I need you to sit with this for a second.

There is a machine in Melbourne, Australia. Inside it, 800,000 human brain cells — real, living neurons grown from human skin and blood samples — sit on a silicon chip. They fire. They listen. They learn.

Nobody programmed them.

In 2022, they taught themselves to play Pong. In 2026, they taught themselves to play DOOM.

And the CIA's venture capital arm just invested.

This isn't a headline from 2045. This is happening right now, and almost nobody is talking about it.

What Cortical Labs Actually Built

The company is called Cortical Labs. Their product is the CL1 — the world's first commercially available biological computer.

Here's how it works: human induced pluripotent stem cells (iPSCs), reprogrammed from actual adult donors' skin or blood, are cultivated into neurons and grown directly onto a silicon chip. A life-support system keeps them alive — temperature control, nutrient supply, waste filtration, fluid circulation. The neurons survive for up to six months.

Fifty-nine electrodes interface with the neural tissue, reading and writing electrical impulses in sub-millisecond feedback loops. On top of this sits biOS — their Biological Intelligence Operating System — which creates a simulated environment for the neurons to inhabit and respond to.

The neurons aren't following instructions. They're learning. Adapting. Self-organizing. In real time.

And they do all of this on roughly 30 watts of power.

Your desk lamp uses more.

The Numbers That Should Stop You

Let's put this in context.

Training GPT-4 consumed an estimated 50 gigawatt-hours of electricity. A single NVIDIA H100 GPU pulls 700 watts. Microsoft is literally reopening Three Mile Island — yes, the nuclear plant — to power its AI data centers. The industry is building toward a wall, and everyone knows it.

A rack of 30 CL1 units? 850 to 1,000 watts total. That's not a typo. Thirty biological computers running on less power than a single high-end GPU.

Each unit costs $35,000. Or $20,000 when purchased in server racks. Or $300 per week for cloud access — what Cortical calls "Wetware-as-a-Service."

The company has raised $11 million. Their investors include Hong Kong's Horizons Ventures, Australia's Blackbird Ventures, and — here's the part that should make you pay attention — In-Q-Tel, the venture capital arm of the U.S. Central Intelligence Agency.

In-Q-Tel doesn't invest in science experiments. They invest in technologies they believe will shape national security. Cortical Labs is one of only a few dozen life science startups in their entire portfolio.

From Pong to DOOM: What the Neurons Actually Did

In 2022, Cortical Labs published a landmark paper in the journal Neuron — one of the most prestigious neuroscience journals in the world. They placed 800,000 human and mouse neurons on a chip and created a system called DishBrain.

The neurons were placed in a simulated game environment. When the paddle hit the ball, they received a predictable electrical signal. When they missed, they received random noise. Within minutes, the neurons reorganized their connections to maximize predictable outcomes — and learned to play Pong.

Nobody wrote a line of code that told them how. They figured it out. The same way your brain figures things out: through feedback and adaptation.

But here's the detail that most coverage misses: in testing, these biological neural networks outperformed deep reinforcement learning algorithms in sample efficiency. They learned faster, from less data, than the AI systems we spend billions of dollars training.

Karl Friston — arguably the world's most influential living neuroscientist and the creator of the free energy principle — called the CL1 a "remarkable achievement" and the "dénouement of years of theoretical and biophysical innovation."

His exact words: "Experimentalists now have at hand a little 'brain in a vat,' something philosophers have been dreaming about for decades."

By early 2026, the neurons were playing DOOM — a vastly more complex environment requiring spatial reasoning, timing, threat assessment, and multi-variable decision-making. The gap between Pong and DOOM isn't incremental improvement. It's a category shift.

Why This Actually Matters: Beyond the Headlines

Here's what the clickbait articles won't tell you.

The CL1 isn't primarily a "cool demo" product. Its most significant near-term application is in drug discovery — specifically neuropsychiatric drugs, which have some of the highest failure rates in clinical trials.

The reason those drugs fail? Existing preclinical models can't capture how brain cells actually process information. They test the chemistry, but not the computation. As Cortical's Chief Scientific Officer Brett Kagan explains: "You actually need a device like ours before you can test this model at all."

In a recently accepted paper, the CL1 demonstrated something remarkable: epileptic neurons — which can't learn to play games effectively — regained function when treated with antiepileptic drugs. The biological computer didn't just run a simulation of epilepsy. It modeled the actual disease, at the cellular level, and demonstrated treatment response in real time.

That's not a gaming demo. That's the future of pharmaceutical development.

The Scaling Economics of Biology

Here's the part that should terrify every chip manufacturer.

"While it cost us quite a bit to make 100,000 neurons," Kagan told IEEE Spectrum, "it only costs a fraction more to make a million and not much more for 100 million, because biology grows exponentially."

Read that again. The cost curve of silicon follows Moore's Law — expensive, predictable, plateauing. The cost curve of biology follows cell division — exponential, cheap, and barely scratching the surface.

A transistor consumes roughly a million times more energy per operation than a neuron. Biology has had four billion years of optimization. We've been building chips for sixty.

Cortical Labs is now building two data centers. Not powered by nuclear reactors or solar farms. Powered by neurons. The first 115 CL1 units ship this summer.

The Bottom Line

We are witnessing the birth of a new computing paradigm.

Not "new" like a faster chip or a better algorithm. New like the jump from vacuum tubes to transistors. New like the jump from analog to digital. Biological computing doesn't compete with silicon on silicon's terms. It operates in a category that silicon can't access: real, adaptive, self-organizing intelligence with a million-to-one energy advantage.

The CIA is investing. The world's top neuroscientist is endorsing it. The neurons are playing DOOM and curing epilepsy models at the same time. Two data centers are being built.

And the company has 22 employees.

If you're in AI, biotech, energy, defense, or pharmaceutical development — this is the signal. Not a trend piece. Not a prediction. A signal that's already transmitting.

Silicon got us here.

Biology takes us further.

---

Sources: IEEE Spectrum, New Atlas, Tom's Hardware, Gizmodo, New Scientist, Cortical Labs (corticallabs.com), Neuron journal (2022). Cortical Labs is based in Melbourne, Australia. The CL1 launched March 2, 2025 in Barcelona.