TBC plans to sell its neuron-taught video model through AWS

The Biological Computing Co. is taking a video generation model that was tuned using measurements of living neurons to market, in a collaboration with Amazon Web Services announced on September 22, 2026.

TBC says its optimized text-to-video model runs five times faster and at 80 percent lower inference cost than the open-source video model it is built on, with better output. The San Francisco company has not published benchmarks for it, and it has declined to name the base model, which it describes as a frontier video generator.

The optimization is a software layer derived from recordings of real neural activity. According to the AWS announcement, the layer adds less than 0.1 percent to the parameters of the underlying model and runs on ordinary AI infrastructure. No biological hardware is involved on the customer's side.

Under the collaboration, TBC plans to run the optimized model on AWS Trainium chips, make it deployable through Amazon SageMaker AI, and sell it through the AWS Marketplace. WIRED reported that select AWS customers gained access as a limited preview starting September 22, and both companies expect the technology to reach all AWS enterprise customers eventually. For now, the public door is an early access form on TBC's site, which promises a beta this fall.

The company's lab work is unusual. TBC keeps rat brain cells and human stem cells alive on multi-electrode silicon arrays made by the Swiss biotech firm 3Brain, sends electrical patterns into the neurons, and records how they respond. Images map onto that electrode grid more naturally than language does, which is part of why the company started with video. Jeff Dean, a prominent AI researcher and a TBC investor, suggested the video fine-tuning path first, according to WIRED.

Two numbers worth reading carefully: five times faster and 80 percent cheaper are, if you need a fifth of the GPU time, the same measurement in different units. The Decoder also noted the commercial model's details are thin: no named base model, no resolution, no clip length, no quality metric. An earlier, larger adapter of roughly three percent extra parameters pushed frame rate from about two to just under ten frames per second, and an outside test by infrastructure provider Bluesky Compute in July measured about 4.4 times lower inference costs and more than three times as much coherent video as the base model.

WIRED also reported that TBC raised $25 million earlier this year in a round led by Primary Venture Partners and an additional $25 million shortly after it closed, for more than $50 million total. That second tranche had not been reported before.

AWS's own global director of technology for startups, Deap Ubhi, put the open question plainly to WIRED. Pushed to the extremes, will the improvements hold, and what happens to fidelity over a ten minute or hour-long generation?

A cheaper, faster video model would change what a creator can afford to attempt, since iteration count is usually the thing that eats a budget. Whether TBC's numbers survive an independent test is the part nobody outside the company has measured yet.

The model is not generally available. Early access is a request form at tbc.co/early-access, with a beta promised for this fall.