Frontier Computing Raises $10M Pre-Seed to Grow AI on Living Neurons

Training AI models is expensive and energy-hungry, largely because GPUs spend most of their power moving data back and forth between memory and compute. Living neurons don’t have that problem — memory and computation happen in the same place. But lab-grown neural tissue has been stuck at roughly a million neurons: past that point, cells at the center of the cluster sit too far from an oxygen supply and die.

A startup claiming to have broken that limit has raised a large seed round. Cambridge, UK-based Frontier Computing has raised $10 million in pre-seed funding, Amino Collective announced. A member of Y Combinator’s Summer 2026 batch, Frontier grows real, lab-cultured neural tissue as a computing substrate — what the company calls “wetware.” Biological neurons learn from far less data than machine learning models typically need, and don’t require retraining from scratch to pick up new information, according to the company. Frontier says its tissue-culture technique gets around the oxygen-supply limit, allowing it to grow neural tissue at a scale well beyond prior lab work. It has already trained its system to play the arcade game Frogger, reaching a 92% success rate after an hour of real-time learning.

Founder Michael Domarkas is 19 and studied natural sciences at Cambridge. He has been experimenting with culturing neural tissue since he was 17, and started the company as an extension of that work.

The round was led by General Catalyst, with participation from Amino Collective, LocalGlobe, KAYA, and Long Journey Ventures. The funding will go toward growing the team and its wetlab operations. Frontier says it’s targeting a 500-million-neuron cluster by January 2027 — 2,500 times larger than the biggest commercially available biocomputing system today. That target is still a plan, not something that exists yet.

The biocomputing field

Biological-neuron computing — “wetware,” or synthetic biological intelligence — is still an early field, but a few companies have already commercialized products. Australia’s Cortical Labs sells its CL1 biological computer, built from neurons cultured from human stem cells on a silicon chip, for $35,000, and has raised $11 million to date. Each CL1 unit runs roughly 200,000 neurons — an order of magnitude below Frontier’s 500-million target. Switzerland’s FinalSpark rents remote access to its Neuroplatform — 16 brain organoids totaling roughly 160,000 neurons — to researchers for $500 a month. All three companies pitch dramatically lower energy use than silicon GPUs, but none has yet been used to train a commercial-scale large language model.