If you have seen a video of a paralyzed person moving a cursor with their thoughts, you have seen a brain–computer interface (BCI) do its job. The idea is simple to state and hard to do: read the brain's electrical activity, decode the intention behind it, and turn that intention into a command a computer or a machine can carry out.
For readers who have never encountered the term, a BCI is a system that sits between the brain and the outside world. It does not read your mind in the science-fiction sense. It reads patterns of electrical activity that correspond to movements or intentions you are already trying to make — "I want to grasp," "I want to move left" — and translates those patterns into action.
Key takeaways
- A BCI has four stages: sense the brain's signal, decode the intention, act on a device, and feedback to the user so they can adjust.
- There are two roads: invasive (electrodes implanted in or on the brain) and non-invasive (sensors on the scalp or outside the skull, like EEG caps or ultrasound).
- Today the clearest, approved wins are medical: helping paralyzed patients regain hand control or supporting rehabilitation.
- The popular "read my thoughts" framing is a myth. Current BCIs detect motor intent, not language or memory.
- China has become an unexpected center of BCI activity, driven by regulatory approvals, a huge chronic-pain population, and heavy capital inflows in 2026.
How a BCI actually works
Strip the hype and a BCI is a pipeline. The brain produces faint electrical signals whenever it plans a movement or processes information. Electrodes — implanted or worn — pick those up. A decoder, usually a machine-learning model, learns which signal pattern maps to which intention. The decoder's output drives something: a cursor, a robotic arm, a wheelchair, or a synthesized voice. The user sees the result and adjusts, which trains both the person and the model.
The hard part has never been the electrodes. It has been the decoding. For years, faint signals for a sentence or a motion could not be turned into reliable meaning. In 2026, AI lowered that barrier by roughly an order of magnitude: multiple researchers publicly noted decoding accuracy rising from around 60% to above 90%, and some industry estimates say parts of the commercialization cycle shrank from three–five years to under one year.
Invasive vs non-invasive
The field splits on a single question: do you open the skull?
Invasive BCIs implant electrodes close to or into brain tissue. The signal is crisp, but every new patient means another craniotomy-level operation, so scaling depends on operating rooms and surgeons, not factories. In March 2026, China's NEO system, from BrainRobotics (博睿康), became the first implantable BCI to win commercial approval — ahead of Neuralink, which remained in a small human trial.
Non-invasive BCIs avoid surgery. EEG caps read signals through the scalp; a newer route uses ultrasound to read and write neural signals across the whole brain without cutting. The signal is noisier, but the addressable population is enormous. Chinese Academy of Sciences academician Zheng Hairong (郑海荣) has pointed to non-invasive ultrasound BCI as a route toward China's more than 300 million chronic-pain sufferers, with multi-center hospital trials underway.
What BCIs can do today
The real, approved use cases are medical:
- Helping spinal-cord-injury patients compensate for lost hand movement — a paralyzed person can "think" a hand toward a cup.
- Stroke and post-surgical neuro-rehabilitation, where feedback loops retrain impaired pathways.
- Non-invasive chronic-pain intervention and early screening, still largely in clinical-validation stages.
These are genuine medicine. They are also narrow: limited patient pools, high per-device prices, and heavy dependence on hospitals.
What they cannot do yet
Two myths are worth killing. First, BCIs do not read thoughts or memories. They detect motor intent and, in research settings, attempt to reconstruct speech from imagined talking — not silent mind-reading. Second, no consumer BCI today has a clear daily "刚需 (must-have)" use that survives contact with real life. Selling a headset is easy; answering "why would I wear this every day?" is not.
Why China matters here
Outside China, BCI coverage centers on Neuralink. But in 2026 the center of gravity shifted. NEO's commercial approval gave China the first cleared implantable device. Policy wrote BCI into the 15th Five-Year Plan (十五五, 2026–2030). And capital followed: one researcher estimated that in the first half of 2026 alone, domestic BCI investment ran into the tens of billions of yuan — on the order of 70 billion yuan (about $9.7 billion / HK$76 billion).
The strategic bet is not a single chip. As AI lowers the decoding bar, the moat moves to data and scenario — whoever sits closest to patients and doctors, and holds real clinical data, holds the entrance.
Honest limitations
This is an explainer, not a clinical or investment guide. Approval-status and comparison claims reflect company disclosures and reporting by outlets such as MIT Technology Review, not an independent regulatory audit reproduced here. The 60%–90% decoding-accuracy and "under one year" timeline figures are attributed to researchers and industry estimates quoted in coverage, not a single peer-reviewed source. The 70 billion yuan capital figure is a rounded industry characterization, not a published statistical total. Non-invasive ultrasound applications described remain in clinical-validation stages and are not yet approved products.
Sources
Reporting by Chinese state media and company disclosures; MIT Technology Review's characterization of NEO's approval status; remarks by CAS academician Zheng Hairong (郑海荣) at the 2026 Purple Mountain Brain Science Conference; and industry estimates cited in 2026 BCI investment coverage.