A future-historical field guide · Current to 21 June 2026

The Brain Interface Atlas

Read / Write / Merge

The BCI race is not one race to read minds. It is a trade-off frontier: every step toward higher bandwidth moves closer to living tissue, while every step toward mass adoption demands less risk, longer stability, clearer agency, and stricter governance.

This atlas separates peer-reviewed evidence, regulator or company reporting, and forward speculation. “Bandwidth” means useful, repeatable agency—not electrode count, a promotional demo, or a model guessing from context.

1commercial implant approval
NEO · China
21Neuralink participants
announced January 2026
56 wpmaverage long-term home speech
one independent study
0general-purpose consumer
high-bandwidth implants
Field view
Filter by route
I · The signature instrument

The bandwidth–burden frontier

Twenty consequential organizations plotted by neural proximity and surgical burden, clinical evidence and regulatory maturity, and intended signal coverage. The empty destination is a mature, low-burden interface with a large bubble: high useful bandwidth without asking the body to pay for it.

Left → right: farther from tissue to penetrating cortex · bottom → top: research to approved use bubble area = intended neural coverage / bandwidth ambition, not proven bitrate ◉ current regulatory landmark
No company occupies the promised corner. Non-invasive systems are safer but information-poor or indirect; penetrating systems can reach richer signals but inherit surgery, tissue response, maintenance, and longevity. Surface and vascular systems are betting that the commercially useful optimum lies between the extremes.
Click any node to open its dossier or locate it in the roster. Position is an editorial synthesis of publicly reported evidence as of 21 June 2026—not a medical recommendation or investment ranking.

What “top 20” means here. Consequence to the field: human evidence, regulatory position, technical distinctiveness, ability to productize, capital and ecosystem influence, and geographic importance. It is not a valuation table and does not imply that #1 has the best clinical device.

Evidence discipline. “Human” may mean acute testing during another surgery, a chronic implant, or a company-reported trial. Those are not equivalent. The stage column keeps them separate.

II · The operating logic

A BCI is a loop, not a mind-reading chip

A useful interface must acquire a signal, identify a task-relevant pattern, translate it into an action, return feedback, and remain usable as both brain and device change. The narrowest link governs the whole system.

Useful agency ≈ signal × stability × decoder × feedback × independence ÷ burden
A thousand electrodes do not create a thousand channels of meaning. Calibration time, drift, false activations, caregiver dependence, surgery, battery life, software support, and the user’s own learning all consume practical bandwidth.
Read

Decode an intention or state

Movement attempts, imagined handwriting, attempted speech, attention, fatigue, or a population-level neural state. Today’s strongest systems are task-specific decoders—not unrestricted access to private thought.

Write

Return information to the nervous system

Electrical, magnetic, optical, chemical, or ultrasonic stimulation can evoke sensation or alter circuits. The write channel remains coarser and harder to validate than the read channel.

Close the loop

Turn decoding into restored agency

The clinically important endpoint is not a neural signal. It is speaking at home, grasping a cup, moving a cursor, standing, reducing symptoms, or controlling a device without a research team in the room.

III · The deep lineage

From electric rhythm to independent use

The history is less a march toward telepathy than a repeated expansion of the control loop: from observing brain activity, to conditioning it, to decoding action, to returning sensation, to sustaining use outside the laboratory.

IV · The global field

The frontier ten—and the next ten shaping the map

The dossiers apply one frame to every company: interface route, human evidence, strategic reading, near-term unlock, and hard bottleneck. The emphasis is on what has crossed into people and daily use, not on the most cinematic promise.

01 · RouteHow close to tissue, and by what surgery?
02 · SignalWhat is read, written, or restored?
03 · EvidencePreclinical, acute human, chronic, or approved?
04 · EdgeWhat could become defensible?
05 · BottleneckWhat can still prevent a real product?
The next ten

Different wedges into the same frontier

These organizations matter because they test distinct materials, therapeutic endpoints, geographies, or low-burden routes—even when their evidence base is earlier or their product is not a general communication BCI.

V · The route map

Nine ways to meet a brain

“Invasive versus non-invasive” hides the actual design space. Interfaces differ in where they sit, what physical signal they measure, whether they can write back, how much tissue they cover, and what risks they import.

The read channel

Already useful in constrained tasks

Motor intent, attempted speech, handwriting, broad cognitive state, and some sensory variables can be decoded when the task and training distribution are known.

The write channel

Still sparse, coarse, and ethically heavier

Electrical stimulation can evoke localized touch or modulate circuits; ultrasound and optical methods are promising. None can safely write arbitrary concepts, memories, or full-fidelity experience.

Active field · June 2026

China crossed the first commercial threshold

On 13 March 2026, China approved Neuracle Medical’s NEO for a narrow hand-function indication in people with cervical spinal cord injury—the first commercial authorization of an implantable BCI. It is a major regulatory milestone, not the arrival of a general “brain chip.” S12S13

The more useful comparison is not “China beat Neuralink.” NEO uses eight epidural electrodes and drives a robotic glove; Neuralink pursues dense penetrating recording and general digital control. China’s near-term advantage is regulatory deployment and cohort scale across lower-penetration routes. The United States still holds much of the deepest independent intracortical evidence.

ApprovedNEO, for a specific motor-compensation indication in China.
InvestigationalNeuralink, Synchron, Paradromics, ONWARD, CorTec and most chronic implants.
Human, but acuteSome surface, graphene, soft-electrode and ultrasound studies during clinical access.
Preclinical ambitionFull biohybrid, molecular-ultrasound and experience-transfer systems.
VI · What works now

Remarkable demonstrations, narrow generalization

The field has crossed from proof-of-concept into meaningful daily agency for a small number of people. It has not crossed into reliable, broadly available, general-purpose neural computing.

VII · The promise and the price

The same channel can restore agency—or become an instrument of power

Medical utility is the strongest reason to build BCIs. It is also the wedge through which neural data, stimulation authority, cloud dependence, and augmentation markets may enter ordinary life.

The promise

The perils

A global line is beginning to form: UNESCO’s 2025 Recommendation on the Ethics of Neurotechnology centers mental privacy, freedom of thought, protection from undue influence, and safeguards against surveillance or social control. These principles will matter most before consumer and workplace uses normalize—not after. S38
VIII · Capability horizons

From restored control to the speculative merge

The responsible way to discuss the future is to change the evidence standard with distance. Near-term claims inherit present data. Far-future claims require multiple unsolved capabilities to compound.

The telepathy ladder

“Brain-to-brain communication” already exists in a technically literal but semantically tiny sense. The ladder from binary task signals to shared experience has missing rungs, especially on the write side.

What “brain-to-cloud” means in 2026: neural activity is decoded by local hardware and software, then used to operate an application that may itself use cloud AI. It is not a transparent upload of thought. As models enter the loop, they can amplify sparse intent—but they also introduce inference error, platform dependence, surveillance risk, and ambiguity over whether a phrase originated in the person, the decoder, or the language model.
Cross-species · plausible path

Translation will probably arrive before telepathy

Multimodal AI can combine vocalization, posture, gaze, physiology and ecology to infer an animal’s likely state or request. Neural recording may validate those models; stimulation or sensory prostheses may return a small learned signal. The result could be useful communication without ever decoding an animal’s inner monologue.

Cross-species · hard limit

There may be no shared dictionary to recover

Brains are shaped by species-specific bodies and senses. A bat’s echolocation, an octopus’s distributed control or a dog’s olfactory world may not map cleanly into human nouns. The realistic target is negotiated signals and shared tasks—not English subtitles for raw animal thought.

IX · 2030 +

Six futures, not one inevitable merge

Technical capability does not choose the dominant future. Reimbursement, surgical tolerance, ownership, regulation, public trust, and whether non-invasive modalities break the frontier will decide what scales.

X · Calibration

Ambition, marked by confidence

The bars are evidence confidence, not numerical probabilities. They ask how much present evidence supports the direction and timetable of each claim.

XI · The recurring loop

How a neural demo becomes infrastructure

Every durable BCI platform must traverse the same sequence. Most projects are still between signal acquisition and chronic independence; public conversation is already arguing about the final steps.

The decisive transition is not brain signal → cursor movement. It is research apparatus → dependable part of a person’s life.

The closing argument

The first BCI revolution is not mind upload. It is agency restored—one cursor, word, hand, sensation, and step at a time. The second begins only if read and write become safe, durable, mutual, and governable enough that the interface still belongs to the person wearing it.

Bandwidth is not sovereignty