A child hasn't seen her mother's face in three years. Her retinas are intact. Her brain is perfectly capable of vision. The only barrier is 1.2 millimetres of damaged tissue in her optic nerve — standing between her and the world. What if that pathway could be bypassed entirely?

The scale of the problem: visual impairment as a signal failure

Vision loss ranks among the most irreversible neurological conditions globally. For millions of patients, the cause is not a failed visual cortex — it is a broken signal pathway. Damaged photoreceptors from retinitis pigmentosa, age-related macular degeneration (AMD), or Leber congenital amaurosis; severed or degraded optic nerves from glaucoma, traumatic injury, or ischemia — these are infrastructure failures. The cortex sits ready. The signal never arrives.

This is the core challenge that visual brain-computer interface (BCI) technology is now positioned to solve: not by repairing the damaged tissue, but by routing around it.

• 285M:People living with visual impairment worldwide
• 39M:Classified as completely blind globally
• 80%:Of cases considered preventable or treatable

The bypass paradigm: inserting signal at multiple levels

The visual system is not a single channel — it is a hierarchical pathway. Visual neuroprosthetics exploit this architecture by inserting electrical signal at different anatomical levels, depending on where the damage lies. This is the defining insight behind the bypass paradigm.

1.Retinal-level intervention — subretinal and epiretinal implantsWhen photoreceptors fail but ganglion or bipolar cells survive, subretinal implants (such as Pixium Vision's PRIMA) and epiretinal devices stimulate remaining retinal neurons directly. Commercially deployed in Europe, these represent the most clinically mature form of visual prosthetics for photoreceptor loss.

2. Optic nerve stimulation — the underexplored corridorFor patients with intact retinas but damaged optic nerves, direct optic nerve stimulationoffers a minimally invasive intermediate pathway. Cuff and intraneural electrodes placed around or within the optic nerve can restore partial signal transmission, bypassing the damaged fibre region while preserving anatomical proximity to natural visual processing.

3.Cortical-level intervention — direct V1 stimulationThe most radical approach: electrode arrays implanted on or within the primary visual cortex (V1) bypass both the eye and optic nerve entirely. The brain interprets stimulation pulses as phosphenes — points of light — which can be spatially patterned to convey rudimentary but meaningful visual information. This is the frontier that Neuralink, Cortigent, and the Monash Gennaris device are competing on.

KEY SCIENTIFIC INSIGHT

Neuroplasticity is the real asset. The human brain, given consistent, spatially coherent, temporally stable signal, will reorganise to interpret it as vision. Engineers do not need to perfectly replicate the retina's output — they need to deliver reliable signal, and the cortex does the rest. This is the biological foundation on which the entire visual BCI thesis rests.

The convergence driving feasibility in 2026:

Three simultaneous advances are shifting visual neuroprosthetics from experimental to clinically translatable:

Miniaturised electrode density. Higher electrode counts per device mean more addressable phosphene "pixels." Science Corp's Science Eye targets 16,000+ electrodes — compared to the 60 in the original Argus II. Resolution is not yet natural vision, but it is approaching functional threshold for navigation, face detection, and text reading.

Wireless transcutaneous power delivery. Eliminating percutaneous cables — the single greatest infection and reliability risk in implantable neurotechnology — is now achievable at the power levels required for cortical stimulation. This is what makes long-term implantation realistic outside of clinical trial settings.

AI-driven neural encoding. Camera input must be translated into stimulation patterns that the individual patient's cortex will interpret as coherent vision. AI models trained on patient-specific neural responses are replacing fixed encoding algorithms — and this is where the most significant IP value in the field is now being created.

"The implant is the platform. The intelligence is the product."

Who is building the visual BCI stack ?

A genuine ecosystem has formed around visual neuroprosthetics, with distinct layers of the intervention stack now competed on independently.

Cortical

• Neuralink:N1 chip, 1,024 electrodes. Blindsight program targeting V1 implantation in visually impaired humans.
• Cortigent:Orion cortical prosthesis. Ongoing clinical trials. Successor to the Argus II programme.
• Monash / Gennaris:Wireless cortical tiles over V1. University spinout with growing institutional backing.

Retinal:

• Pixium Vision:PRIMA subretinal photovoltaic implant. CE-marked for dry AMD in Europe.
• Science Corp:Science Eye — 16,000+ electrode high-density retinal prosthetic. Founded by ex-Neuralink team.

Optic nerve:

Emerging players:Bioelectronics companies targeting the optic nerve — the underexplored middle layer.

Regulatory infrastructure is maturing alongside device development. FDA Breakthrough Device designation is accessible for visual neuroprosthetics with strong safety profiles, and the EU's Medical Device Regulation is generating commercial precedent following PRIMA's approval. Reimbursement coding, surgical training networks, and vision rehabilitation protocols are being built in parallel — the commercialisation layer is forming.

The Celvion perspective: personalised neural encoding is the real moat

At Celvion, we believe hardware is no longer the limiting factor in visual BCI outcomes. Electrode counts are sufficient to begin conveying meaningful perceptual information. What constrains results is the absence of a personalised neural encoding layer — a system that accounts for the fact that no two visual cortices reorganise identically after vision loss.

Every patient who loses vision through a different mechanism, at a different age, over a different timeline, develops a differently reorganised cortex. Applying a generic stimulation pattern to that cortex is analogous to issuing the same prescription lens to every patient without an eye examination. It will work for some. For most, it will fail.

The visual BCI companies that define this decade will treat the neural encoding algorithm as a continuously learning software product — adaptive, patient-specific, and improving with use. The implant is the hardware platform. The personalised encoding intelligence is the product with genuine defensible IP.

We are also directing research attention toward optic nerve stimulation — a largely underexplored anatomical target with significant clinical potential. 

For patients with intact retinal function but degraded optic nerve transmission, nerve-level stimulation may offer a less invasive, more anatomically proximal path to restored vision than full cortical implantation. 

This corridor deserves serious scientific and commercial attention.