> ## Content Index
> Fetch the complete content index at: https://nexi.fund/llms.txt
> Use this file to discover other available public pages before exploring further.

# NeuroXess Is Filing to Go Public Before It Has an Approved Brain Implant
- URL: https://nexi.fund/neuroxess-simlife-brain-model-2026/
- Published: 2026-11-20T14:00:00.000Z
- Updated: 2026-11-20T14:00:00.000Z
- Description: A Shanghai brain-implant maker closed a RMB 700 million round and moved toward a listing while its device is still in a Class III trial. Its pitch is a 14,300-square-metre factory and 18,000 hours of neural data a year.
- Author: Nexi.fund Labs
- Tags: Biotech & Health, #mode-5, #hook-paradox, #track-F, #layout-chronicle-flow

A man paralysed from the chest down leaves his apartment, steers a wheelchair through a gate barrier, finds a shelf in a shop and pays for an item without moving his hands. The control signal comes from electrodes sitting on the surface of his brain. NeuroXess, the Shanghai company behind the implant, says the full trip was completed in a single attempt 84% of the time, end to end, at a decoding latency under 50 milliseconds.

That is a good demo. It is also, read closely, a business plan.

On 16 November 2026, NeuroXess published results from its SimLife testbed — an arena where one patient, identified only as Mr Song, carries out continuous tasks across travel, home, social and work scenarios. Two weeks earlier the company closed a RMB 700 million Series B, roughly $98 million. And on 16 November Chinese outlets reported it had filed a prospectus. A neurotech firm is heading toward a public listing while its flagship device is still inside a Class III registration trial.

The judgment that matters more than any decoding score: NeuroXess is not being priced on the science. It is being priced on a regulatory fast lane, a factory floor and a dataset.

## Seven years from a lab bench to a load-bearing trial

The corporate arc is short and unusually linear for a brain-interface company.

```

CHRONICLE: NeuroXess
────────────────────────────
2021    ·  Founded in Shanghai by Dr. Tiger H. Tao
2025.07 ·  Nature reports 256-channel flexible BCI; first real-time Mandarin decoding, ~50 words/min, <100ms
2026.01 ·  Breaks ground on the "giga-factory" in Ganjiang, Jiangxi
2026.06 ·  Two tetraplegic participants play Chinese chess across 800km on neural signals alone
2026.07 ·  Class III GCP trial starts at Huashan Hospital; enters NMPA green channel
◉ NOW   ·  RMB 700M Series B (28 Oct) + SimLife data testbed + prospectus filing
🔥 NEXT ·  Giga-factory ramp toward "tens of thousands" of units a year

```

Sources: NeuroXess corporate releases, 2025–2026; Nature news, July 2025.

The July milestone is the one to watch. On 7 July 2026 the "Triple-F" system — fully implanted, fully wireless, fully functional — began a GCP registered trial at Huashan Hospital, led by Professor Ying Mao and spread across 15 Grade-A tertiary hospitals. It was the first subdural cortical-surface BCI in China to enter the registered clinical phase for Class III devices, the highest risk class. The data from those hospitals feeds directly into a registration filing with the National Medical Products Administration.

Ten days before that, the same device entered the NMPA's Special Review Procedure for Innovative Medical Devices — the so-called green channel. That is not a marketing label. It is an administrative shortcut reserved for products with core invention patents and demonstrable clinical value, and it can shave time off a review that otherwise runs for years.

## The implant is the easy part to explain

The device's engineering choices are sensible. The electrodes go under the dura but do not penetrate brain tissue, which trades some signal quality for safety. The battery and processing modules sit in the chest, away from the brain, so waste heat never warms the cortex. There are no cables through the skin; charging is wireless. The surgical method borrows the well-established deep-brain-stimulation routine and needs no bespoke robot.

Follow-up numbers, from the company's own NMPA release, are credible on their face: a 28-year-old Shanghai participant with C3 tetraplegia hit 96.1% decoding accuracy after more than six months; a 26-year-old in Nanchang with C4 injury reached 91.2% at four months. In June 2026 the two played chess against each other from 800 kilometres apart, one driving a robotic avatar, the other a physical board through an exoskeleton glove.

Impressive. Still a demo.

## The money is really about the factory

What the RMB 700 million buys tells you where the company thinks the risk sits. The round, announced on 28 October, was co-led by SDIC Unity Capital, Shenzhen Capital Group and Triwise Capital, and the proceeds are earmarked for three things: multi-centre registration trials, the giga-factory, and "BCI+AI" work. The factory is 14,300 square metres in Jiangxi's Ganjiang New District, targeting annual delivery in the tens of thousands of units. R&D stays in Shanghai; manufacturing moves inland.

$98M Series B, Oct 2026 

#### NeuroXess Series B

RMB 700 million, co-led by state-backed and healthcare funds, aimed at registration trials and a giga-factory rather than new science. · *NeuroXess, Oct 2026*

That is a strange priority for a company whose product is not yet approved. It becomes coherent once you read SimLife as a data programme. NeuroXess says the testbed can train 20 people at once and accumulate roughly 18,000 hours of real-world interaction data a year. That data is meant to train models that understand intent, environment and task state together — what the company calls a "real-world physical model".

As we wrote in November, the whole-brain interface is, at bottom, [a data problem](https://nexi.fund/precision-neuroscience-brain-atlas/). NeuroXess has arrived at the same conclusion from the opposite direction: instead of a hospital atlas, it wants a factory that ships implants and a testbed that harvests behaviour.

## Everyone is collecting the same thing now

The convergence is not local. Neuralink reported more than 50,000 hours of human neural recordings in October, including 9,000 hours from a single participant, and is using the unlabelled data to pretrain decoders; for some users recalibration fell from about ten minutes a day to ten minutes a week, and one hit 11.32 bits per second on cursor control. Precision Neuroscience published an atlas drawn from 70 patients across 12 centres, over 87,000 microelectrodes and 180-plus hours of synchronised recording, on the back of a $250 million Series D. Stanford's Digital Brain Project, launched on 2 November, plans ten imaging campaigns and five years to train foundation models of the human brain on 7T fMRI and wearable OPM-MEG.

Different instruments, one thesis: the durable asset is the recording, not the electrode.

The honest gap is that none of these models generalise across people yet. The best public data still leans on participant-specific training, and the connectome frontier is far smaller than the marketing implies. Complete synaptic-resolution connectomes exist only for tiny organisms — the male fruit fly's central nervous system, at 166,000 neurons and 125 million synapses, is the largest. A cubic millimetre of mouse cortex is about 0.2% of the brain. The same volume in a human is roughly 0.000001%.

## What the prospectus has to prove

So the risk is not that the decoding is fake. It is that the timeline is. A Class III trial across 15 hospitals, a green channel, a factory rated for tens of thousands of units and a public listing all have to stay in sequence. If the trial data holds, the green channel shortens the wait and the factory has something to build. If any link slips, the factory is a very expensive empty room and the dataset is a research asset with no product attached.

> We will use this momentum to accelerate multi-centre clinical trials and regulatory approval, enabling this safer, more precise, and highly accessible BCI solution to benefit patients as swiftly as possible.— Dr. Tiger H. Tao, founder, CEO and chief scientist, NeuroXess

The implant is the demo. The factory is the pitch. The approval, whenever it lands, will be the first moment all three have to agree.

## Sources

[ NeuroXess SimLife trial: continuous real-world tasks at 84% success The November results behind the 50ms latency, 10.5-to-2-hour training cut and 18,000 hours of annual data. VCBeat Health ](https://www.vcbeathealth.com/article/64810?ref=nexi.fund) 

Primary account of the SimLife testbed and its clinical numbers.

[ NeuroXess completes RMB 700 million Series B Co-leads, use of proceeds, the 14,300-square-metre giga-factory and the clinical pipeline. NeuroXess ](https://en.neuroxess.com/news/neuroxess-completes-rmb-700-million-series-b-financing-co-led-by-sdic-unity-capital-scgc-and-triwise-capital/?ref=nexi.fund) 

Company disclosure of the round and the industrial plan.

[ Triple-F BCI enters the NMPA special review procedure The July green-channel entry, the Huashan-led Class III trial and the 96.1% / 91.2% follow-ups. NeuroXess ](https://en.neuroxess.com/news/neuroxesss-triple-f-bci-system-enters-nmpa-special-review-procedure-for-innovative-medical-devices?ref=nexi.fund) 

The regulatory milestone that underwrites the whole commercial story.

[ BCI companies report milestones in data and durability Neuralink's 50,000 hours, Precision's 70-patient atlas and Paradromics' four-year recordings. Neurofounders ](https://www.neurofounders.co/articles/bci-companies-demonstrate-progress-at-scale?ref=nexi.fund) 

Shows the data-scale race is global, not a single-company quirk.

[ Stanford Digital Brain Project to measure and model the human brain Ten campaigns, 7T fMRI and OPM-MEG, and a five-year plan to train brain foundation models. Wu Tsai Neurosciences Institute ](https://neuroscience.stanford.edu/news/stanford-digital-brain-project-measure-and-model-human-brain?ref=nexi.fund) 

The academic version of the same data bet — useful as the honest benchmark.