One of artificial intelligence's most uncomfortable problems is not intelligence at all. It is energy.
The systems we are building require increasingly large amounts of electricity to train, operate and scale. The International Energy Agency expects global data-centre electricity consumption to more than double by 2030, with AI among the main forces behind that growth. For all the sophistication of modern AI, there remains an awkward comparison sitting inside our own heads: the human brain accomplishes an extraordinary amount of computation using roughly the power of a small household light bulb.
Biological computing asks whether some of that efficiency can be borrowed rather than merely imitated.
This is no longer entirely theoretical. Cortical Labs has developed systems in which living neurons are grown directly over electronic interfaces and incorporated into closed-loop computing environments. Its original DishBrain research, published in Neuron, demonstrated this approach using neural cultures embodied in a simulated Pong environment.
Its more recent CL1 platform takes that idea further, combining an in-vitro neural culture, multi-electrode interface and life-support system into a device designed specifically for real-time interaction with biological neural networks.
FinalSpark is experimenting with neural organoids that researchers can access remotely for wetware-computing research. Its published Neuroplatform research describes an infrastructure capable of continuously recording neural activity, electrically stimulating organoids and allowing researchers to run experiments remotely through software.
Neither technology is remotely capable of replacing today's AI infrastructure, and claims about dramatic energy savings still need to be demonstrated at meaningful computational scale. But the direction is real: instead of continually building larger machines that imitate biological neural networks, researchers are beginning to explore computation using biology itself.
That is where I think the conversation becomes larger than energy.
If biological or hybrid computing eventually becomes practical, we should not assume its ultimate purpose will be to make today's data centres cheaper. Technology rarely stops at solving the problem that justified its invention.
Consider a distant but increasingly imaginable progression. Today, human cells can be reprogrammed into induced pluripotent stem cells and subsequently differentiated into other cell types, including neurons. A cheek swab would therefore not literally power a computer, but cells obtained non-invasively could conceivably become the biological starting material for a personalised neural system.
Now imagine that idea several technological generations from today.
You buy a personal AI device whose biological processor was grown from your own cells. It learns continuously, consumes little energy compared with conventional computational infrastructure, and becomes increasingly adapted to you. Not merely another cloud assistant serving millions of people, but persistent computational intelligence that exists specifically alongside one person.
Then connect that trajectory with another field developing independently: brain–computer interfaces.
Present-day BCIs are primarily restorative technologies. Researchers have already demonstrated systems that allow people with severe paralysis to communicate and control computers through neural activity, including systems capable of sustained independent use outside the laboratory. We are still a long way from installing intelligence, increasing someone's concentration on command or creating a cognitively enhanced human. But once biological computing, AI and neural interfaces are placed on the same long technological horizon, the destination becomes difficult to ignore.
AI may eventually stop being something we use and become part of the cognitive environment through which we think.
That possibility changes the energy question.
If computation remains enormously expensive, powerful intelligence remains concentrated in data centres, corporations and governments. If computation becomes radically more efficient, intelligence can become distributed. If it becomes small enough and efficient enough, it becomes personal. If neural interfaces mature alongside it, personal intelligence could eventually become cognitively integrated.
At that point, we are no longer discussing better computers. We are discussing the possibility of changing the effective capabilities of a human being.
A person might retain an artificial memory larger than anything biology could naturally provide. Knowledge could be retrieved almost as quickly as it is recalled. Translation might become effectively instantaneous. Complex calculations could occur alongside ordinary thought. An AI could continuously challenge assumptions, model consequences, identify forgotten information and compensate for cognitive weaknesses.
And this raises a question I have found increasingly difficult to separate from conversations about AI: what are we actually trying to evolve into?
There is an assumption beneath much of technological progress that more capability is inherently desirable. More intelligence. More productivity. More memory. More knowledge. More control.
But removing limitations does not tell us what we should do once those limitations are gone.
Recent AI already hints at this problem. Used well, it can extend human cognition. Used passively, it can substitute for it. The distinction is subtle but important. A technology that helps us think can also gradually become a technology that thinks instead of us. Biological integration would make that distinction considerably harder to see because eventually the boundary between tool and user may itself become ambiguous.
Then there is an even stranger consequence.
Much of human civilisation has been organised around material scarcity. We compete for land, resources, labour, objects and access because all of them are limited. Knowledge was once scarce too. Expertise took decades to acquire. Memory was limited. Calculation was expensive. Communication was slow.
AI has already begun reducing some of those cognitive scarcities.
If intelligence itself eventually becomes abundant, personalised and integrated into human cognition, the hierarchy of what humans value could change again.
The material world would not disappear. We would still require food, energy, shelter, infrastructure and a functioning planet. Biology does not abolish physics. But material accumulation might become less central to what separates one person's capabilities from another's.
The more interesting scarcity could become attention, agency, experience, purpose and control over one's own cognition.
Perhaps that is where the conversation about biocomputing ultimately leads.
Its greatest contribution may indeed be energy efficiency. But solving AI's energy problem could unlock consequences far beyond electricity. It could help move intelligence from enormous external machines toward increasingly intimate, persistent and eventually biological systems.
And before celebrating that as the next stage of human progress, we should probably decide what we believe progress means.
Because the important question may not be whether we can build enough energy to support ever more powerful intelligence.
It may be what happens to humanity when intelligence no longer needs very much energy at all.