’s seen in other lab settings.
This squares with my reporting. I’ve seen the same group of rhesus macaque monkeys living at Fremont for three years now. They’ve all had implants in their brain at various points. The devices can be removed, and a couple have been upgraded to newer models. Seventeen of the monkeys are still active, healthy and feeding Neuralink brain data on-site; three retired to a sanctuary; and one was euthanized during a planned terminal procedure.
Before you can enter the animal facility, you have to don gloves, a gown, booties, a mask and protective eyewear. You also get a briefing on how to approach the animals: slowly, carefully, sans eye contact. If you look a monkey in the eye, a signal of aggression, the monkey might freak out. Inside, spacious playpens are filled with toys, faux trees and playground equipment. Music is often playing throughout the facility, and TVs are on hand, showing mostly nature programs.
The monkeys’ primary role is to demonstrate that both the device and the surgical robot can work as intended. When they feel like it, they also contribute to the company’s thought translator by playing computer games.
Yes, Neuralink has a room full of monkeys that sit in front of computers and have their mind read, and it looks weirder than it sounds. For a couple of hours a day, they stare at laptop screens that are wheeled into position just outside the cages. They can choose from games that involve joysticks and touchscreens (such as tracing letters and spelling words) or games that depend on brain-controlled clicking. In one example, a 35-by-35 grid of small boxes appears on the screen, then one box suddenly lights up. The monkey’s goal is to move a cursor onto the lit box just by thinking it. The monkeys get faster at the tasks over time, and the expectation is that humans will, too.
It’s common for contract research centers to perform similar work by withholding food and water from the animals and pinning their skulls to metal rigs to hold their head in place. Instead of experiencing such Clockwork Orange-style torments, Neuralink’s monkeys snack on fruit and smoothies while they do their jobs and stop when they want. “Anytime they leave what we call the consent area, that means they’re done and that we need to back off,” Sorrells says. (The animals pick their food and soon their TV programs from an iPad.) She’s unapologetic about Neuralink’s desire to move fast if that means a human gets a life-altering device sooner. “It’s unethical not to be hyperfocused on this,” she says.
The company is beginning to shift the balance of its animal testing and much of its operations from California to Texas, where there’s a newer, bigger campus on 37 acres of ranchland outside Austin. This location has a surgery facility with multiple operating rooms, barns, a pathology building and a sci-fi-themed employee bar. Neuralink says it plans to build an indoor-outdoor space for its primates. Today, it has dozens of sheep and pigs penned there. During a recent visit, the pigs were wearing little backpacks that held batteries and fed power to a patch atop their head to keep their implants charged. The animals also have buttons in their pens they can press with their snouts to ask for food or a trip outside the barn.
It’s a long way from the snout button to 22,000 human implants. Just as with a rocket malfunction, a surgery gone wrong or an implant that leaks chemicals into someone’s brain would set the company back years. And beyond basic safety, the device has to live up to its promise. Humans will be able to tell the world things about the implant that monkeys can’t, including where its limits are. So far, downloading kung fu into your brain, let alone doing battle with an evil super-AI, remains sci-fi. Seo says that future implants will likely have 128 or more threads and that the next version of Neuralink’s custom chip should extend battery life to as long as 11 hours. “The goal is to get to a full day,” he says, at which point patients will be able to recharge their implant overnight via a charging pad built into a pillow.
The Elon of it all makes a lot of things tougher. And yet, if his bluster pays off here, he’ll have played a central role in massively improving the life of a great many people. He’ll deserve some of the credit even if it’s Synchron or Onward or somebody else that becomes the industry standard. If that thought makes your eyes roll, just try not to think about it too hard.
