Did Neuralink make the wrong bet?

Elon Musk promised Neuralink would bring superhuman abilities and minds merged with AI. Then he fueled a runaway hype train for his brain implant technology, which ended up with a grisly record for implants in monkeys and some success with human subjects. But for all of the hype, he’s still further away than Mars from his goal. And that’s because his relentless ambition is once again hitting the wall of scientific reality.

The heart of the issue is how brain-computer interfaces (BCIs) translate thought into results. Neuralink’s products have all been brain-to-cursor interfaces, which allow patients to control a mouse with their minds. But Neuralink’s competitors have raced ahead with newer BCIs that translate thought directly to speech. Turns out that’s a more promising approach — enough to convince Neuralink to quietly invest in BCIs that focus on speech.

Musk has a strong record of overpromising and underdelivering, and his biggest quagmire may end up being his pursuit of a grand, unified vision of a human-AI-hybrid technology. When it comes to the human mind, he’s underestimated and oversimplified the steps it will take to make meaningful brain-computer interfaces a reality for patients who really need them.

BCIs are similar, but there’s a big difference

All BCIs connect a brain to a computer with wires or Bluetooth. They stalk the tiny bursts of electricity your neurons use to talk to each other and then try to make sense of them so that they can predict what you might want to do in the future. The key difference between BCIs is the type of behavior they’re trying to emulate.

A motor BCI, like the one Neuralink has been building, helps users guide a cursor across a computer screen. Unlike those, speech BCIs translate brain waves into sounds and small sections of words called phonemes. In the span of five years, speech BCIs have reached impressive milestones that rival the achievements of the two-decade-old motor BCI technology. A 2019 study reported that a speech BCI could predict what a person planned to say when given only a few options. By 2024, a 45-year-old ALS patient could speak naturally with 97 percent accuracy using his speech BCI.

In November 2025, Neuralink patient Brad Smith showed The Verge his motor BCI. He thought about moving his arm, which he could no longer move due to ALS, and instead the computer cursor moved across the screen. For speech BCIs, it’s words or chunks of words. Patients think about speaking the word “good,” for example, and the word appears on the screen. It is not mind reading — it is detecting what they’re trying to say.

Here is the catch: Both versions are technically motor BCIs. The underlying neuroscience is the same. If you move your finger, your brain is sending signals down into the muscles in your pinky. If you talk, your brain sends similar signals down into your tongue and other muscles that help you form sounds. The BCI detects what muscle the user is thinking about moving, whether tongue or finger, and predicts what they’re trying to do or say.

Neuralink is now course-correcting to be in line with the rest of the BCI community: In May, Neuralink began recruiting patients for a clinical trial to study speech restoration at the Cleveland Clinic Abu Dhabi hospital in the United Arab Emirates; in October, it launched a speech restoration trial in the United States at the University of Texas Southwestern Medical Center. The patients will use the same hardware as the current Neuralink patients but for the goal of turning their thoughts into speech rather than cursor movements. The company has already claimed success in a video posted to X on March 24th of a speech BCI trial participant who can still speak but whose speech is hard to understand because of ALS.

Speech BCIs seem to be the future of the field, but it remains to be seen whether the technology will speed past motor BCIs to market or simply offer another technology option to patients with different needs.

Neuralink has been making moves to step into its commercial era. The company hired a former director of the FDA office that oversees medical devices like BCIs to head its medical affairs, and Musk announced that Neuralink will begin “high-volume production” of the devices in a post on X on December 31st, though any Musk production predictions need to be taken with a grain of salt. As Musk’s medical company falls in line with the broader BCI field, does it also drift further from his vision of human enhancement and back to regular medical assistance for those who need it most? It is unclear.

Space is hard; the brain is harder

Sergey Stavisky was one half of the leadership team for the 2024 speech BCI research study out of the University of California, Davis that set a high bar for speech BCI accuracy. Stavisky was a former motor BCI researcher but pivoted to speech BCI in 2019 to make rapid progress in a field that looked to him ripe for success. “It seemed like it was a bit of an untapped opportunity,” he said. This has borne out, he said, noting how speech BCIs quickly expanded the size of their vocabulary from only 50 words to “being able to say any word in the dictionary,” he said.

But he doesn’t think that Neuralink made the wrong bet to focus on motor BCIs when the company formed in 2016. At that time, academic research into motor BCIs had matured enough for industry to step in, he said. “I think at that time, cursor control was sufficiently de-risked by academic trials that it was clear that with better hardware, a very useful medical device could be built,” he said. (Stavisky has been a paid consultant for Neuralink in the past, but he did not provide details because he signed a non-disclosure agreement. It is not uncommon for academic BCI researchers to consult with for-profit BCI companies. Stavisky is tangentially working with Neuralink’s competitor Paradromics on its upcoming clinical trial through his coinvestigator at Davis.)

Matt Angle, CEO of Paradromics, disagrees. Neuralink did make a mistake by focusing on motor BCIs, he told The Verge. Paradromics started one year earlier, in 2015, with speech as its first priority. Like Stavisky, many top Paradromics scientists come from the motor BCI research field.

Speech is a better first application of BCI technology than motor restoration, from Angle’s perspective, because it’s “the biggest quality-of-life deltas that you can imagine,” he said, “being able to talk to your loved ones again — and it’s something that BCI can do today.”

I asked Angle why a motor BCI might not be as valuable to a patient unable to talk as a speech BCI given that both result in words spoken aloud by a computer program. I witnessed Neuralink patient Brad Smith use his motor BCI to communicate in a real-time conversation with me and his wife in November. Smith typed out answers to my questions letter by letter, word by word, with his mind-controlled computer cursor. Smith told me that Neuralink changed his life for the better.

Speed limits motor BCIs, according to Angle. (Smith typed out his 16-word response to my question in one minute and 17 seconds.)

“If I lost the ability to communicate and my primary means of communication was the BCI, I would like to have speech back,” he said. Still, he is quick to note that all BCIs, speech and motor, should exist: “I don’t think it’s for us to armchair what someone with a disability would or wouldn’t want,” Angle said.

Looking further, AI chatbots seem like an obvious complement to speech BCIs. The two technologies are tangentially related: BCIs are already built on algorithms similar to the large language models powering AI chatbots, and many people with speech impairments use predictive word software — again, somewhat related to LLMs — to pick out which words or phrases they most likely want to say next. (Smith used text-to-speech app Proloquo4Text in conjunction with his Neuralink BCI.) Speech BCIs could make it easier and faster, with fewer clicks, to input prompts into AI chatbots, and access the benefits of agents and agentic browsers (when they work) to navigate the virtual world.

Patients want all types of BCIs

Former BCI user Ian Burkhart was unable to speak or move during the two weeks following a diving accident in 2010 that resulted in a spinal cord injury. Communication emerged as “a huge, huge priority” during that time, more so than being able to move, he said. Burkhart now appears to speak with relative ease and has recovered partial movement of his hands. But he said he would still like a speech BCI today, just for the ability to rapidly input text into a computer.

This seems noteworthy given that Burkhart is one of the several dozen people in the world to actually use a motor BCI. He was part of a roughly seven-year-long clinical trial at The Ohio State University, where he controlled a computer cursor and played Guitar Hero with this brain. He also became the first person to reanimate some muscles in his body using electrical stimulators controlled by his thoughts.

If forced to choose between speech or motor BCIs, ALS patient Spero Koulouras told The Verge in a written comment: “for me it’s motor by a mile.” A former software engineer and entrepreneur, Koulouras says that he is “effectively quadriplegic and mute” over six years after his diagnosis with ALS in 2019. He communicates entirely through his computer and spends much of his day writing code and doing 3D design, all of which contribute to his preference for a mind-controlled computer cursor rather than a brain-to-speech BCI.

Both technologies come with downsides, Koulouras noted. Speech BCIs cannot enable him “to be fully functional in [his] virtual environment,” he said. “But family gatherings are torturous,” he said, even though he uses prerecorded phrases to make a point within a conversation. “The inability to joke, snark, and harass friends and relatives in real time is emotionally devastating… Motor control today can’t provide the communication speed to be an active participant.”

Koulouras was not selected to join Neuralink’s motor BCI trial after the company evaluated him in February 2025. He is not sure why but guesses that his existing technology works well — too well, perhaps. He uses a motion tracker device called Cato that attaches to his glasses and translates subtle head movements into cursor movements on a screen. Koulouras is the cofounder of the company behind the device, Auli.Tech. “I believe my proficiency with my current tech may have factored into Neuralink’s decision. As a clinical trial I may not have had as much potential for improvement, negatively impacting reported results,” he said. In June 2025, Neuralink contacted him again for its speech BCI trial, but his low respiratory scores would have required him to get a tracheostomy, which he declined.

Koulouras’ experience highlights just how inaccessible BCI technology is for most patients. Potential BCI users need to meet a long list of criteria to be considered for a trial, after meeting the most obvious criterion of simply living near a trial location. Advocacy groups the ALS Association and the ALS Network, which connected Spero to The Verge, include information or host events about BCIs on their websites, but the bulk of their efforts are focused on advocating for insurance reimbursement for necessities like wheelchairs, navigating healthcare denials, and increased research funding.

“From a cursor guided by thought to speech restored directly from the mind, every advance in brain–computer interfaces represents real progress for real people,” ALS Network president and CEO Sheri Strahl wrote to The Verge. “Each breakthrough – whether restoring movement, communication, or autonomy – expands dignity and quality of life. It all matters, and it’s encouraging to see so many innovative scientists taking different approaches toward the same deeply human goal.”

“What’s the market?”

There is the question of what patients want, and there is another question of how many patients might benefit from it. In other words, “What’s the market?” associate professor Kip Ludwig asked when speaking to The Verge. Ludwig leads an institute focused on neuroengineering at the University of Wisconsin-Madison, where he studies how electrical zaps to the body’s nerves can treat heart failure and other complex disorders. For all BCIs, “it’s incredibly small for an incredibly expensive technology,” he said. Most motor BCI patients have either ALS or a paralysis from a spinal cord injury. There are roughly 30,000 ALS patients and 300,000 patients with traumatic spinal cord injury in the US, according to recent estimates. In order to enter a BCI clinical trial, participants must also live within a several-hour drive of the trial site, have a caregiver who can assist them, and not have other serious medical conditions like epilepsy or anything requiring regular MRIs.

Motor BCI companies, therefore, have to find other patient populations that might benefit from their technology. Stroke patients with a less severe motor dysfunction than full quadriplegia are an obvious target population. But, if the spinal cord is doing its job and sending signals from the brain and out to the nerves, then these same patients don’t really need a brain surgery, Ludwig argued. Motor BCIs are “an invasive version of something I can do less invasively in the periphery,” he said.

Speech BCIs, in contrast, might be a good fit for stroke patients, according to Paradromics CEO Angle. The company is first focusing on a small group of patients with ALS or an injury that affects muscles or nerves. As the trials of speech BCIs located in the motor cortex progress, Angle said the company plans to launch more clinical trials in other parts of the brain, like the superior temporal gyrus, which has been shown to encode spoken speech and internal speech, like an inner monologue. Tapping into the STG can open up the patient pool to those with strokes in the motor region of their brain, and who can no longer speak. After these small feasibility studies show that speech BCIs are safe, like all clinical trials, later studies will include more and more patients so that enough data can convince the FDA that the tech is so useful that it should come to market.

The reality of augmentation

Perhaps the largest divide within the BCI industry is not speech versus motor, but augmentation versus medical assistance. At the company’s 2019 launch event, Musk set Neuralink’s ultimate goal as a “full brain-machine interface,” which he defined as “a sort of symbiosis with artificial intelligence.” Motor BCIs were the necessary stepping stones to his eventual goal of augmenting any human who wants a BCI to achieve superhuman AI incorporation. Neuralink first needed to “solve” several “issues” related to “brain disorders” like Alzheimer’s or dementia, as well as paralysis resulting from broken or injured spines.

But the theory behind augmentation has a major flaw: Evolution capped how much information can flow from the brain to the body, associate professor at University of Wisconsin-Madison Kip Ludwig told The Verge. “In reality, we’re limited by our own physiology,” he said. Even if BCIs got super fast at decoding the brain’s signals, we would not be able to make the most of it, he said. “Evolution did a great job.”

“There’s this false assumption that they can get so good at brain-machine interfaces that they can decode from the brain faster than we can encode with our natural body typing or swinging a baseball bat or things like that,” Ludwig said. He is quite familiar with the “natural rate” of information transfer — he measures the brain-to-organ latency rate as part of his own research exploring the ways that electrical zaps to the body’s nerves can treat complex disorders like heart failure. Motor BCIs could, in theory, shave 200 milliseconds or so off someone’s reaction time, he said. That is roughly how long it takes for a command from the brain to travel down nerves into muscles and cause a movement. But that isn’t that useful to people trying to regain independence in doing tasks at home, he said.

For now, speech BCIs don’t seem to fit into the futuristic vision of human augmentation, Ludwig noted. It could get more sci-fi if the technology moves from motor regions of the brain that control the mouth to areas that tap into abstract ideas of language — and could decode someone’s inner monologue.

The “bummer” of commercial BCI efforts

Technical success does not necessarily translate into commercial success, as seen by the boom-and-bust cycles of many medical device companies attempting experimental technologies. A pair of companies providing retinal prostheses to partially blind patients offer two unrelated examples. Both Second Sight Medical and Pixium Vision went bankrupt and left patients stranded with unserviceable technology; both also had their IP bought, and their patients rescued, by newer medtech ventures, one of whom was Science Corporation, founded by Neuralink cofounder Max Hodak.

Blackrock Neurotech may boast over 19 years of testing in humans, but the company has pushed back the year that it expects to commercialize its at-home motor BCI system called MoveAgain. In 2021, the company predicted that it could bring MoveAgain to market within the year. In 2022, I spoke to the company’s cofounder and then-president, now chief science officer, Florian Solzbacher for STAT News. Only one document required by the FDA stood between the company and its commercialization goal of 2023. “We are quite confident that this will work,” Solzbacher said at the time.

But the deadline came and went. In 2024, when the investment arm of crypto company Tether took a majority stake in Blackrock Neurotech, the announcement lacked mention of a timeline for commercialization. Blackrock Neurotech did not respond to The Verge’s multiple requests for comment on the commercialization delay.

“It’s a bummer,” Burkhart said of the delay. While he occasionally consults with Blackrock Neurotech, he can only surmise the reason for the delay. Medical insurance reimbursement tops his list. Home devices are always a pain to get reimbursed by insurance companies, which disabled people know all too well. Motor BCIs are a particularly unique device with no precedent, he said. “There’s no medical justification that says people need to be able to use a computer or use a robotic arm or use a muscle stimulation device, anything like that,” he said.

Speech, in contrast, does have precedent. “But there is medical justification for people being able to accurately convey their health needs,” he adds. The vast number of speech generators or alternative communication devices already FDA-approved and reimbursable by insurance might make the reimbursement pathway for speech BCIs a “little bit cleaner” compared to motor BCIs, he said.

As of June 2025, Neuralink has implanted between five and 12 humans — reports vary and Neuralink did not respond to our requests for an exact count — since the first patient was implanted in January 2024. While impressive, Neuralink trails Blackrock Neurotech’s 52 total patients by several dozen.

It remains to be seen whether speech BCIs can leap-frog traditional cursor-based motor BCIs to the commercial market. Motor BCIs have the advantage of patient use at home, which the FDA will use to evaluate the safety of the technology. Speech BCIs, meanwhile, have only been used in controlled lab settings.

And yet, Angle is unconcerned about which type of BCIs will come to market first. He is convinced that whenever patients have the option to speak again with a speech BCI, they’ll choose to get the device. It’s the adoption of the technology that matters more to him.

“It’s about making sure that we’re launching not a gee-whiz gadget but an actual medical device that meets an important unmet medical need and is delivering value to the people who get it.”