Huberman says AI labs are really biotech

Neuroscientist Andrew Huberman went on Invest Like The Best with a blunt reclassification: Meta, Anthropic and OpenAI are not LLM companies. They will all become biotech companies, he said, because every one of them wants to read from and write to the human brain without surgery.

That is the punchline, not a throwaway.

Huberman, a Stanford neuroscientist and host of Huberman Lab, argued the tell is who these labs hire and what they study in private. He said leaders at the big AI labs may lack formal neuroscience training but they understand a lot of neuroscience and they have converged on the same non-invasive brain interface goal. In his telling, large language models are a waypoint, not the destination.

The comparison sharpens the claim. Meta has Reality Labs and its EMG wristband research, OpenAI has explored hardware and bio-adjacent bets through its personnel and partnerships, and PayPal mafia-adjacent founders now funding AI labs openly talk about cognition and longevity. Huberman puts Elon Musk’s Neuralink in the same conversation even though its current product is invasive, to argue the field will race toward doing the same thing without drilling a hole. That is a very different competitive map than model-vs-model benchmarks.

The second half of the conversation explains why he thinks biology is the bottleneck to that future. Host Patrick O’Shaughnessy framed top founders as bandwidth limited, able to generate leverage from AI but constrained by energy and sleep. Huberman answered with a protocol stack, not a hack: honest assessment of sleep need, long exhales to slow heart rate via respiratory sinus arrhythmia and the vagus nerve, shifting attention from rumination to pure sensation, and even confusing proprioception to fall asleep faster. He cited rocking beds that work in studies, voluntary eye movements under closed lids, and a new MIT-linked eye mask that measures REM directly and uses gentle stimulation behind the ears to induce the eye movements that help people drop off and amplify REM. He said versions are 7 to 12 months from release, and he has no stake.

The health piece is not separate from the biotech thesis. Huberman predicted the same non-invasive stimulation will move from sleep into waking, with eyeglasses or wearables that nudge autonomic state toward focus or calm at different times of day. Today that is cold showers, caffeine, exercise and breathwork. Tomorrow he expects hardware that does it more precisely. The limitation is obvious in his own telling: none of this replaces sleep need, and most founders who try to run on four hours because Musk does will crater. The labs may want to write to your brain, but they still have to prove they can do it safely, non-invasively, and outside a lab.