Brain-computer interfaces are missing a "Lobster Pool"! Tsinghua University's Gao Xiaorong responds to a reporter from the 21st Century Business Herald: The large-scale brain electroencephalogram model foundation is most likely to be developed in China.

Covering reporter: Zhang Rui | Editing by: Wei Guanhong

Since this year began, warm government-policy winds have been blowing frequently for brain-computer interface policy. For the first time, the government work report has included “brain-computer interface.”

On March 13, the National Medical Products Administration approved the world’s first invasive brain-computer interface medical device—“Implantable Brain-Computer Interface Hand Motion Function Compensation System” by Borelink Medical Technology (Shanghai) Co., Ltd., for上市.

During the “2026 Zhongguancun Forum Annual Conference” held from March 25 to 29, at the “Brain-Computer Interface Innovation Development Forum,” Deputy Minister of Industry and Information Technology Ke Jixin clearly pointed out that brain-computer interfaces are at a critical stage of moving from technology R&D to large-scale application. More efforts are needed to pool strength and jointly推动 brain-computer interfaces to accelerate from the laboratory to real-world applications.

During the forum, reporters from 《每日经济新闻》 interviewed multiple experts and people from the industry around questions such as brain-computer interface commercialization, technical routes, and future prospects.

Policy empowerment: Commercialization accelerates, entering the “brain renovation” stage

“Brain-computer interface” being written into the government work report for the first time—does this mean commercialization is about to speed up?

To this, multiple interviewees said, “This is definitely the case.”

Gao Xiaorong, a long-term professor at Tsinghua University and one of the main founders of the discipline of neuroscience engineering and brain-computer interfaces, has led brain-computer interface research in China since 1998. He told 《每日经济新闻》 reporters that this means, “we’re entering the ‘brain renovation’ stage now— the whole world is becoming an era where people want to ‘renovate the brain.’”

Speaking about the上市 of the world’s first invasive brain-computer interface medical device, he believes it is “very meaningful.” “From the time the concept was proposed to now, it’s been 50 years. Finally, a product has landed.”

Jie Fu, CEO (Chief Executive Officer) of Shanghai Jinzhi Biotechnology Co., Ltd., told 《每日经济新闻》 reporters that only market demand can truly drive the industry’s development. Once the medical side’s application entry is opened, it will become the “low-lying area” that attracts all kinds of technology transformation and implementation. Resources, capital, and technology will naturally converge in that direction. If all parties cannot see a clear path to monetization in the long term, then all prior R&D work will lack an effective exit. “We can see that the state is already very actively and pragmatically pushing this forward.”

In May last year, Beijing Tiantan Hospital, affiliated with Capital Medical University, opened a specialized outpatient clinic for brain-computer interfaces. Zhao Jizong, an academician of the Chinese Academy of Sciences and a professor at Beijing Tiantan Hospital affiliated with Capital Medical University, told 《每日经济新闻》 reporters that, “At present, the brain-computer interface outpatient clinic is quite popular. Dr. Yang Yi—who sees patients there—is often unable to get off work on time.”

Zhao Jizong said that there are two purposes for setting up an outpatient clinic: one is to recruit patients for research; the other is to prepare for future promotion, which requires building a case database. At present, it mainly targets three groups: people with hemiplegia, paraplegia, and ALS (amyotrophic lateral sclerosis).

“Different from ordinary outpatient care, it’s necessary to evaluate many things, including family circumstances, income, the relationship between spouses, and so on,” Zhao Jizong said. “In the past, we didn’t focus on those. We thought that once patients came, it would be enough. In fact, the problems are very complex—it’s not just a matter of disease. Long-term paralysis often leads to family problems and social issues such as poverty caused by illness.”

He said that opening outpatient clinics means brain-computer interfaces have entered the view of ordinary patients, but whether they can be done is another matter. They are still in the clinical trial stage, supported by scientific research funding.

Ecosystem to be built: currently missing****brain-electroencephalogram AI model foundation

Currently, AI (artificial intelligence) development is booming. Zhao Jizong believes that in the process of developing brain-computer interfaces, AI technology is needed. Adding AI will help accelerate device updates and iteration, as well as post-implant training—for example, whether AI can be used to produce templates with stronger applicability that patients with different conditions can use.

In Gao Xiaorong’s view, what the brain-computer interface field lacks most at present is foundation-layer construction. An ecosystem similar to CUDA (a parallel computing platform and programming model developed by NVIDIA) has not yet been established. “It’s like building a ‘lobster pond’—once we build the ‘lobster pond,’ everyone can ‘raise lobsters.’”

Gao Xiaorong said what we need to do now is foundation-layer construction. But at present, no one is willing to do this kind of ‘dirty and tiring work,’ which requires processing massive amounts of data. “We started doing brain-computer interface competitions in 2010 and accumulated a huge amount of data. Now we are working on foundation models and basic computing power, and we will also cooperate with relevant institutions to invest resources in building infrastructure. Just like the development path of large models, someone needs to lay the foundation first.”

Gao Xiaorong said that building this “lobster pond” requires many tasks. “In simple terms, you need data, algorithms, and computing power—and you also need application scenarios. Only after you prepare all of these can you build the ‘lobster pond’—that is, the foundation of an EEG model.” Gao Xiaorong said he believes the foundation of an EEG model will very likely be born in China, because our work is relatively ahead of the curve. “Just like language models have foundation models, EEG also needs a foundation model.”

Route debate: “inclusive products** must be non-invasive****”**

Brain-computer interfaces roughly fall into two categories: one is invasive, which requires surgery to implant electrodes; the other is non-invasive, which collects signals through external devices such as head-worn setups.

In Zhao Jizong’s view, non-invasive approaches are the easiest to promote. Invasive signals have better quality, but they have higher technical requirements and are more difficult. Also, implants may cause problems over the long term, such as immune reactions, fiber encapsulation, and signal attenuation.

“Many companies in China are making external headgear, but most of what they do are applications like improving sleep and helping students concentrate. It might be better if the focus were truly on运动功能 rehabilitation. But the downside is that signal quality is not as good as with invasive approaches.” In Zhao Jizong’s view, “the simplest is the best.” Whether it’s semi-invasive or fully invasive, it requires opening the skull. It’s impossible for implantation to have zero side effects 100%—this also depends on the individual situation.

The global share of non-invasive versus invasive brain-computer interfaces is roughly 8:2. Is it because non-invasive is not as difficult?

To this, Jie Fu believes that it is not because the difficulty is small. The core of brain-computer interfaces is real-time “reading” and “writing” of signals. Currently, most companies are focused on multimodal data acquisition (reading) and neural modulation (writing), and these are often separated. As industry heat increases, these directions are often collectively referred to as the brain-computer interface track. “At present, around 80% of companies are still at the stage of signal acquisition or one-way writing. But to achieve truly closed-loop control and personalized adjustment capabilities with non-invasive brain-computer products, I believe the industry still needs to go through a fairly long period of development.”

Jie Fu frankly said she is more optimistic about non-invasive approaches, because chronic brain health issues are becoming a global “silent pandemic.” The real value of non-invasive brain-computer interfaces is not that the technology is flashy, but that it responds to a social reality: children are trapped by attention deficiency, sleep delay, and anxiety; middle-aged people are wrapped layer by layer by stress, insomnia, and comorbidities; and for the elderly, poor sleep leads to neurodegenerative diseases, and the social care burden is 1:2.5.

“From everyone’s life, chronic brain diseases are unavoidable. Most of these chronic brain problems are not suitable for resolution with invasive methods. In terms of cost-effectiveness and risk, the risk-reward ratio doesn’t really match. So non-invasive brain-computer interface solutions that target chronic brain diseases can definitely become that beam of light,” she said.

In Gao Xiaorong’s view, now there needs to be普惠 products, and you can’t say only billionaires can “renovate the brain” while others can’t “renovate.” The world’s first implanted product currently on the market is not yet a普惠 product. “普惠 must be non-invasive, something everyone can afford. Invasive is more expensive and more complex than non-invasive.”

Promising outlook: The “Fifteenth Five-Year Plan” stage may be rolled out nationwide, but still facing multiple tests

When asked to what extent brain-computer interface technology will develop during the “Fifteenth Five-Year Plan” period, Zhao Jizong said that in that stage it may be rolled out nationwide, but it must be carried out in qualified hospitals—not every organization can do it.

Zhao Jizong emphasized that brain-computer interface technology is still in the trial stage, and it is not a substitute for traditional treatment approaches. Instead, it mainly provides an additional rehabilitation path. Its promotion still needs to address many issues, such as shortages of specialized training personnel and standards formulation.

He gave an example: after device implantation, professional personnel are needed for training. There is a shortage of talent in this area. Currently, it’s mostly computer professionals helping decode and train patients, and the training time is long. Patients don’t just stay in the hospital for three or five days and then leave; they first need to learn how to operate a computer, and understand what different signals represent as instructions.

“Right now, our方案 is: after surgery, stay in the hospital for one month; after discharge, stay near the hospital for two more months; only then can they go home. Because it’s in the research stage, the number of patients is limited, basically with follow-ups—if there’s a problem, they can come back at any time to resolve it. If this is to be rolled out nationwide, who will do this work? That’s why we can only do it one case at a time. It’s not because we don’t have equipment—everything is there, and implantation itself is also simple. It’s because the training work after implantation can’t keep up,” he said.

Zhao Jizong explained to the 《每日经济新闻》 reporters that after extracting signals, it’s necessary to analyze which signal corresponds to which action. Many signals have no value, or they are not the signals from the primary hand. To extract effective signals, it is currently done by computer professionals—guiding patients to move the cursor and telling them it’s too high, too low, to the left, or to the right, so adjustments are needed. “The process of adjustment is the training process—training them how to move.”

The reporter learned that among patients who come to the outpatient clinic, the proportion of groups such as farmers and workers is relatively high. For these patients, rehabilitation training must first include learning how to use a computer.

Is computer-based training definitely necessary? In the future, could it be replaced with a phone? Gao Xiaorong said that there shouldn’t be too much difference between a computer and a phone, and in the future it will definitely become phone-based. “Our lab is already developing toward the direction of glasses.”

Besides that, funding is also very important. Zhao Jizong mentioned that last year, the U.S. said the average cost per patient was 5,000 USD. I think it sounded very promising—about 30,000 to 40,000 RMB, which Chinese patients can still accept. But this year, the U.S. mentioned it was close to 50k USD. At that price, it is still a high barrier for ordinary patients.

As for when large-scale commercial applications might appear, Jie Fu believes the key is still to return to brain health issues or the disease itself. Building a set of scientific dialogue logic between clinical and approval departments, and proving what advantages this technology has over existing treatment methods—for example, a certain therapy was effective for 50% of patients in the past; now it can be improved to 75%; and in the future, it may reach 90%. “I think the limiting step lies in the approval itself. The market has opened a crack, but it doesn’t mean you can just fly directly to the commercial endpoint. You still need to validate it step by step with solid clinical data.”

Cover image source: 《每日经济新闻》 media resources database