Brain Cells on Chips: The Biocomputing Revolution

Living neurons grown on silicon chips can learn tasks like Pong with far less data and energy than traditional AI, offering a potential solution to computing limits and energy crises. Companies like Cortical Labs and FinalSpark are commercializing biocomputing, while researchers explore applications in drug testing and neurodegenerative disease modeling—though significant engineering, ethical, and scalability challenges remain.

The Biocomputing Breakthrough

Neurons Learn Pong

In 2022, Cortical Labs grew 800,000 neurons on a silicon chip (about the size of a bumblebee's brain) and trained them to play Pong by sending electrical impulses. When neurons missed the ball, they received chaotic stimuli; when they hit it, they received predictable stimuli. Within minutes, the neurons started playing more consistently, suggesting they naturally minimize surprise and unpredictability.

Why Biology Beats Silicon for Learning

Biological brains require far less training data and energy to understand real-world environments compared to traditional machine learning. This efficiency comes from evolutionary pressure—organisms that couldn't quickly learn and adapt died out. The promise of reduced data and energy consumption is considered the holy grail for AI advancement.

The Computing Crisis

Moore's Law Hits a Wall

For over 50 years, the semiconductor industry has doubled transistor density every two years (Moore's Law). Today, transistors are just a few nanometers apart, and some chip layers are one atom thick. Since you cannot go thinner than one atom, the physical limits of silicon-based scaling are being reached, forcing the industry to seek alternative solutions.

Energy Demand Explosion

Data centers are consuming massive amounts of power. By 2034, global data centers are expected to use roughly 1,580 terawatt hours yearly—equivalent to all of India's electricity consumption. A supercomputer can draw up to 40 megawatts, while the human brain runs on just 20 watts, making biological systems vastly more efficient.

AI Investment Surge

In 2025, private sector companies invested more than $500 billion in AI infrastructure alone, reflecting intense competition for computational advantage. The DeepSeek release in January 2025 reminded markets of the economic stakes and geopolitical tensions surrounding AI development.

Brain Architecture vs. Silicon

Neurons Outperform Chips in Complexity

The human brain contains approximately 86 billion neurons forming over 100 trillion connections. Unlike silicon chips, which process information sequentially (one thing after another), the brain is a lattice structure where information flows in multiple directions simultaneously. Additionally, in biological systems, the distinction between hardware and software blurs—cells can change their structure and function dynamically.

Commercialization and Startups

Cortical Labs: From Research to Product

Cortical Labs, based in Melbourne, Australia, raised approximately $10 million in 2023 from investors including In-Q-Tel and Horizon Ventures. Their first commercial product, the CL1 unit, is a processor containing human brain cells (neurons on chip, organoids, or bio-engineered intelligence). Each CL1 costs around $35,000 and was expected to launch in March 2025. The unit acts as a life-support system, feeding cells, disposing of waste, and maintaining a 37-degree-Celsius environment.

FinalSpark: Cloud-Based Biocomputing

FinalSpark, a Swiss biotech startup with under 10 employees, has bootstrapped with just over $1.5 million in their own funds (no outside investment). They created the Neuroplatform, a cloud computing network offering remote access to 16 brain organoids housed in incubators in Vevey, Switzerland. Researchers worldwide can observe real-time neural activity 24/7, send stimulation to electrodes, and use the organoids for robotics research or university teaching. The platform now has 200 registered users.

Engineering and Scalability Challenges

The Hybrid Environment Problem

Scaling biocomputing faces immense engineering obstacles. Biological cells require constant feeding, waste removal, and precise temperature control (37 degrees Celsius). Computer chips generate hundreds of watts and require cooling systems that produce extreme heat. Combining biological material with this thermal environment would literally cook the cells. Any hybrid technology must solve these fundamental incompatibilities.

Manufacturing Reliability

Moving from lab prototypes to mass production is a major hurdle. Building one prototype differs vastly from having hundreds roll off a production line reliably and stably. Investors in biocomputing are not expecting quick profits; they recognize this is a long-term, high-risk venture comparable to the semiconductor industry's 50-year development trajectory.

Medical and Research Applications

Brain Organoids for Drug Development

Thomas Hartung's lab uses standardized brain organoids (mini brains) to replace animal testing in toxicology and drug development. Animal studies often fail to predict how toxins or drugs affect humans. Brain organoids offer a more human-relevant model. For pharmaceutical companies, being one day earlier to market is worth approximately $1 million, making faster, more accurate testing highly valuable.

Modeling Neurodegenerative Disease

Researchers are using brain organoids to model diseases like Parkinson's and Alzheimer's. One approach involves training organoids to learn tasks, then observing whether disease-context organoids forget faster or differently than healthy controls. This could reveal disease mechanisms and accelerate therapeutic discovery for conditions where current treatments only manage symptoms.

Neurological Disease Burden

More than one-third of the global population will suffer from neurological conditions like Parkinson's, Alzheimer's, or dementia at some point in their lives. By 2050, brain disorders are projected to increase by 22% due to population growth and aging. Parkinson's alone costs the U.S. over $50 billion annually (direct and indirect costs), with about 10 million patients worldwide and 90,000 new diagnoses yearly in the U.S.

Ethical and Philosophical Questions

Consciousness and Moral Status

A central ethical concern is whether brain organoids could become self-aware or conscious. Researchers currently believe tiny balls of cells are too simple to develop consciousness, but as organoids grow more complex, questions arise: Could they suffer? Experience pain? If they become self-conscious, is it ethical to stop feeding them or kill them? These questions parallel broader biomedical ethics around stem cell use and donor consent.

Informed Consent and Future Scenarios

Stem cell donors often remain alive when their cells are used to create brain organoids. Donors may not have anticipated their cells being used to create thinking tissue. As organoids become more sophisticated, the validity of original consent becomes questionable. Researchers must grapple with whether donors truly understood the implications of their contribution.

The Long-Term Outlook

A Hybrid Computing Future

The future likely involves a world where some computing is biological and some is artificial. This hybrid approach could address both energy efficiency and processing power limitations. However, replicating the semiconductor industry's 50-year development trajectory with a new medium will take considerable time, investment, and scientific effort. Many innovations take years in the lab before becoming commercially viable.

Notable quotes

My life has been focused on how can you elicit intelligence from brain cells in a dish? — Brett Kagan, Chief Scientific Officer, Cortical Labs
Anything with biology can learn to navigate its environment incredibly quickly, incredibly efficiently. — Brett Kagan
In the future, we'll have to live in a world where there will be some parts in computing which are living, and other which are just artificial. — FinalSpark representative
Bloomberg Originals
24 min video
3 min read
Brain Cells on Chips: The Biocomputing Revolution
You just saved 21 min.
The big takeaway
Living neurons grown on silicon chips can learn tasks like Pong with far less data and energy than traditional AI, offering a potential solution to computing limits and energy crises. Companies like Cortical Labs and FinalSpark are commercializing biocomputing, while researchers explore applications in drug testing and neurodegenerative disease modeling—though significant engineering, ethical, and scalability challenges remain.
The Biocomputing Breakthrough
Neurons Learn Pong
In 2022, Cortical Labs grew 800,000 neurons on a silicon chip (about the size of a bumblebee's brain) and trained them to play Pong by sending electrical impulses. When neurons missed the ball, they received chaotic stimuli; when they hit it, they received predictable stimuli. Within minutes, the neurons started playing more consistently, suggesting they naturally minimize surprise and unpredictability.
1
Grow 800,000 neurons on silicon chip
2
Divide chip into sensory and motor sections
3
Send chaotic stimuli when neurons miss ball
4
Send predictable stimuli when neurons hit ball
5
Neurons learn to play Pong within minutes
How DishBrain learned to play Pong
Why Biology Beats Silicon for Learning
Biological brains require far less training data and energy to understand real-world environments compared to traditional machine learning. This efficiency comes from evolutionary pressure—organisms that couldn't quickly learn and adapt died out. The promise of reduced data and energy consumption is considered the holy grail for AI advancement.
The Computing Crisis
Moore's Law Hits a Wall
For over 50 years, the semiconductor industry has doubled transistor density every two years (Moore's Law). Today, transistors are just a few nanometers apart, and some chip layers are one atom thick. Since you cannot go thinner than one atom, the physical limits of silicon-based scaling are being reached, forcing the industry to seek alternative solutions.
50+ years
Duration of Moore's Law doubling
Transistor scaling is hitting physical limits
Energy Demand Explosion
Data centers are consuming massive amounts of power. By 2034, global data centers are expected to use roughly 1,580 terawatt hours yearly—equivalent to all of India's electricity consumption. A supercomputer can draw up to 40 megawatts, while the human brain runs on just 20 watts, making biological systems vastly more efficient.
Supercomputer
40 megawatts
Human brain
0.00002 megawatts (20 watts)
Power consumption: supercomputer vs. human brain
AI Investment Surge
In 2025, private sector companies invested more than $500 billion in AI infrastructure alone, reflecting intense competition for computational advantage. The DeepSeek release in January 2025 reminded markets of the economic stakes and geopolitical tensions surrounding AI development.
$500 billion
AI infrastructure investment in 2025
Private sector AI spending in 2025
Brain Architecture vs. Silicon
Neurons Outperform Chips in Complexity
The human brain contains approximately 86 billion neurons forming over 100 trillion connections. Unlike silicon chips, which process information sequentially (one thing after another), the brain is a lattice structure where information flows in multiple directions simultaneously. Additionally, in biological systems, the distinction between hardware and software blurs—cells can change their structure and function dynamically.
Neurons
86 billion
Connections
100 trillion
Scale of the human brain
Commercialization and Startups
Cortical Labs: From Research to Product
Cortical Labs, based in Melbourne, Australia, raised approximately $10 million in 2023 from investors including In-Q-Tel and Horizon Ventures. Their first commercial product, the CL1 unit, is a processor containing human brain cells (neurons on chip, organoids, or bio-engineered intelligence). Each CL1 costs around $35,000 and was expected to launch in March 2025. The unit acts as a life-support system, feeding cells, disposing of waste, and maintaining a 37-degree-Celsius environment.
Cortical Labs funding (2023)
10 million dollars
CL1 unit price
35 thousand dollars
Cortical Labs investment and product pricing
FinalSpark: Cloud-Based Biocomputing
FinalSpark, a Swiss biotech startup with under 10 employees, has bootstrapped with just over $1.5 million in their own funds (no outside investment). They created the Neuroplatform, a cloud computing network offering remote access to 16 brain organoids housed in incubators in Vevey, Switzerland. Researchers worldwide can observe real-time neural activity 24/7, send stimulation to electrodes, and use the organoids for robotics research or university teaching. The platform now has 200 registered users.
FinalSpark self-funded capital
1.5 million dollars
Brain organoids available
16 units
Neuroplatform users
200 researchers
FinalSpark scale and adoption
Engineering and Scalability Challenges
The Hybrid Environment Problem
Scaling biocomputing faces immense engineering obstacles. Biological cells require constant feeding, waste removal, and precise temperature control (37 degrees Celsius). Computer chips generate hundreds of watts and require cooling systems that produce extreme heat. Combining biological material with this thermal environment would literally cook the cells. Any hybrid technology must solve these fundamental incompatibilities.
Manufacturing Reliability
Moving from lab prototypes to mass production is a major hurdle. Building one prototype differs vastly from having hundreds roll off a production line reliably and stably. Investors in biocomputing are not expecting quick profits; they recognize this is a long-term, high-risk venture comparable to the semiconductor industry's 50-year development trajectory.
Medical and Research Applications
Brain Organoids for Drug Development
Thomas Hartung's lab uses standardized brain organoids (mini brains) to replace animal testing in toxicology and drug development. Animal studies often fail to predict how toxins or drugs affect humans. Brain organoids offer a more human-relevant model. For pharmaceutical companies, being one day earlier to market is worth approximately $1 million, making faster, more accurate testing highly valuable.
$1 million
Value of one-day-earlier market entry for pharma
Economic incentive for faster drug testing
Modeling Neurodegenerative Disease
Researchers are using brain organoids to model diseases like Parkinson's and Alzheimer's. One approach involves training organoids to learn tasks, then observing whether disease-context organoids forget faster or differently than healthy controls. This could reveal disease mechanisms and accelerate therapeutic discovery for conditions where current treatments only manage symptoms.
Neurological Disease Burden
More than one-third of the global population will suffer from neurological conditions like Parkinson's, Alzheimer's, or dementia at some point in their lives. By 2050, brain disorders are projected to increase by 22% due to population growth and aging. Parkinson's alone costs the U.S. over $50 billion annually (direct and indirect costs), with about 10 million patients worldwide and 90,000 new diagnoses yearly in the U.S.
Ethical and Philosophical Questions
Consciousness and Moral Status
A central ethical concern is whether brain organoids could become self-aware or conscious. Researchers currently believe tiny balls of cells are too simple to develop consciousness, but as organoids grow more complex, questions arise: Could they suffer? Experience pain? If they become self-conscious, is it ethical to stop feeding them or kill them? These questions parallel broader biomedical ethics around stem cell use and donor consent.
Informed Consent and Future Scenarios
Stem cell donors often remain alive when their cells are used to create brain organoids. Donors may not have anticipated their cells being used to create thinking tissue. As organoids become more sophisticated, the validity of original consent becomes questionable. Researchers must grapple with whether donors truly understood the implications of their contribution.
The Long-Term Outlook
A Hybrid Computing Future
The future likely involves a world where some computing is biological and some is artificial. This hybrid approach could address both energy efficiency and processing power limitations. However, replicating the semiconductor industry's 50-year development trajectory with a new medium will take considerable time, investment, and scientific effort. Many innovations take years in the lab before becoming commercially viable.
Worth quoting
"My life has been focused on how can you elicit intelligence from brain cells in a dish?"
— Brett Kagan, Chief Scientific Officer, Cortical Labs, at [1:06]
"Anything with biology can learn to navigate its environment incredibly quickly, incredibly efficiently."
— Brett Kagan, at [2:47]
"In the future, we'll have to live in a world where there will be some parts in computing which are living, and other which are just artificial."
— FinalSpark representative, at [14:16]
Made with Glimpse by Wozart
glimpse.wozart.com/v/gg7uenfy
Share this infographic

More like this