TBC's neuron-derived adapter claims 5x faster AI video on AWS, but its baseline is unnamed
- The Biological Computing Co. (TBC) and AWS announced on September 22, 2026 a "neuron-derived" adapter that plugs into existing diffusion video generators, adds less than 0.1% to model size, and claims 5x faster inference with 80% lower cost.
- The Decoder points out that the 5x speed-up and the 80% cost cut are the same improvement stated twice, since cloud inference cost scales directly with compute time.
- TBC grows cortical nerve cells on chips wired with 4,096 electrodes, records how activity propagates across the culture, and converts those signal patterns into compact adapter modules that slot into a diffusion model without retraining its core weights.
- TBC did not name the open-source base model, so every performance number is measured against an uncheckable baseline, and the video quality claim carries no standardized metric.
- A related variant used 156,000 parameters on a 600-million-parameter video model to roughly double visual coherence at under 0.5% overhead, and the AWS release will run on Trainium chips and be licensed via Amazon SageMaker AI and the AWS Marketplace; TBC raised $25 million in February 2026.