A rat-neuron-inspired video model from a Baltimore startup reaches an AWS preview, with its baseline still unnamed

WIRED's Lauren Goode reports that The Biological Computing Company (TBC) is putting a model derived from rat brain research into a limited Amazon Web Services preview for select customers, starting Tuesday, September 22, aimed at video generation. Amazon and TBC say a broader enterprise rollout is expected soon. Until now the model ran only through Bluesky Compute.

Cofounder Alexander Ksendzovsky describes the method as encoding images into biological material, observing how neurons process them, then building software that mimics that processing. The shipped product is software, not living tissue. A San Francisco lab of about 35 people works with rat brain cells and human stem cells on 3Brain multi-electrode arrays.

TBC says its model is up to five times faster at video generation than the open-source frontier video model it compares against, with lower inference cost. It has declined to name that baseline. Jeff Dean, listed as an investor, reportedly pushed the company to prove value on video benchmarks before making broader claims. Performance figures are TBC's and AWS's as reported by WIRED; no public benchmark is cited.

AWS's Deap Ubhi says he likes the approach of improving existing visual models rather than reinventing the transformer. He also raised the main caution himself: whether consistency holds if customers push toward ten-minute or hour-long video. A speed gain on short clips is a narrower result than one that holds over long-form coherence.

TBC raised $25 million in a round led by Primary Venture Partners and then another $25 million after that March close, for more than $50 million in total. WIRED reports the second round was previously undisclosed. AWS already works with Cortical Labs' wetware-as-a-service offerings, so this is not Amazon's first biological computing partnership.

Cloud distribution matters for a field that has mostly lived in labs and academic demos. A listing in an enterprise cloud gives the approach paying customers who will test it against ordinary generative models on their own workloads. That raises the evidence bar too: customers, not only investors, will be checking the claims.

Video is one of the most compute-intensive generative modalities, and efficiency is the competitive theme across our recent coverage. Anthropic's Opus 5.5 claims Fable-level performance at about 40% lower running cost, and OpenAI's GPT-6 Sol and Luna halve API prices. Nvidia is pushing efficiency from the hardware side with local AI at IFA 2026. TBC's pitch is that biology can supply another source of efficiency, but unlike those vendors it has not named what it is beating.

Using rat neural tissue in research will also draw scrutiny from animal-research watchers, even though customers receive software. More in our research coverage and generative AI coverage.