Andrew Huberman: Why Every AI Lab Will Become a Brain Biotech Company

Andrew Huberman tells Patrick O’Shaughnessy how to switch your mind off for sleep, and why he thinks OpenAI, Meta and Anthropic will all end up as brain biotech companies. The conversation, from Invest Like the Best, starts with a practical question about energy and burnout and then spends most of its 78 minutes on neurotechnology: reading from and writing to the nervous system without opening the skull.

TLDW

Huberman argues that for the most capable people the bottleneck is now personal energy, and that the skill nobody trains is switching off at night. He gives a three-part method for falling asleep (quiet the thoughts, slow the heart, lose track of your limbs) and describes an eye mask from an MIT group that does part of it for you. From there he predicts that the big AI labs are really racing to read and write brain activity non-invasively, that implantable vagus nerve stimulators will go mainstream among founders, and that a gene delivered by virus plus a wearable cap could one day let people dial their own motivation.

Thoughts

Huberman spends most of this conversation predicting devices that will turn brain states up and down, and then, about 37 minutes in, describes an experiment suggesting nobody knows how the dial works. As Huberman tells it, researchers tagged the neurons active while an animal learned something, then reactivated them later and got the behavior back. The surprise was the control: fire those same neurons in reverse order, or all at once, and the behavior still appears. If the order of firing does not matter for memory, a core assumption in the textbooks is wrong. He frames this as good news for engineers, who may need less precision than they feared, and in the same breath calls it a huge gap in understanding. It is hard to hold “we will soon dial motivation up and down” next to “there is no chapter that explains how thoughts are made.” He puts both on the table, which is to his credit.

That gap is the right lens for the headline claim, that every major AI lab will become a biotech company. It deserves both attention and a discount. It rests on Huberman’s private conversations with the people running those companies, not on anything they have announced, so treat it as his read. The reasoning for why it would start with sleep is good, though. We already let devices measure us while we are unconscious, and a mattress that changes temperature to shift sleep stages is already a crude form of writing to the nervous system. The timelines are where I would be careful. A sleep mask in 7 to 12 months is plausible. One in five entrepreneurs wanting a vagal implant within a year is the kind of number that sounds right in Palo Alto and nowhere else, a bias he cheerfully admits to twice.

The proposal that should make people stop and think comes around the 50 minute mark: an injected viral vector that makes chosen neurons respond to light or ultrasound, and a cap that does the stimulating. Huberman first calls this a “vulnerability” and then swaps the word for “accessibility.” His defense is that we already have caffeine, and nearly every adult uses it. But caffeine does not come with a vendor, firmware and a network connection. “You would hope” the companies add safeguards is carrying a lot of weight in that passage. His own story about He Jiankui, the scientist who edited the genomes of twin babies, makes the opposite point: once a capability exists, someone uses it before the ethics committee meets. (Huberman repeats a rumor that He now runs a lab in Austin and says several times that he cannot confirm it. Neither can I.)

The practical advice holds up better than the predictions, and the best of it is a protocol Huberman declines to give. Asked how founders should manage their energy, he does not prescribe eight hours and breath work. He says the first four years of a company are about not dying and not injuring yourself, and that the real work is an honest read of how much sleep you personally need to feel great, to get by, and to fall apart. Then he names the thing most high performers never train: the off switch. O’Shaughnessy admits it can take him four hours to come down from a good day. That reframing is worth more than any routine. The limit on a long career may not be how hard you can run. It may be how quickly you can stop.

The last ten minutes will outlast the devices. Huberman says he is done repeating the basics, that his book Protocols is the thesis turned in, and that 2027 and 2028 will be quiet years for health content before better trained people arrive. Then he tells the story of his graduate adviser, Barbara Chapman, who caught him sneaking into her lab at night after he had joined a more fashionable one, and told him the only thing that would decide his success was how badly he wanted the answers. It is a little funny to hear the man most associated with optimization say the real variable is curiosity. It is also the most transferable idea in the conversation, and it costs nothing to test.

Key Takeaways

Know your own sleep number. Huberman feels great on seven hours, groggy on eight, fine on six for a day, and “cratered” after two nights of four or fewer.
Having lots of energy is not the same as being in a healthy state. Some high-output people are redlining, with or without stimulants, and cannot switch off at night, so they never reset.
Falling asleep has three requirements: turn off your thoughts, slow your heart rate, and forget where your limbs are in space. The order does not matter, and combining all three works best.
To quiet thinking, move into pure sensation. Shift attention from your feet to your legs to your breathing to sounds in the room and back. Huberman says this is not meditation, which holds attention on one point.
Long exhales until the lungs are empty, followed by passive inhales, slow the heart through respiratory sinus arrhythmia and the vagus nerve.
Slow eye movements under closed eyelids (side to side, up and down, circles) mimic what rocking does to a baby and help the brain lose track of body position. An eye mask from an MIT group does this with mild stimulation behind the ears and measures REM directly.
He calls public health and longevity discourse a disaster with a bar for entry of “basically zero,” and expects a downturn as people tire of theatrics and of doing too many things each day.
His prediction: Meta, OpenAI and Anthropic are not really LLM companies. The goal, he says, is to read and write from the brain non-invasively, probably with ultrasound, and “that’s what the real arms race of AI is about.”
Most vagal pathways are excitatory, so vagus nerve stimulation increases alertness. Implanted stimulators are already used for depression, and he expects founders to start asking for them once the procedure feels no more invasive than a piercing.
Neuroscience understands the senses and the deep switch-like circuits for hunger, anger and hormones fairly well. That is why GLP-1 drugs work, and why very high doses can blunt drive for other things too. Memory and thought are where the model breaks down.
Heart rate and HRV are indirect. A former postdoc of his is building a cuff that reads the autonomic nervous system directly and can tell good stress from bad, and he expects focus to be measured from pupil size through eyeglasses.
The advice he has carried since graduate school: success is determined by how precisely you match your genuine curiosity to the work you do.

Chapters

00:41 Energy Is the New Bottleneck for Talented People

O’Shaughnessy opens with an observation from founders: technology now unlocks so much that the constraint is their own energy. Huberman sets Musk aside as an outlier with no peers and says everyone else needs an honest self-assessment of their sleep need at this age and stage of life. He recalls 80 to 100 hour weeks in graduate school with two workouts a week, and says the first years of a company are about getting as much done as possible without injury or chronic illness.

06:52 How to Turn Off Your Thoughts and Fall Asleep

The brain wants to plan and predict, so stopping it is a skill. Huberman’s method is to move attention around the body and the room to get into pure sensation, use long exhales to drop the heart rate, and then confuse the system that tracks limb position. Studies show a bed rocking at the right frequency puts adults to sleep quickly, because rocking drives compensatory eye movements that feed the balance system. You can copy that by moving your eyes slowly under closed lids, or wait for the MIT mask that does it for you and then amplifies REM.

19:47 Why Public Health Discourse Is a Disaster

Asked for the biggest problem in his field, Huberman goes straight at the influencer economy. Showing your abs is enough to build a following, while in neuroscience a PhD is necessary but nowhere near sufficient. He says the gap between the top tier of health educators and the rest keeps widening, and that he needs more serious peers with training in science or medicine. He expects them to arrive after a downturn.

23:51 AI Labs as Biotech and the Race to Write to the Brain

On the research side, the big problem is writing to the nervous system without surgery. Huberman describes watching his childhood friend Eddie Chang, chair of neurosurgery at UCSF, operate on awake patients, and says breaching the skull will stay a clinical last resort. He predicts ultrasound will be the route and that the AI labs are all chasing it. Vagus nerve stimulation is the near-term example: he recounts Karl Deisseroth describing a suicidally depressed patient who began talking about applying for a job as the stimulation was turned up.

33:49 What Neuroscience Knows and the Memory Result That Breaks the Model

Huberman maps the state of knowledge from the outside in. The sensory organs and the hypothalamus are well understood, and the prefrontal cortex is known to handle context, strategy and impulse control. Then he describes experiments from the labs of Mark Mayford and Susumu Tonegawa in which tagged memory neurons produce the same behavior whether they are reactivated in order, in reverse, or all at once, like banging every key of a piano and still hearing the song. Chang’s speech work succeeds, he says, because it reads from the planning areas and lets a computer do the rest.

46:28 The Cyborg Line: Direct Measurement, a Virus and a Hat

O’Shaughnessy asks what excites and scares him. Huberman says people got used to phones listening and to driverless taxis, and direct measurement of the waking nervous system is the next hurdle. Beyond that he wants access to all his neurons by brain structure, under his own control. The mechanism he sketches is a harmless viral carrier that adds a light or sound sensitive channel to chosen cells, a cap that stimulates them, and AI that adjusts the dose, possibly with a pill as a third safeguard.

55:48 From Blunt Tools to Precise Ones

Morning sunlight is not biohacking, Huberman says, because a circuit from the eye to the hypothalamus exists for exactly that purpose. Drugs are a different matter. SSRIs, psilocybin and MDMA all work by raising neuromodulators such as serotonin to open a window of plasticity, and he expects magnetic stimulation to be paired with them for depression and PTSD. Raising creativity or IQ on demand is “not even close,” because nobody knows which circuit to target.

1:01:10 CRISPR Babies and Where Human Genetic Engineering Stands

Huberman tells the story of He Jiankui, who announced he had used CRISPR to delete the HIV receptor in twin babies, and of colleagues waiting to see whether he would be celebrated or jailed before deciding how close to stand. He raises the suggestion that the same receptor is linked to memory, which left an open question about whether the aim was also enhancement. He says he is more excited than worried, provided people talk about it openly, and that nothing stays secret for long anymore.

1:06:32 Waking Brain States and the Technologists Turning Neuroscientist

Sleep stages are well mapped, but Huberman says there is no good language for waking states, including the focused, enjoyable one he is in during the interview. Defining those states, and which suit which activities, is what he would work on first. He notes that Dario Amodei came up through neuroscience recording from retinas, and says Sam Altman, Mark Zuckerberg and Musk are all sharp on the subject. He calls them the athletes of brain exploration.

1:13:22 The Kindest Thing: Follow Your Genuine Curiosity

For O’Shaughnessy’s closing question, Huberman describes joining a fashionable molecular genetics lab while sneaking into Barbara Chapman’s lab at night to do the experiments he actually loved. She told him to switch, and they went on to publish eight first author papers together. He applies the same test to starting a podcast, and to Musk, whose real gift he thinks is ignoring what others say he should work on. Telling him to skip Mars is “like telling Metallica to play the Grateful Dead.”

Notable Quotes

“The Oracle said it best, like know thyself. You really have to be honest with yourself about what you need in order to do something for a long period of time.”
Andrew Huberman, on what he tells young people who ask how much sleep and training they need

“It’s very easy to get attention. It’s very hard to keep respect.”
Andrew Huberman, on the widening gap between serious health educators and the rest

“Anyone that thinks that those are LLM AI companies is perhaps going to be surprised at this prediction. Every single one of those is going to be a biotech company.”
Andrew Huberman, on Meta, OpenAI and Anthropic

“I’m not worried about peptides. I’m thinking implantable devices is where it’s going to go next.”
Andrew Huberman, on vagus nerve stimulators moving from the clinic to founders

“What if I activate all those neurons at once, like banging on the keys of the piano? It’s not the same as playing the song.”
Andrew Huberman, setting up the memory experiment in which scrambled activation still produced the learned behavior

“I want to understand waking brain states. I don’t know how to describe the state that we’re in.”
Andrew Huberman, on the problem he thinks neuroscience most needs to solve next

“Success is determined by how precisely you match your genuine curiosity to the work that you do.”
Andrew Huberman, restating the advice of his graduate adviser Barbara Chapman

The sleep section alone is worth the first twenty minutes, and the neurotechnology argument is best heard in Huberman’s own words, hedges and all. Watch the full conversation here.