A person who lost their sight does not lack a visual cortex. The pathway from the eye is gone, but the brain's screen is often still there, dark. The boldest corner of the brain–computer interface (脑机接口) field asks whether a chip could write pictures directly onto that screen — not restore the eye, but replace the signal that used to reach it.
That is the wager behind Xinsheng Vision (芯生视界), a startup that in June 2026 closed a near-100-million-yuan (≈ US$14m / HK$110m) seed round led by Matrix Partners China (经纬创投), with Tsinghua-linked Shui Mu Ventures, Starlink Capital and Yanyuan Ventures following. The money is earmarked for an invasive visual cortical prosthesis (视觉皮层假体).
Why sight is the hardest target
Visual restoration is widely described inside the field as among the most difficult directions in brain–computer interfaces (脑机接口). Speech or movement decoding can tolerate error; a mistaken word is embarrassing, a missed grip is recoverable. Vision is unforgiving: the brain expects edges, colors, depth and motion in a precise spatial grammar, and a wrong pattern is not a stutter but noise.
Xinsheng Vision (芯生视界) was spun out of a multidisciplinary brain–computer interface team at Peking University and ShanghaiTech University. Its stated approach is to build what it calls a "ten-thousand-channel neural graphics card" — a high-channel-count neurostimulation chip that renders external images into the stimulation language a visual cortex can interpret, paired with reconstruction algorithms and a fully implantable system. The company says the funds will push preclinical validation and a core engineering team.
The data problem nobody can skip
The startup's argument rests on a biological fact more than a hardware one. Animal models do not transfer: rodents barely have color vision, and the layout of early visual cortex differs from humans. A system meant to give a person real sight, the company argues, must be built on human visual-cortex data, not extrapolated from mice. Xinsheng Vision says its founders spent more than a decade recording from the human visual cortex — single neurons, electrical stimulation, and the subjective reports of what a person "sees" when a spot is triggered.
That claim is the company's own and is not independently verified. It is worth stating plainly: a seed-stage company describing a decade of proprietary human neural data is making a credibility bet, not presenting evidence. The interesting question for readers is whether such data, if it exists, can be turned into a device that clears a regulatory bar rather than a demo reel.
What "seeing" might mean in practice
Phosphenes — points of light the brain reports when a spot of visual cortex is stimulated — are the crude alphabet here. Early visual prostheses have let users count objects or trace outlines. The leap Xinsheng Vision is describing is from isolated light points to structure: boundaries, color and spatial layout written at scale. The "ten-thousand-channel" framing is the company's name for the channel count it believes that leap requires. Treat the number as an engineering target, not a measured result.
The broader field is moving on parallel tracks. At a Shanghai technology fair in June 2026, an implantable visual-reconstruction system from Mingshi Medical (明视医疗) was demonstrated alongside other brain–computer interface devices, a sign that several Chinese teams now treat sight restoration as a live commercial target rather than a distant dream.
Why China is leaning in
Visual cortical prostheses sit squarely inside the national push that put brain–computer interfaces (脑机接口) into the "15th Five-Year" planning framework as a future industry, and the field benefits from the same June 30, 2026 device-classification rules that placed implantable brain–computer interfaces (脑机接口) under unified Class III management. For a country with millions of people living with blindness or severe visual impairment, a domestic visual prosthesis is a strategic as well as a humanitarian prize.
Trend reading
According to Li An, chief scientist at BrainNet (脑机网), visual restoration will be gated less by chip fabrication than by the quality of human visual-cortex data a team can ethically and legally accumulate — without that, even a perfect electrode decodes into static.
Honest limitations
- The funding round, lead investor and participating funds are reported by China Daily, a state outlet; the corporate valuation and use-of-funds detail are company-disclosed and not independently audited.
- The "ten-thousand-channel neural graphics card," the human-data advantage and the high-process chip claims are the company's own statements, not peer-reviewed results or regulatory filings.
- No clinical trial, patient outcome or device prototype performance has been published; the work is at the preclinical and seed-financing stage.
- Comparisons to foreign efforts such as a rival's process node are company framing and should not be read as verified benchmarking.
What readers can do now
- Researchers: the bottleneck is human visual-cortex data with matched stimulation-and-perception labels — partnerships with ethics-cleared neurosurgical centers, not just chip design, are where progress will be made.
- Clinicians: treat any visual-prosthesis claim as investigational until a registered trial posts endpoints; ask what "useful vision" means to the patient, not just channel count.
- General readers: follow the national device-classification and pricing rules — when a visual prosthesis gets a billing code, that, not a press release, is the moment to take it seriously.