The difference between hype and reality in neurotech right now
Paralyzed patients are texting with their thoughts while CEOs promise to download your soul — here's what's actually happening.
If you’ve followed neurotech for more than five minutes, you’ve probably felt the whiplash. One week, Elon Musk is suggesting his brain chip will let you “save and replay memories.” The next, researchers quietly publish a study showing a 47-year-old woman who hadn’t spoken in 18 years just decoded speech at 47.5 words per minute using a cortical implant. The Musk quote gets three million impressions. The actual science gets a press release on a government website.
That gap — between what’s being sold and what’s being built — is the most important thing to understand about neurotech right now. This field is simultaneously more boring and more astonishing than the headlines suggest. Less “merge with AI by Thursday,” more “a paralyzed man can now text his family without using his hands.” Which is, if you think about it for a second, genuinely extraordinary.
The question worth asking isn’t “is this hype?” It’s “where exactly does the hype end and the real stuff begin?” The line shifts almost monthly. So let’s draw it.
What’s actually working right now
Start with the clearest proof point. An NIH-funded team led by Dr. Edward F. Chang at UCSF and Dr. Gopala Anumanchipalli at UC Berkeley implanted an array of electrodes over the speech-encoding area of a 47-year-old woman’s brain. She had been unable to speak or produce any vocal sounds for 18 years following a stroke. Their system, published in Nature Neuroscience in March 2025, decoded words and synthesized speech in increments of 80 milliseconds — less than a quarter of a second — and achieved a more than 99% success rate. For a 50-word vocabulary, it reached 90.9 words per minute. 🧠
That’s faster than many people type. And it’s not a one-off.
Synchron’s Stentrode is the other device that deserves real attention right now, partly because it reaches the brain without requiring a single drill to the skull. The Stentrode is delivered via catheter through the jugular vein and lodged in the motor cortex’s draining vein, where it records brain signals through the vessel wall. In Synchron’s COMMAND trial — the first FDA-approved investigational device exemption trial of a permanently implanted BCI — all six patients with severe bilateral upper-limb paralysis successfully met the primary endpoint: no device-related serious adverse events such as death or permanent increased disability.
The real-world implications are already showing up. Synchron recently demonstrated its BCI controlling Apple Vision Pro using only brain signals. That’s not a lab demo with a wired rack of equipment behind it. That’s a paralyzed person using a consumer device with their mind. 🔬
The things that are working right now:
Motor BCIs that let paralyzed patients control cursors, type, and navigate digital interfaces
Speech neuroprostheses decoding silent attempted speech into real-time audio
Endovascular implants (Synchron’s Stentrode) that avoid open-brain surgery entirely
Deep brain stimulation for movement disorders, already FDA-cleared and in use for decades
EEG-based neurofeedback for epilepsy monitoring, ADHD, and PTSD symptom management
None of these are science fiction. All are in human trials or cleared for use.
Where Neuralink actually stands
Let’s be precise about this, because imprecision has cost the field credibility. Neuralink is not a fraud. It’s also not what Elon Musk describes at his presentations. Two years into actual human trials, the company has successfully demonstrated brain-to-cursor control in human patients. What Neuralink has achieved is essentially a very expensive assistive mouse.
That framing stings a little, but it’s fair. The brain doesn’t think in binary code or discrete commands. Translating the messy, analog output of neural activity into precise digital signals requires massive computational interpretation, and that interpretation gets exponentially harder as you move beyond simple motor commands. The leap from cursor control to “downloading knowledge” isn’t an engineering challenge. It’s a categorical mismatch between the promise and what neurons actually do. ⚡
Durability is a real technical problem too. Utah arrays — the bed-of-nails style electrode implants used in many research BCIs — often lose signal from over 60% of their electrodes within one year, as scar tissue forms around the implant and muffles the signal. This isn’t a Neuralink-specific problem. It’s a fundamental biological fact: the brain’s immune system treats implanted electrodes as foreign invaders, because they are.
What Neuralink has contributed:
Proving the surgical robot can implant thousands of fine electrodes with precision
Demonstrating that patients with severe paralysis can use the system safely
Pushing the entire BCI field toward higher electrode counts and miniaturization
Attracting enough capital to fund serious competing research
The problem isn’t the science. It’s the gap between “paralyzed patients can control a cursor” and “you’ll be able to stream music directly into your brain.” Those are not on the same timeline. One exists today. The other may never exist as described.
Have you adjusted your own mental model of where BCI technology sits? If you came in thinking neural lace was five years away, has any of this shifted your expectations?
The signal degradation nobody talks about
Here’s the part that gets glossed over in investor decks. The brain is not a passive substrate for electronics. It’s a dynamic, immunologically active organ that remodels around implants. Any technology claiming long-term viability has to deal with neuroimmunology, not just microfabrication.
Today’s BCIs must be painstakingly trained to each user’s unique neural patterns, and performance can fluctuate daily with fatigue or mood. This is not a software problem you can patch overnight. It means every implanted BCI user is essentially running a custom-trained system that has to be periodically recalibrated. 🔬
The Columbia and Stanford teams published something striking in late 2025: a chip with 65,536 electrodes and 1,024 channels, streaming wirelessly. The lead researcher called it “a fundamentally different way of building BCI devices,” with “technological capabilities that exceed those of competing devices by many orders of magnitude.” That’s a genuine leap in channel count. Whether it translates into proportionally better clinical outcomes is the question that only long trials can answer.
The honest technical picture, broken down:
Invasive systems have better signal quality but cause scarring and signal loss over time
Endovascular systems (like the Stentrode) are safer but pick up lower-resolution signals
Non-invasive EEG is accessible but has latency of 800–1,200 ms and accuracy problems in noisy environments
Signal decoding is improving fast thanks to AI, but daily recalibration is still often required
Long-term durability remains the hardest unsolved engineering problem in the whole field
The good news is that AI-powered decoders are genuinely changing what’s possible with the signals that do come through. The UCSF speech neuroprosthesis works partly because modern deep learning can extract meaning from noisy, partial neural signals in ways that would have been impossible five years ago. The hardware is improving, but the software is arguably improving faster.
The consumer market: mostly noise, some signal
This is where the hype concentration is highest. Walk into any tech conference and you’ll see EEG headbands promising to “supercharge your meditation,” focus-tracking earbuds claiming to prevent burnout, and sleep wearables that offer to “decode your brain state.” Most of it deserves healthy skepticism.
“The problem with a lot of consumer EEG tech is that it doesn’t work well or at all,” says Ramses Alcaide, co-founder of Neurable. His company spent a decade and collected data from around 7,000 people to design EEG technology accurate enough to fit inside headphones while retaining meaningful signal quality. That’s a remarkably honest thing for a company selling brain-reading headphones to admit. But it’s also probably true of most of their competitors.
Professor Karl Friston at University College London, one of the world’s most influential neuroscientists, says the best way for a consumer to approach wearable EEGs is by treating them with the same level of reverence as a household thermometer. He’s right. These devices can tell you something real about your brain state — like a thermometer tells you something real about your body — but they’re nowhere near the diagnostic tools their marketing implies. 💡
Some specific things to watch in this space:
Neurable’s MW75 Neuro headphones, which embed EEG sensors in noise-canceling headphones and have actual signal processing research behind them
Muse S Athena, which launched in early 2025 combining EEG with fNIRS for a genuinely deeper picture of brain oxygenation during meditation
NAOX earbuds, a French startup building clinical-grade EEG into true wireless earbuds for longitudinal monitoring of conditions like epilepsy
Consumer neurotech firms now account for 60% of the global neurotechnology industry, outnumbering medical ones since 2018. Since 2010, consumer firms have proliferated more than four-fold. That’s a lot of companies. The signal-to-noise ratio in consumer neurotech is, fittingly, not great.
The thing that does deserve attention: the Centre for Future Generations’ 2025 neurotech market atlas notes that miniaturization and AI are genuinely accelerating integration. EEG sensors are getting small enough to live in earbuds. The question of “when does passive brain monitoring become part of everyday wearables” is not crazy to ask. It’s probably a decade away, not two.
What the next 18 months actually look like
Paradromics received FDA IDE approval for its Connexus system to start the Connect-One early feasibility study, targeting speech restoration and computer control in people with severe paralysis via a high-bandwidth, fully implantable BCI. Unlike cursor control, this trial is designed around speech restoration as a primary endpoint — which means if it succeeds, we’ll have direct evidence of a device designed from the start to give people back their voice. That matters. 🚀
The realistic outlook for the next two years: expansion of endovascular BCIs into post-stroke aphasia and upper-limb amputation, with a focus on reliability rather than bandwidth. Expect two or three new FDA De Novo clearances for communication-focused systems.
What’s not coming in the next 18 months:
FDA approval of any BCI for general consumer use
Reliable long-term implants without signal degradation
Anything resembling “thought uploading” or memory augmentation
Consumer brain chips that actually work the way the pitch decks describe
What is coming:
More human trial data from Synchron, Neuralink, Precision Neuroscience, and Paradromics
Improved AI decoders that extract more from the same electrodes
Real competition between companies driving faster iteration
Growing regulatory clarity on what “BCI safety” actually requires long-term
MIT Technology Review called BCIs one of its 10 Breakthrough Technologies of 2025, and research scientist Michelle Patrick-Krueger at the University of Houston characterizes this as “the translation era” — a period of considerable private investment focused on moving BCI from research into clinical practice.
That framing is exactly right. This is a translation era, not a deployment era. The science is outrunning the infrastructure, the regulation, and in some cases the honesty of the people selling it.
The question that keeps me up at night isn’t whether BCIs will eventually work — the UCSF speech results make that essentially certain. It’s whether the companies racing to be first will cut corners on long-term safety data to get there. Because the brain deserves more patience than a quarterly earnings cycle allows.
What would it take for you to actually trust a brain implant enough to consider one?