Product1 publisher3 min readPublished
Amazon starts previewing a video model reverse-engineered from rat neurons
Selected AWS customers can try The Biological Computing Company's video model from Tuesday. The cells stay on silicon arrays in San Francisco, and what customers get is software built to imitate how those cells handled images.
The Product Desk · Product desk

What happened
- From Tuesday, selected Amazon Web Services customers can use The Biological Computing Company's "rat brain" AI model, which the startup built to improve video generation, as part of a limited preview.
- Amazon and TBC both say they expect the technology to reach all AWS enterprise customers soon, though neither has put a date on general availability.
- AWS's Deap Ubhi said the cloud provider already works with Cortical Labs, the Australian company that pairs lab-grown neurons with silicon chips and sells it as wetware as a service.
Compiled by The Product DeskSomething wrong?How this is made
Why it matters
- decision Because the neurons never leave TBC's lab, a buying team should review this as an ordinary model licence covering weights, latency and output liability.
- constraint Without a published price or an efficiency figure, nobody can build a business case for swapping video inference onto this model, so preview participation is paid for in staff hours spent benchmarking.
- capability Listing biology-derived vendors in the AWS marketplace lets them reach enterprise buyers through an existing cloud agreement, which is a far shorter path than signing a four-year-old lab company directly.
- precedent With Cortical Labs keeping neurons alive in a device and TBC shipping software copied from neurons, "biological computing" in Amazon's catalog now covers two products with almost nothing operationally in common.
What arrives in a customer's account is a model. The rat cells sit on multi-electrode silicon arrays made by the Swiss firm 3Brain, where TBC researchers send in electrical stimulation patterns and record how the neurons respond [14]. The company then looks through that activity for computational patterns it can turn into software and apply to existing video generation models [15]. Cofounder Alexander Ksendzovsky described the loop to Wired: "We figured out a way to code information, like images for example, to the biological material," he said, and "We then observe how the biology processes that information, and then we build a tool that mimics that process." [4][5]
The biology, then, sits upstream of the product. No living cells run in the buyer's inference path [1]. Anyone filling in a vendor questionnaire on Monday can handle it the way they handle any other weights-and-API purchase.
It also separates TBC from the other biology vendor in Amazon's catalog. Deap Ubhi, AWS's global director of technology for startups, said TBC is not the first biologically derived computing platform Amazon has made available in its marketplace, and that AWS also works with Cortical Labs, an Australian company that combines lab-grown neurons with silicon chips and sells the result as "wetware as a service" [6][7]. Cortical Labs also sells a multi-thousand-dollar low-power "biological computer" for laboratory use that can supposedly keep neurons alive for six months [8].
The pitch is efficiency. Ubhi said of TBC that "they're working within existing standards of the generative AI space and saying, 'How can we make the current visual models more efficient?'" [9]. Wired's account carries no price for the preview and no measured efficiency comparison [20].
TBC raised $25 million earlier this year in a round led by Primary Venture Partners, then secured another $25 million shortly after that round closed in March, taking its total above $50 million [11][12]. The San Francisco office and lab it opened last year holds 35 employees working with rat brain cells and human stem cells [13]. Divide the funding by that headcount and you get roughly $1.4 million per person in the lab [2]. TBC is four years old, founded in Baltimore by two neuroscientists who are also neurosurgeons, Ksendzovsky as CEO and Jon Pomeraniec as president and COO [10].
Video came first for a physical reason: electrode placement on the array changes how each electrode interacts with the neurons, and images mapped onto that grid more readily than text or language [16]. Pomeraniec said the suggestion to start with fine-tuning video generation models came from Jeff Dean, an AI researcher who is also a TBC investor [17].
Two questions sort this for a buying team. First, is the biology in the loop at runtime? Here the answer is no, so the diligence is ordinary: license terms, latency, cost per second of rendered video, who stands behind the output. Second, is the claimed gain measured on your prompts and your hardware, or on the vendor's? In a limited preview, a buyer still has the leverage to insist on the first and refuse the second.
Who this is for: teams already generating video at enough volume that inference cost shows up as a line item, and who have someone willing to run a controlled comparison and write the number down.
What to watch
- Whether general availability for AWS enterprise customers arrives with a published price or a measured efficiency comparison.
- Whether the AWS marketplace listing states that the model was derived from rat brain cells and human stem cells, which legal and ethics reviewers will ask about.
- Whether AWS files TBC's software and Cortical Labs' wetware service under the same catalog category, and how it describes the difference.