Wetware Computing: Why Startups Are Building AI Processors Out of Human Brain Tissue

We are currently trying to brute-force the path to Artificial General Intelligence by feeding entire nuclear reactors to metal boxes. It is loud, expensive, and thermodynamically unsustainable. But what if the ultimate AI hardware isn’t etched from silicon? What if it is grown in a petri dish?

Before you read: We recently covered how the massive power draw of modern AI is melting Nvidia’s flagship server racks and forcing hyperscalers to revive dormant nuclear power plants. This article looks at the radical biological alternative attempting to solve that energy crisis.

Table of contents

• The One-Millionth Power Bill
• How Do You Boot Up a Brain?
• The Bio-Degradation Problem
• Fun Facts
• Sources

The One-Millionth Power Bill

The human brain is an engineering marvel that we completely take for granted. It contains roughly 86 billion neurons, processes incredibly complex multimodal data in real-time, and runs on about 20 watts of power. That is less electricity than the dim bulb sitting inside your refrigerator.

Compare that to Nvidia’s NVL72 rack, which draws up to 120,000 watts just to run a large language model that still struggles to count the number of Rs in the word ‘strawberry’. The efficiency gap between biology and silicon isn’t just large; it is a chasm of about six orders of magnitude.

This is why the field of “organoid intelligence” (OI) is quietly gaining traction. Startups are no longer just theorizing about biological computing—they are actively building it. A Swiss startup called FinalSpark recently launched the world’s first “Neuroplatform,” a cloud computing system where developers can rent access to 16 human brain organoids. Yes, you read that correctly. You can SSH into a cluster of living human brain cells.

How Do You Boot Up a Brain?

An organoid is a three-dimensional cluster of lab-grown tissue. In this case, scientists take human stem cells and coax them into becoming neurons. These neurons naturally reach out to one another, forming complex synaptic networks.

To turn this biological soup into a computer, you place the organoid on a Microelectrode Array (MEA). This is a specialized silicon chip studded with microscopic gold pins that can both read the electrical signals firing from the neurons and send electrical pulses back into the tissue.

Training an organoid isn’t like compiling C++ or running backpropagation on a GPU. It is closer to training a microscopic, highly autistic dog. You give the tissue an electrical pulse representing a task. If the neurons fire in the desired pattern, you flood the dish with a chemical reward like dopamine. If they fail, you give them chaotic electrical noise—essentially biological negative reinforcement. Imagine getting a PagerDuty alert at 3 AM not because your AWS instance crashed, but because your server farm literally caught a virus and died.

The Bio-Degradation Problem

I honestly think wetware is the most fascinating hardware paradigm shift of our lifetime, but let’s stay grounded. It is not going to replace your RTX 4090 tomorrow.

Silicon’s greatest advantage is its stability. A CPU will happily crunch numbers for a decade if you keep the dust out of the fans. Biological tissue, on the other hand, is incredibly needy. It requires a constant flow of nutrient-rich microfluidics, precise temperature control, and a sterile environment.

Even under perfect laboratory conditions, the organoids on FinalSpark’s Neuroplatform currently only live for about 100 days. When the cells die, your “processor” is gone forever. You cannot simply reboot it. Furthermore, the results are currently non-deterministic. Two identical organoids given the exact same chemical rewards will not necessarily learn a task at the same speed or in the same way. Standardizing biological output is an absolute nightmare for software engineers who rely on predictable, deterministic execution.

But the economic incentives to figure this out are massive. If we can stabilize the lifespan of these organoids and scale the microfluidic infrastructure, wetware could completely shatter the AI data wall. We wouldn’t need to strip-mine the earth for copper or build new nuclear reactors; we would just need to get really good at plumbing.

Fun Facts

• Did you know that the neurons used in these organoids are often derived from human skin cells? Through a process called induced pluripotency, adult skin cells are reprogrammed back into an embryonic state before being coaxed into becoming brain tissue.
• Did you know the world’s fastest supercomputer, Frontier, consumes 21 megawatts of power to achieve exascale computing? A human brain achieves roughly the same computational complexity using the energy equivalent of half a digested sandwich.

Sources

• FinalSpark: Neuroplatform Architecture and Organoid Intelligence Documentation
• Frontiers in Science: Organoid intelligence (OI) - the new frontier in biocomputing and intelligence-in-a-dish
• Nature: Stem-cell-derived brain organoids and microelectrode arrays
• IEEE Spectrum: The Reality of Biological Computing and Wetware