TL;DR
Amazon Web Services is now distributing a "rat brain" AI model from Baltimore startup The Biological Computing Company, marking a significant commercial milestone for the fringe field of biological computing.
What happened
- AWS launched a limited preview of TBC's biologically-derived video generation model on September 23, 2026, with a full enterprise rollout expected soon.
- TBC encodes images onto living rat brain cells placed on multi-electrode silicon arrays, records how neurons respond, then translates those patterns into software that improves existing AI video models.
- The startup claims its model is up to 5x faster at video generation and meaningfully lowers inference costs compared to the open-source frontier model it runs on (name undisclosed).
- TBC has raised more than $50 million total: a $25 million round led by Primary Venture Partners closed in March, followed by an unreported second $25 million raise shortly after.
- The company, founded four years ago by neuroscientists Alexander Ksendzovsky and Jon Pomeraniec, has about 35 employees and operates a wet lab in San Francisco using 3Brain silicon arrays.
Why it matters
- AWS is the distribution lever that could scale biological computing from a lab curiosity to an enterprise product, reaching a customer base far beyond TBC's prior neocloud partner Bluesky Compute.
- TBC is not trying to replace transformers: it works within existing generative AI standards, fine-tuning video models rather than rebuilding architecture, which lowers adoption friction for enterprise buyers.
- AWS already carries Cortical Labs (lab-grown neurons on silicon, sold as "wetware as a service"), signaling Amazon is building a deliberate portfolio in biological compute, not making a one-off bet.
- Jeff Dean's involvement as investor and strategic advisor (he suggested the video fine-tuning approach) adds credibility and connects TBC to the top tier of AI research networks.
- A 5x inference speed claim on video generation, if it holds at scale, would be directly relevant to cost-sensitive enterprise media and advertising workloads where video AI spend is surging.
What to watch next
- Fidelity at scale: AWS's own director flagged the open question of whether quality holds for longer-form video (10 minutes, one hour). Early enterprise results will either validate or deflate the benchmark claims.
- Disclosure of the baseline model: TBC refuses to name the open-source frontier video model it benchmarks against. Naming it, or a third-party audit, would be the credibility signal serious buyers need.
- Expansion beyond video: TBC's rat-brain approach was chosen partly because images map cleanly onto electrode grids. Watch for announcements applying the method to language or multimodal tasks, which would signal the technology is genuinely generalizable.
Originally published on Present of AI, a daily source-linked AI news timeline. Read the full timeline or browse the open dataset.