Brain-Computer Interface Race Heats Up: Musk Unveils 'Three-Year Whole-Brain Plan' as Chinese Firms Accelerate Catch-Up

Technical Competition Among Leading Global Brain-Computer Interface Companies Shows Signs of Acceleration

Earlier this month, Neuralink founder Elon Musk held a press conference to announce the company’s new three-year roadmap for its brain-computer interface technology, highlighting the goal of achieving a whole-brain interface by 2028 and gradually increasing the number of electrode channels from the current thousands to tens of thousands.

In contrast, in the Chinese market, as the national government has continuously emphasized research and practical application in frontier disciplines related to brain science in recent years, brain-computer interfaces (BCIs), as a key application scenario, are garnering increasing attention. Whereas domestic companies previously favored non-invasive approaches, a growing number of invasive BCI firms are now accelerating the process of clinical trial validation.

In a recent exclusive interview with a reporter from 21st Century Business Herald, Tao Hu, founder and chief scientist of NeuroXess, pointed out that significant changes in both the internal and external environments have occurred in China’s brain-computer interface (BCI) industry over the past two years, propelling its development to a new peak. From the perspective of technological iteration logic, the evolution of BCI resembles the progression of the autonomous driving industry from Level 1 (L1) to Level 5 (L5), representing a comprehensive competition of capabilities.

In his written response to a reporter from 21st Century Business Herald, Chen Tianqiao, founder, chairman, and CEO of Shanda Group and founder of the Tianqiao and Chrissy Chen Institute for Neuroscience, stated that cultivating world-class companies requires patient capital. He urged China’s sci-tech investors not to treat brain-computer interfaces merely as a fleeting trend for profit-making. Meanwhile, he expressed hope that more support would be extended to scientists like Tao Hu, who have burned their bridges to fully dedicate themselves to high-tech entrepreneurship.

Insights from Neuralink

According to Neuralink’s roadmap, the expansion of brain-computer interface (BCI) capabilities follows a progression from motor functions to language, vision, and ultimately whole-brain integration. The number of electrodes implanted in the brain is planned to increase from 3,000 in 2026 to over 25,000 by 2028, while exploring deep integration with artificial intelligence (AI) technologies.

Given that Neuralink currently has only seven trial participants, with the earliest brain-computer interface electrode implantation case dating back to January 2024, and considering that neural implant devices typically require a 5–10 year safety evaluation period, this appears to be an aggressive timeline.

In response, Tao Hu analyzed for a reporter from 21st Century Business Herald that Neuralink’s recently announced “whole-brain interface” does not imply literal whole-brain coverage. Rather, it refers to the simultaneous deployment of high-density electrode arrays across multiple brain regions—including those responsible for motor control, vision, hearing, and language—thereby enabling more complex functional coordination. The number of channels has jumped from the thousands to the tens of thousands, with the aim of achieving an exponential increase in input-output efficiency.

However, he emphasized that for the number of channels connecting electrodes in brain-computer interfaces, more attention should be paid to the effective channel count.

“The approach adopted by Neuralink involves using implanted electrode arrays deep within the cerebral cortex; in fact, the acquired EEG signals are often redundant and repetitive,” he analyzed. Rather than simply quantifying the number of electrode channels, it is preferable to conduct a systematic, comprehensive evaluation from perspectives such as electrode implantation and arrangement. This strategy aims to maximize the acquisition of effective information while minimizing damage to the brain, even with a relatively smaller number of electrode channels.

Furthermore, he believes that Musk’s mention of significantly shortening the electrode implantation time while achieving a higher level of safety represents an important development direction for the industry.

However, this raises a common question among the public: Why is it difficult for other companies to replicate Neuralink’s seemingly rapid increase in channel density?

Tao Hu pointed out to reporters from 21st Century Business Herald that, regarding the high throughput of brain-computer interfaces (BCIs), greater attention should be paid to the metric of “effective high throughput.” Since the core components of BCIs are fabricated using integrated circuit processes, the manufacturing of micrometer-scale electrode arrays is not constrained by advanced process nodes. Therefore, rapidly increasing the number of channels is technically feasible.

“The real challenge lies in systematic engineering capabilities,” he analyzed. Effective high-throughput performance is not a matter of showcasing isolated technical feats; rather, it requires long-term optimization across electrode design, algorithms, and signal matching. Increasing the number of channels triggers a cascade of interrelated issues, including power consumption, heat generation, wireless transmission, and biocompatibility. Neuralink has succeeded by making substantial trade-offs at the product implementation level.

Tao Hu cited Neuralink as an example, noting that it does not transmit all raw data in real time; instead, it reduces bandwidth and power consumption through methods such as local compression and feature extraction. “This involves rational trade-offs in product engineering, which is something we need to learn from.”

Although Chinese brain-computer interface (BCI) companies initially favored non-invasive approaches, they have become increasingly active in the invasive BCI sector over the past two years, with clinical trials even showing signs of acceleration.

NeuroXess is one of them. According to Qichacha, NeuroXess was established in 2021 and has undergone two rounds of public financing, with a cumulative amount reaching hundreds of millions of yuan. It is at a leading level domestically, with investors including Shanda Group, Sequoia China, Zhongping Capital, Lianxin Capital, and other institutions.

As an investor, Chen Tianqiao pointed out to reporters: “When I invested in Tao Hu, his company, BrainCo (Naohu Technology), was virtually the only invasive brain-computer interface (BCI) enterprise in China at that time. This field was also widely recognized internationally as a lonely, niche, and high-risk endeavor. Four years later, the situation has changed significantly. Brain-computer interfaces have garnered substantial global attention, with a growing number of startups emerging in this sector in China. The field has attracted investment from leading state-owned capital and private investors, resulting in increasingly large financing rounds and a series of high-impact achievements, which is truly encouraging.”

Brain-Computer Interface Dimensional Ascension

The shift in market sentiment over the past two years stems from significant fundamental improvements across technology, policy, and capital.

“More innovative and disruptive technologies have emerged in the field itself,” Tao Hu pointed out to a reporter from 21st Century Business Herald. For example, brain-computer interfaces (BCIs) have shifted from previously rigid designs to currently adopted flexible interfaces; the introduction of integrated circuit manufacturing processes has significantly increased the number of BCI channels; and continuous breakthroughs have also been made in core materials, algorithms, and other areas.

From a policy perspective, brain science is prominently featured in the 14th Five-Year Plan and the Outline of Long-Range Objectives Through the Year 2035, with brain-inspired computing and brain-computer integration technologies identified as key priorities.

“Capital from various sectors is paying increasing attention to the brain-computer interface (BCI) industry; as a result, leading BCI companies have frequently secured substantial financing over the past two to three years. Overall, the industry has reached a new peak in recent years across multiple dimensions, including clinical progress, technological and performance advancements, and capital and resource investment,” he summarized.

New external changes stem from the integration of AI technology applications. “Over the past two years, particularly with advancements in AI technology, we have significantly improved both the quantity and quality of electroencephalographic (EEG) signal acquisition from the brain. Meanwhile, these high-quality acquired data can be better analyzed and decoded. This facilitates the realization of thought-controlled synthesis of movement, language, and other functions,” he further analyzed.

Earlier this year, NeuroXess became the world’s first invasive brain–computer interface (BCI) company to achieve real-time decoding of the Chinese language. Reportedly, this marks China’s first clinical trial of real-time Chinese speech synthesis using a high-channel-count, implantable, flexible BCI. Two days after surgery, the patient began relevant training, and by seven days post-operation, achieved a 71% decoding accuracy across 142 common Chinese syllables, with a single-character decoding latency of under 100 milliseconds.

As an investor in BrainCo, Chen Tianqiao pointed out to reporters that the difficulty of language decoding is no less than any technological breakthrough achieved by Neuralink, and its value far exceeds obtaining approval for a single device. “The Tianqiao and Chrissy Chen Institute will assemble a world-class team focused on EEG large models, EEG decoding algorithms, and data, leveraging the latest artificial intelligence technologies to empower BrainCo. We encourage BrainCo to compete directly with Neuralink and boldly explore ultimate challenges such as visual reconstruction and memory downloading.”

Tao Hu pointed out to a reporter from 21st Century Business Herald that, overall, brain-computer interface (BCI) technology is a microsystem involving hardware and software fundamentals such as electrodes for acquiring electroencephalogram (EEG) signals, chips for signal processing, processors for neural encoding and decoding algorithms, and batteries powering the entire system. However, considering that the human brain is prone to shock in high-temperature environments, this entire system must be kept within limits of sufficiently small size, low power consumption, and controllable temperature rise.

“NeuroX Corporation has devoted significant effort over the past few years to integrating the entire supply chain. Currently, we are one of the few companies in China capable of achieving full independent control over all core components of the system,” he continued.

From a technical roadmap perspective, the early industry achieved real-time motor decoding, and currently, some companies have realized real-time speech decoding. The next step, as mentioned by Neuralink, will focus on the restoration of visual capabilities—thus gradually revealing a practical implementation roadmap for brain-computer interface technology.

Tao Hu pointed out to a reporter from 21st Century Business Herald that, building on the L1–L4 developmental stages for brain-computer interfaces proposed by Chen Tianqiao, NeuroXess has mapped each stage to the decoding difficulty and information density of different brain functional areas, in line with its own research focus.

Among these, L1 pertains to motor decoding. As early as 20 years ago, U.S. companies conducted application demonstration and validation. Due to the involvement of a single brain region, even brain-computer interfaces with single-digit channel counts can currently achieve functions such as mind-controlled wheelchair operation and robotic arm manipulation.

L2 is language decoding. Since this stage involves the coordination of multiple brain regions, high information density, and rapid transmission, leveraging brain-computer interfaces is expected to bypass the vocal organs and achieve rapid information exchange.

L3 will focus on visual decoding. The visual cortex, located in the posterior region of the human brain, is extensive in area and handles a large volume of information; both Neuralink and BrainCo have already made strategic investments in this domain.

At the L4 stage, the decoding of human emotions, memories, and cognition will be involved, entailing more complex brain functions and neural circuits in specific brain regions.

Tao Hu pointed out that this pathway does not follow the linear L1–L5 progression seen in the autonomous driving industry, but its evolutionary logic and technical difficulty are positively correlated with clinical needs.

Commercial Exploration

The application of brain-computer interface (BCI) technology will not be limited to the clinical dimension. One perspective holds that, as current human-computer interaction is predominantly based on touchscreens, BCIs have the potential to become the ultimate mode of human-computer interaction. This implies that there is considerable room for further expansion in the commercialization of this industry in the future.

Tao Hu also highly endorses this view. He analyzes that the future definition of brain-computer interfaces (BCIs) lies in how to rapidly output human intentions and thoughts while enabling the rapid input of external information, thereby accelerating information transmission.

“This requires consideration of dimensions such as the safety and efficiency of brain-computer interface (BCI) technology and the universality of the interface operating system. ‘Unlike practices in the medical device industry, we began laying out our ecosystem early on to enable connectivity between the brain and embodied intelligence, large language models, and smart home systems, thereby amplifying the utility of BCIs,’ he pointed out.”

According to reports, NeuroXess is fully embracing AI. On one hand, it leverages AI to advance brain-computer interface (BCI) technology; as more BCI data accumulates, the integration of large models enables rapid algorithmic iteration, benefiting users. On the other hand, it uses BCI to empower AI, not ruling out the realization of scenarios akin to those depicted in the movie Avatar, where the human brain directly interfaces with large models.

The establishment of a comprehensive, self-sufficient system is, to some extent, linked to Tao Hu’s personal background. In 2024, he resigned from his position as Deputy Director of the Shanghai Institute of Microsystem and Information Technology, Chinese Academy of Sciences, completely abandoning the “dual-track” model to go all-in on Neuralink Technology full-time. This transition not only prevented the dilution of resources between scientific research and entrepreneurship, but also enabled him to deeply focus his academic expertise on commercial implementation.

He told reporters that the greatest difference between scientific research and commerce lies in the establishment of “systems thinking”: scientific research emphasizes breakthroughs at specific points, whereas commerce demands system optimization, yield rates, cost control, and regulatory compliance; while scientific research allows for free exploration, commercial ventures require clearly defined indications and market demand. Chen Tianqiao commented on this matter, stating, “He (Tao Hu) is among the very few scientists who have made the firm decision to resign from prestigious research positions at the pinnacle of the ‘ivory tower’—positions they had dedicated half their lives to—thereby cutting off their retreat, to devote themselves full-time to entrepreneurship in the brain-computer interface field.”

Overall, Tao Hu pointed out to reporters that the development of brain-computer interfaces (BCIs) should focus on three key elements: high throughput, minimal invasiveness, and long-term in vivo stability. “The most critical among these is long-term in vivo stability,” he analyzed. While high throughput and minimal invasiveness can be validated in the short term, long-term in vivo stability requires accumulation over time, with long-term safety potentially serving as a “veto” factor.

From the perspective of capital markets, Chen Tianqiao pointed out to reporters that Neuralink, despite being founded eight years ago and only recently generating revenue, has not been hindered in becoming a globally renowned super unicorn. He stated, “If we continue to apply the investment practices common in the internet sector—requiring valuation adjustment mechanisms, immediate regulatory approvals, instant revenue generation, and rapid IPOs—such an investment approach will result in a lose-lose outcome for genuine science and technology innovation enterprises. We hope to demonstrate greater patience and sincerity, working together to help more ‘brain-computer interface’ startups grow into world-leading high-tech companies.”

(By Author: Luo Yiqi; Editor: Zhang Weixian)

Responsible Editor: Jiang Yuhan