A biotech startup is literally growing neurons to help power the next wave of AI

TBC emerges from stealth with a bold, strange bet

The Biological Computing Company — out of San Francisco’s Mission Bay — surfaced from stealth in February with a $25 million seed round and a team of about 23 people, led by former neurosurgeons Dr. Alex Ksendsovsky and Dr. Jon Pomeraniec. It has been reported that the staff includes ex‑engineers and researchers from Meta, Apple and Amazon. Walk into their lab and you don’t see racks of GPUs. You see chips populated with living cells. Creepy? Cool? Both.

How the living chips are said to work

It has been reported that each chip contains between 100,000 and 500,000 neurons grown from reprogrammed stem cells into frontal‑cortex‑like cells and housed on multi‑electrode arrays. The company allegedly used rat neurons early on but is “moving away” from that. Data — images, video — are encoded into the biology, the neurons respond, and those responses are decoded back into richer representations for downstream AI tasks. The chips reportedly last about a year and require waste cleanup every few days. Yes, there is actual waste. Welcome to wetware maintenance.

The pitch: energy savings, faster training, early customers

TBC claims the biggest upside is energy: brain cells, they say, can produce useful computations with far less power than brute‑force silicon. It has been reported that models trained on these “biological neural responses” reached peak performance about three times faster, implying a threefold reduction in compute and energy needs. For now the company isn’t selling chips; it is using them to improve visual AI — generative video, game rendering, computer vision — and, allegedly, is in talks with foundational model labs, cybersecurity firms and data companies. “All of the data flows through the biology,” Ksendsovsky tells visitors. Bold words.

The long view: fascination, hurdles, and the ethical twinge

TBC’s thesis sounds outlandish, bordering on sci‑fi — a little Frankenstein with lab coats — but it answers a very real problem: AI’s appetite for power. The road to real‑time, inference‑stage biological compute is still long; Ksendsovsky pegs that as a five‑to‑ten‑year goal and flags hard engineering problems like automating waste handling and extending cell lifespans. There’s an emotional tug here: awe at a novel approach and an uneasy sense that the era of “neural” networks may be coming home to roost — literally. Will investors, researchers and regulators be ready to let living tissue into the AI stack? That’s the question TBC will have to answer.

Sources: thedeepview.com