The Pulse
AWS Takes TBC’s Neuron-Derived Video Model Toward Market
Amazon Web Services is partnering with The Biological Computing Co. to commercialize a video model optimized using measurements from living neurons. TBC says the software delivers 5x faster generation and 80% lower inference costs on conven
AI.info Team ·
Amazon Web Services is backing The Biological Computing Co. as the startup moves a video-generation model shaped by living neural activity from laboratory research toward commercial distribution.
The companies announced the collaboration on September 22, 2026. TBC says its optimized text-to-video model generates video five times faster than the underlying model, cuts inference costs by 80% and improves output quality. The figures come from the companies’ announcement, which does not identify the open-source text-to-video model or publish a benchmark protocol.
The software does not contain living cells. TBC measures how real neurons respond to signals, extracts computational principles from those responses and turns them into a lightweight software layer that runs on standard AI hardware.
“Our partnership with AWS takes neuron-derived AI optimization to commercial scale,” Alex Ksendzovsky, TBC’s CEO and co-founder, said in the announcement. “We’re turning discoveries from real neurons into faster, cheaper AI for businesses and creators.”
AWS Targets Trainium, SageMaker and Its Marketplace
The partnership covers several parts of AWS’s AI stack. TBC plans to run the optimized model on AWS Trainium, make it deployable through Amazon SageMaker AI and pursue distribution through AWS Marketplace.
That route would put TBC’s software inside services that companies already use for model deployment and inference. The announcement does not say when the model will become generally available. TBC is inviting customers to request early access.
Jason Bennett, AWS’s vice president and global head of startups and venture capital, described the work as an attempt to transfer biological efficiency into conventional computing. “TBC’s insight is that we can learn from the original computer—the human brain—to make AI faster, more efficient, and more economical,” Bennett said.
The commercial pitch is straightforward: lower inference costs can let a video platform produce more clips from the same infrastructure budget, while faster generation can shorten the time between a prompt and a usable result.
Neurons Stay in the Lab
TBC’s customer-facing system uses silicon hardware, not a biological processor. The company says its software layer adds less than 0.1% to the underlying model and requires no biological hardware or new customer workflow.
On its website, TBC describes a three-stage process. Researchers encode an AI problem into electrical signals, measure neural responses and identify useful computational patterns, then deploy those patterns as code around an existing model.
The company positions that approach as different from simply scaling model size or adding more accelerators. Its longer-term plan includes applying the method to additional architectures and workloads, followed by work toward systems in which biological components and silicon operate together.
Earlier Tests Focused on Interactive Video
TBC has previously described a related experiment built around Oasis, an open-source Minecraft video-generation model. In a technical post, the company says its Hypercolumn Neural Optimizer adds about 14 million parameters, or roughly 3% of the base model, and introduces local spatial mixing alongside time-based memory.
TBC evaluated that system across 200 randomly seeded action sequences, each 600 frames long, and compared the results with real gameplay. The company says the optimized model retained visual detail and stability longer than the unoptimized version.
In a separate serving test, TBC says the system increased single-stream output from roughly two frames per second to nearly 10 frames per second, a 4.4-fold improvement. The company attributes the speed increase to the adapter combined with a streaming cache, a higher-order sampler and multi-session serving across GPUs.
TBC also acknowledges limits in that work. The evaluation centers on one Minecraft environment, and the company says it still needs to test whether the gains transfer across other environments, action patterns and inference settings.
The Commercial Test Is Still Ahead
The AWS announcement presents the text-to-video system as TBC’s first commercial product created through its biological discovery process. It does not disclose pricing, customer commitments, model size, hardware configurations or independent validation of the claimed fivefold speed improvement and 80% cost reduction.
Those details will matter because video inference costs depend heavily on clip length, resolution, sampling steps, hardware utilization and serving design. TBC’s earlier Oasis results show how much of the speed gain can come from the combined software and serving stack rather than from the biological adapter alone.
For now, AWS is giving TBC a path into its infrastructure and distribution channels, while TBC supplies a software method built from experiments with living neurons. The immediate product is a conventional AI model; the biology remains part of how its optimization was discovered.