What kind of business is a brain-machine interface, after all?
Author: Zan Zhu
Handmade Quantity | 100%
By 2026, brain-machine interfaces have already gained significant attention.
According to incomplete statistics, in just the first half of 2026, the financing in the domestic brain-machine interface sector exceeded 7 billion yuan, with more than 60 events occurring. The density of policies is equally noticeable: In August 2025, seven departments, including the Ministry of Industry and Information Technology, jointly issued the “Implementation Opinions on Promoting the Innovative Development of the Brain-Machine Interface Industry”; that same year, for the first time, the National Medical Insurance Administration separately priced brain-machine interface projects; and the “14th Five-Year Plan” outlines brain-machine interfaces alongside quantum technology, nuclear fusion, embodied intelligence, and 6G as one of the six major future industries.
Hot money, policies, and talents are all flowing in the same direction. But beneath this excitement, a more fundamental question has not been seriously addressed: What kind of business is a brain-machine interface, after all?
On October 30, at the Shanghai Pudong International Talent Port, a salon titled “After the Brain-Machine, Before the Future” provided a close observation opportunity. This event, co-hosted by Dao Tong Investment, the Shanghai Future Industry Fund, and the Shanghai Future Qidian Community, invited the head of the cutting-edge intersection at the Shanghai Municipal Science and Technology Commission, gathering a prominent selection of the most noteworthy startups in this sector—representing invasive and non-invasive, electrical and ultrasonic, medical and industrial safety; every mainstream technology route had a representative team present.
What’s more interesting is that if you look back at the financing history of these companies, a recurring name appears. This has led the market to question: why has an investment institution known for life sciences and medicine become one of the most active players in the hard technology field of brain-machine interfaces?
“The bottom of the barrel is medical devices, but the ceiling is not”
Before getting back to this question, one must discuss a topic: Is a brain-machine interface a medical device?
According to this logic, summing up the market space for various indications of neuroregulation results in just a few billion, insufficient to support today's valuation. However, in response to this query, Peng Lei, founder of the ultrasound brain-machine company Gestalt, has offered one perspective: “The bottom of this sector is medical devices, but the ceiling is not.”
Breaking down this statement: The so-called “bottom” is calculable—each indication corresponds to patient populations, willingness to pay, and alternative therapies, all of which can lead to a real market, determining the foundation on which a brain-machine company can survive. The so-called “ceiling” is the imaginative scope of brain-machine interfaces as the next generation of human-computer interaction infrastructure, as well as the potential for symbiosis with AI, determining whether this sector is worth heavy investment.
A frequently mentioned contrast is: Over the past decade, the world has invested trillions in AI to simulate human intelligence; yet, the investment in researching how the "only AGI in the universe"—the human brain itself—can be explained and enhanced has differed by several orders of magnitude. The disproportionate investment in these two matters has led more and more investors to rethink this in the past year.
Another judgment arises from endgame thinking: brain-machine interfaces represent the fusion of human-computer interaction. This is an “interface” technology—just like after the emergence of ChatGPT, if a company lacks its own large model, it will lose not just a product, but the position of an era. This is also why brain-machine interfaces have entered the strategic sequence of national future industries alongside quantum, nuclear fusion, and 6G.
In other words, medical devices are merely the first foothold for the commercialization of brain-machine interfaces, but not their boundary. They represent the accounts that can currently be settled, but also an entry point with yet unseen limits.
This also explains why understanding brain-machine interfaces requires crossing aptitude circles that naturally span both medical and hard technology domains.
A grid, a hard technology structure
Currently, the most popular classification of brain-machine interfaces is "invasive" and "non-invasive." This binary classification is the most widely disseminated and can easily lead to misunderstandings—it suggests a zero-sum battle between two opposing routes, as if they are competitors destined to determine a winner.
However, to practitioners, the true structure of this industry resembles a grid: the horizontal axis consists of four physical methods—sound, light, electricity, and magnetism—while the vertical axis includes three levels of intervention: invasive, semi-invasive, and non-invasive. Each cell in the grid could potentially house one or several companies—the key lies in what indications are targeted and whether the provided value can surpass existing therapies.
Furthermore, the cells are not in a competitive relationship. Electrical solutions boast the highest temporal and spatial resolution, yet the implanted location limits brain regions, while many brain functions are completed via loops rather than single points; ultrasound cannot visualize individual neuron discharges, but it can see the overall brain blood flow changes resulting from neurovascular coupling... This means that acoustics and electricity, invasion and non-invasion are less about replacement and more about approaching the same "system" from different angles—what they aim to do is expand the market together rather than declare an individual victory.
The same applies upstream. Electrodes must be high-throughput, flexible, and long-term stable; chips must balance between channel count, power consumption, biocompatibility, and computational precision; algorithms must account for individual differences and signal drift. The brain-machine interface is likely one of the industries that align best with the “barrel principle”—the ultimate barrier is not a singular technology but rather the ability to continuously iterate the entire link and optimize the system.
This presents typical hard technology characteristics: interdisciplinary convergence, long engineering chains, extended verification periods, and barriers that cannot be easily replicated.
Because of this, the investment logic surrounding brain-machine interfaces cannot merely be “understanding one medical device” or “understanding one algorithm.” It requires investors to simultaneously grasp neuroscience, clinical medicine, materials, chips, AI, as well as the most realistic aspects of registration, clinical trials, and payment.
So, what kind of institutions happen to stand at the intersection of these capabilities?
Seven companies, two intersecting capability circles
If we pull the timeline back a bit, we can see that brain-machine interfaces came into Dao Tong's view much earlier than the current wave of interest: three to four years ago, they appeared at Dao Tong's annual mid-year strategy meetings and year-end summary meetings, but the internal judgment always remained "the timing is not right"; at the year-end review meeting in 2024, the team unanimously felt the market's rising heat; in the first half of 2025, Dao Tong conducted intensive industry research; and in the second half of the year, a batch of projects were invested in densely.
Now, looking at the deployment method. If we spread out the previously mentioned “technology × scenario” grid, we will find that the most noteworthy batch of new companies in the sector is each occupying different cells:
✦ Active Technology, an industrial transformation platform for brain-machine interfaces at the Lingang Laboratory, follows the invasive electrical route, self-developing the entire chain from electrodes, chips, systems, to algorithms; its “Tianshu” series 1024-channel system has already passed inspection and covers three directions: motion, vision, and emotion.
✦ Gestalt, a re-venture by Peng Lei, founder of Brain Tiger Technology, uses ultrasound for whole-brain scale reading and writing, with its first product targeting chronic pain demonstrating positive early clinical data, and assembling a team to develop a multimodal brain data foundational large model.
✦ Kong Shan Ci, focusing on mental and cognitive disorders, has established a matrix of neuroregulation products using multiple energy sources including magnetism, sound, and electricity, with its transcranial magnetic navigation system undergoing multi-center registration clinical trials at multiple top-tier hospitals and its transcranial focused ultrasound system already beginning overseas deliveries.
✦ Brain Language Technology, founded by a returning scientist who led global clinical trials algorithms for aphasia brain-machine interfaces, employs a “signal tokenization + large model” approach for neural language decoding, aiming to allow aphasia patients to “speak” again at a speed close to natural speech.
✦ Hua Nao Anquan, responsible for the key laboratory of brain-machine interfaces under the Ministry of Emergency Management, applies brain-machine interfaces in industrial safety scenarios such as mining, metallurgy, petrochemicals, ports, and power, accumulating around hundreds of thousands of hours of real industrial EEG data and achieving global firsts in coal mine and mining safety approvals for brain-machine interfaces.
✦ Xinyun Medical, entering from chronic pain, has created the country’s first AI + ECAP closed-loop spinal electrical stimulation system.
✦ Maikewei, relying on Shanghai Jiao Tong University and several top clinical centers, follows the non-invasive route, transitioning from hospital treatment to home scenarios.
Seven companies, seven routes, which hardly conflict with each other, but a common point is that they all have Dao Tong Investment behind them.
If we combine the previous two points—brain-machine interfaces necessitating a crossing of medical and hard technology capability circles, and a medical background investment institution becoming one of the most active players in this sector—then the question becomes: are these two matters coincidental or inevitable?
A more reasonable explanation is that brain-machine interfaces just happen to be at the intersection of these two types of capabilities. Looking forward, electrodes, chips, materials, and algorithms are purely hard technology issues; looking back, indication selection, clinical validation, registration and approval, pricing, and payment are each classic medical issues. The absence of either side distorts investment judgments: understanding only hardware can lead to overestimation of engineering prototypes and underestimation of the true cycles of clinical and registration processes; understanding only medicine can result in a diminished view of brain-machine interfaces, mistaking them as merely a new category of neuroregulation devices.
The value of this “dual capability” becomes clearer when looking at exit paths. The IPO of brain-machine companies is not a standardized action: invasive companies needing three types of certifications must weigh the differences in timing and requirements between the fifth set of standards of the Science and Technology Innovation Board and Hong Kong Stock Exchange’s 18A; non-regulatory companies may find 18C or “first H then A” to be a more realistic path; upstream supportive enterprises are more suitable to wait for revenue and profit to materialize before going public on A-shares. Making such judgments requires knowledge of both the growth rhythms of hard technology companies and understanding of the regulatory logic of medical devices—missing either side can lead to a wrong timeline.
In reality, Daotong's earlier layout at this intersection is not its first practice. Before the AI4S concept became popular, the organization had already completed systematic layout. From AI4S to brain-machine interfaces, what you can see is not just a medical investment institution "crossing boundaries," but that it has consistently guarded the intersection of life sciences and real industry needs, watching technology repeatedly push the boundaries of this intersection outward.
Therefore, rather than saying that Dao Tong has crossed into hard technology, it is more accurate to say that brain-machine interfaces have torn down the wall between “medical” and “hard technology”.
Beneath the fervor, there are several unavoidable barriers
Of course, portraying brain-machine interfaces as a story of only rising trends would be irresponsible. Any industry at the brink of commercialization will face several unavoidable barriers, and brain-machine interfaces are no exception.
The first barrier is the payment barrier. The pricing of brain-machine interfaces has not yet been established across most provinces, and under the DRG payment reform, the motivation for hospitals to procure brain-machine devices is not strong. However, this barrier is loosening—the National Medical Insurance Administration has already set separate pricing for brain-machine interfaces, and once the pricing pathway is established, hospitals can reevaluate their accounts.
The second barrier is the approval barrier. There are no mature evaluation precedents for brain-machine interfaces; from model inspections to clinical trials to registration, each step requires taking the "first bite of the crab." Yet precedents are being created: the first registered certificate in the country has already been issued, and the review paths for subsequent products will become much clearer as a result.
The third barrier is the engineering barrier. The biocompatibility of electrodes, long-term stability of chips, reliability of packaging, energy transfer and power consumption—all of these have not truly been resolved. These are classic "iterative problems"—there are no shortcuts, but each issue can be methodically resolved given time and engineering accumulation.
The fourth barrier is the capital structure barrier. Money in the primary market is increasingly concentrating on leading and invasive projects, while many non-invasive projects remain stuck in early rounds. This indicates an imbalanced distribution of industry resources in the short term, also suggesting there are undervalued positions within it.
These barriers act as signposts rather than roadblocks. The cycle of medical devices will not be compressed by capital; the evolutionary path of brain-machine interfaces is likely to be serious medical care, home healthcare, and ultimately consumer electronics—this requires a period of patience after nurturing and growth. It does not need money chasing trends but capital that understands cycles and is willing to keep pace.
The biggest variable: AI? Embodiment?
As we reach the end, if one must find a candidate for the brain-machine interface's "iPhone moment," the answer most talked about in the industry is AI. However, discussing only AI may not be comprehensive enough.
First, let’s talk about AI. Two directions of fusion are happening simultaneously.
One direction is AI empowering brain-machine interfaces: Clinically, in the past, doctors needed an average of about 239 minutes to interpret a patient's epileptic state, but with AI pattern recognition, it now takes just 5 minutes, improving both efficiency and accuracy; invasive companies are training pre-trained models using thousands of hours of primate neural data, while some teams aggregate multimodal brain data, including ultrasound, MRI, and electrical signals, to develop a foundational large model for brain data. The other direction is brain-machine interfaces feeding back into AI: the human brain is a sparse, event-driven computing system, understanding its operations could save orders of magnitude in power consumption for AI calculations; the "error-related potential" in the brain—feedback signals that occur at the moment of unexpected events—might become a data source for training the next generation of AIs to align with human intentions and values.
On the data front, the industry is gradually forming several consensus points: first, leveraging each user’s long-term data might prove more effective than pursuing generalization through multi-task training; cross-population and cross-device transferability needs rigorous testing; high-quality state-signal paired data is more valuable than merely increasing sampling volume. As the disparities in hardware, such as electrodes and chips, gradually lessen with the maturity of the supply chain, the gap between algorithms and data is widening—this may indeed represent the true dividing line in the latter half of brain-machine interfaces.
Now, regarding “embodiment.” Today’s brain-machine interfaces essentially function as one-way streets: the brain sends intentions to machines, but what the machines do and encounter cannot be sent back to the person. However, actions in the real world almost entirely depend on such feedback—hands need to know the weight of the items being picked, and feet need to gauge the solidity of the ground being stepped on. Closing this feedback loop precisely embodies the utility of embodied intelligence: enabling machines to sense their environment, make decisions, execute actions, and then return results to humans. The brain-machine interface reads intentions, AI is responsible for understanding and planning, and embodied intelligence handles action and feedback; only when these three are integrated can a truly closed-loop human-computer system emerge.
This integration is not merely conceptual. In the rehabilitation field, non-invasive brain-machine interfaces combined with exoskeletons have already entered rehabilitation departments of numerous hospitals, helping stroke patients retrain their motor functions; brain-controlled wheelchairs and robotic arms are also beginning to leave the laboratory. According to industry assessments, human-computer integration will initially start from “thought remote control” and gradually evolve towards machines actively sensing human states, ultimately culminating in a long-term synergistic collaboration developed over time.
It is noteworthy that the next phase of brain-machine initiatives is seriously considering integration with AI and embodied intelligence. The ultimate story of brain-machine interfaces is likely to no longer be just “a smarter medical device,” but rather the next generation of human-computer interaction.
Returning to this salon, a notable detail is that this event originally intended to host an internal exchange for 50 people, but the sign-ups far exceeded expectations, necessitating multiple expansions. According to information from Zhongxin Network, the initial aim of organizing this event was to create “some chemical reactions” among teams with different technological paths and scenarios—some companies hold extensive EEG data from healthy populations, while others possess top intracranial signals, and some focus on developing brain-machine dedicated AFE chips, indicating ample space for complementarity between data, hardware, and algorithms.
This implies that beyond investment actions, another level of role is starting to emerge: when an organization stakes positions in multiple cells within the same sector, it has the opportunity to become a connector within the industry puzzle, rather than just a financial holder.
Fruit trees have flowering periods, and industries have their seasonal cycles
“Fruits never grow too high; we have merely entered the orchard.” This metaphor was used as the conclusion of the day's event.
The trees in the orchard do not bloom at the same time; the same goes for industries; they have their own seasonal cycles. The maturation of engineering and the commercialization transition have their inherent patterns. For investment institutions, the true issue posed by brain-machine interfaces is: when a sector embodies medical devices, hard technology, and future industries simultaneously, what posture should one adopt in this space?
The seven companies presented in that grid do not provide seven answers. They resemble more like seven questions: how exactly should humanity connect with the brain? Should it be through implanted electrodes or ultrasound? Through neuroregulation or algorithmic decoding? Or perhaps, by transforming brain signals into a brand-new form of industrial perception?
No one knows the ultimate answer. But the capital market has already placed its bets in advance. Dao Tong’s chosen approach is to discover those companies that might formulate the answers before they emerge and stand behind them.
免责声明：本文章仅代表作者个人观点，不代表本平台的立场和观点。本文章仅供信息分享，不构成对任何人的任何投资建议。用户与作者之间的任何争议，与本平台无关。如网页中刊载的文章或图片涉及侵权，请提供相关的权利证明和身份证明发送邮件到[email protected]，本平台相关工作人员将会进行核查。
