Announced September 22, 2026, the partnership commercializes a text-to-video model tuned by measurements of living neurons. The Biological Computing Co. claims 5x faster generation and 80% lower inference costs, sold through AWS on Trainium, SageMaker, and the Marketplace.
2026-09-22 · by Moe Ameen
On September 22, 2026, The Biological Computing Co. (TBC), a San Francisco startup that studies living neurons to improve AI, announced a partnership with Amazon Web Services to bring what both companies call the world's first neuron-derived AI video model to market. The pitch is not a new generator built from scratch: TBC took an existing open-source text-to-video model (it has not named which one) and added a lightweight optimization layer, then teamed with AWS to sell the result. TBC says the tuned model generates video about 5x faster and at roughly 80% lower inference cost than the base model, with improved quality.
The unusual part is where the optimization comes from. TBC grows living rat neurons — and human stem-cell-derived neurons — on multi-electrode arrays and records how those biological networks respond to encoded images and video. Software turns those biological responses into adapters that layer onto conventional transformer models. TBC says the resulting layer adds less than 0.1% to the underlying model's size and runs entirely on standard AI hardware — no biological hardware in the loop at inference time, and no change to a customer's workflow. The neurons are used during discovery; what ships is ordinary software.
Under the AWS deal, the model is set to run on Amazon's Trainium chips, be offered through Amazon SageMaker AI, and be listed on AWS Marketplace. TBC CEO and co-founder Alex Ksendzovsky framed it as taking neuron-derived optimization "to commercial scale"; an AWS executive, Jason Bennett, cast it as learning from "the original computer — the human brain — to make AI faster, more efficient, and more economical." The company, founded by neurosurgeons Ksendzovsky and Jon Pomeraniec, emerged from stealth earlier in 2026, raised a $25 million seed round led by Primary Venture Partners, and runs a wet lab in San Francisco's Mission Bay. An earlier proof-of-concept on Decart's OASIS video model reported quality and cost gains as well.
Two important caveats. TBC has not published peer-reviewed benchmarks for the video model, and reporters covering the announcement noted they could not independently verify the 5x-faster and 80%-cheaper figures — treat them as vendor claims until third-party numbers exist. And the details around availability, pricing, and which base model was used are thin; confirm specifics against TBC's and AWS's own pages before relying on them.
The right way to read this is as a plumbing story, not a product you'll open. A neuron-tuned model that makes text-to-video cheaper and faster is an upgrade to one stage — the raw render — sold through AWS to the people who build tools. What it hands back is still a bare clip: no captions, no 9:16 / 1:1 / 16:9 versions, no brand voice, no schedule, no destination. That last mile is exactly what [Kompozy](/) owns, and it doesn't get shorter because the model underneath got faster. So the practical move today is to cover the news, not chase the model: drop your take on neuron-derived AI into Kompozy as a source and it fans one point of view into a Blog Article explaining what a "biological" optimization layer actually is, a brand-exact [Carousel](/glossary/hyperframes) contrasting the hype with the unverified benchmarks, [Quote Graphics](/glossary/output-buckets) of your sharpest line, and a captioned [Persona Short](/glossary/persona-shorts) your face-locked avatar reads to camera — each held to your voice by the [Persona Brief](/glossary/persona-brief) and scheduled across the eight social platforms plus blog and email on [Autopilot](/glossary/autopilot). Being early and clear on a confusing story is worth more than being first to the model.
There's a longer-run angle too, and it's why this is upside rather than a threat. Kompozy treats every generator as an interchangeable input — its own video already draws on HeyGen for avatars and Pexels for B-roll, and copy runs on Claude and OpenAI. If a neuron-optimized video model lands on AWS Marketplace and proves its numbers, it becomes one more render engine that could sit under the pipeline, not a competitor to it. You wouldn't rebuild your workflow around it; you'd keep [generating persona and clipped video](/ai-video-generation), captioning, reframing, and publishing in Kompozy, and inherit a cheaper render underneath. The model layer keeps getting faster and stranger; the part that decides whether anyone sees the work — the captioning, the brand consistency, the multi-platform scheduling — is the layer Kompozy runs.
It is an open-source text-to-video model optimized by a software layer that The Biological Computing Co. derived from measurements of living neurons grown on electrode arrays. Announced with AWS on September 22, 2026, it runs on conventional AI hardware — the neurons are used during research, not at inference — and TBC claims roughly 5x faster generation and 80% lower inference cost versus the base model.
Not independently. TBC has not published peer-reviewed benchmarks for this video model, and reporters covering the announcement noted they could not verify the figures. Treat them as vendor claims until third-party numbers appear, and confirm specifics on TBC and AWS pages.
Not as a consumer app. The model is being distributed through AWS — Trainium, Amazon SageMaker AI, and AWS Marketplace — so it targets developers and platforms first. Most creators would encounter it indirectly, as a faster or cheaper model inside a tool they already use. It also only generates raw video; captioning, per-platform sizing, brand voice, and publishing are separate steps a content engine like Kompozy handles.