# The Biological Computing Company

> Source: https://aiwiki.ai/wiki/the_biological_computing_company
> Updated: 2026-10-31
> Fact-checked: 2026-10-31
> Categories: AI Companies, AI Inference, AI Research, Video Generation
> License: CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/) - attribute to "AI Wiki (aiwiki.ai)"
> Cite as: AI Wiki. "The Biological Computing Company." aiwiki.ai, 31 Oct 2026. https://aiwiki.ai/wiki/the_biological_computing_company
> From AI Wiki (https://aiwiki.ai), the free encyclopedia of artificial intelligence. Reuse freely with attribution.

**The Biological Computing Company** (TBC; styled The Biological Computing Co. and formerly called Biological Black Box) is an American startup that grows living neurons on multi-electrode arrays, records how the cultures respond to electrical stimulation, and turns those measurements into software modules ("adapters") for conventional AI models running on ordinary GPUs and accelerators.[9][11][16] It was founded in 2022 by neurosurgeon-neuroscientists Alex Ksendzovsky (chief executive) and Jon Pomeraniec (chief operating officer), started in Baltimore, Maryland, and is now headquartered in San Francisco.[12][20][2] Its first publicly available product, the image-to-video model **TBC mischief 1**, went on sale through [fal.ai](https://aiwiki.ai/wiki/fal_ai) and [OpenRouter](https://aiwiki.ai/wiki/openrouter) on 30 October 2026. The company says the model is built on the open-weight Wan 2.2 video model and generates video five times faster than that base.[1][2][3][4]

TBC says no neurons are involved when customers use its products. The cells stay in its labs as a discovery tool, and what ships is software that, by the company's account, adds less than 0.1% to the parameter count of the model it modifies.[2][7] The company says it had raised $52.5 million in total by September 2026, including a $25 million seed round led by Primary Venture Partners and a $25 million seed extension led by Builders VC.[18][10][12]

## Overview

| Field | Detail |
|---|---|
| Name | The Biological Computing Co. (TBC)[9] |
| Former name | Biological Black Box (BBB)[9][11] |
| Founded | 2022 (Baltimore, Maryland)[20][12] |
| Founders | Alex Ksendzovsky, MD, PhD (CEO); Jon Pomeraniec, MD, MBA (COO; also described as president)[8][12] |
| Headquarters | San Francisco, California; the website lists Baltimore and San Francisco[2][7] |
| Employees | About 35 (Wired, September 2026)[12] |
| Funding | $25 million seed (Primary, February 2026) plus a $25 million seed extension (Builders VC); $52.5 million total according to the company[9][18] |
| First public release | TBC mischief 1 image-to-video model (30 October 2026)[2] |
| Distribution | fal, OpenRouter; collaboration with [Amazon Web Services](https://aiwiki.ai/wiki/amazon_web_services) for Trainium, SageMaker AI and AWS Marketplace[2][17] |

## History

### Origins

Ksendzovsky told GamesBeat that he had been growing neurons on electrodes since about 2005. As a student in a lab that cultured brain cells on electrodes, he proposed using the cells to predict the stock market. His adviser rejected the idea, and Ksendzovsky went on to earn a doctorate in neurophysiology and become a neurosurgeon, and his postdoctoral research recorded data from electrodes implanted in patients' brains.[14][20] Primary, the company's lead investor, describes the two founders as neurosurgeons "from UMD and Penn" who worked together through medical school, surgical training and research.[21]

Fortune reported that in 2021 the pair tested the stock-market idea in a Washington, D.C. Airbnb. They encoded daily S&P 500 prices as electrical patterns, fed them to a dish of living neurons and recorded the response. Pomeraniec called the experiment "the first proof point."[10] Ksendzovsky told GamesBeat he concluded in 2022 that the idea had become practical, helped by longer-lived cell cultures and tools from [brain-computer interface](https://aiwiki.ai/wiki/brain_computer_interface) research and companies such as [Neuralink](https://aiwiki.ai/wiki/neuralink).[14] Upstarts Media dates the founding to 2022, and Wired reported that the company was founded in Baltimore "four years ago." The founders left academic careers and moved from Baltimore to San Francisco to build it.[20][12]

### Seed round and relaunch (February 2026)

On 12 February 2026 the company came out of stealth as The Biological Computing Co., replacing the name Biological Black Box.[14] It announced a $25 million seed round led by Primary and the opening of a flagship lab in San Francisco's Mission Bay neighborhood.[9] Fortune, which broke the funding story a day earlier, listed Builders VC, E1 Ventures, Proximity, Refactor Capital, Tusk Ventures and Wonder Ventures as participants.[10] At the time the company presented itself more broadly than video: the release described adapters that map decoded neural activity onto foundation models, plus an "Algorithm Discovery" platform for AI designs "beyond transformers."[9] SiliconANGLE reported a plan to launch hybrid neuro-silicon clusters for cloud environments in 2027.[11] Wired later reported that TBC had opened San Francisco offices and a small research lab the previous year.[12]

### AWS collaboration (September 2026)

On 22 September 2026, Wired, GamesBeat, Nikkei Asia and The Decoder reported a collaboration between TBC and Amazon Web Services. Wired said select AWS customers would get a limited preview of TBC's video model.[12][13][15][16] In its own release, TBC said it plans to run the model on [AWS Trainium](https://aiwiki.ai/wiki/aws_trainium), make it deployable through [Amazon SageMaker](https://aiwiki.ai/wiki/amazon_sagemaker) AI and sell it through AWS Marketplace. The company described that model as a text-to-video system built on an unnamed open-source base, with 80% lower inference cost and 5x faster generation.[17] Wired reported that the technology had previously been available only through a neocloud provider, Bluesky Compute, and that TBC would not yet name its base model.[12]

Wired also reported that Pomeraniec credited [Jeff Dean](https://aiwiki.ai/wiki/jeff_dean), a TBC investor, with first suggesting that the company try fine-tuning video generation models before applying its methods more broadly. Wired further reported that shortly after the seed round closed, the company raised another $25 million, taking its total to more than $50 million.[12] In a later announcement of Atish Banerjea, formerly chief information officer of Meta and NBCUniversal, as a strategic advisor, TBC gave its total funding as $52.5 million. It said the latest money was a $25 million seed extension led by Builders VC, with contributions from Wonder Ventures, Jeff Dean and Tusk Ventures.[18]

### Recognition and launch (October 2026)

On 15 October 2026, TBC announced that TIME had named its neuron-derived adapter technology to the magazine's Best Inventions of 2026 list.[19] On 30 October 2026, timed to "Mischief Night," it launched TBC mischief 1 on fal and OpenRouter.[2] Announcing the launch on X, Ksendzovsky wrote that the team had spent four years "brewing real brain cells on electrodes," and he called mischief 1 "the cheapest image-to-video model on @fal and @OpenRouter." He said the model would soon be on AWS Marketplace.[1]

## Technology

### Neural cultures

Wired reported that TBC works with rat brain cells and with neurons derived from human stem cells. The cells sit on multi-electrode silicon arrays made by the Swiss company 3Brain.[12] Ksendzovsky gave the same two sources, rats or human stem cells, in an EE Journal interview, and said that the company's gains so far were mostly in inference rather than training.[22] Nikkei Asia described roughly 100,000 neurons, some dissected from rat brains and others grown from human skin cells, fused to a grid of 4,096 electrodes in a single dish. The Decoder also gives the 4,096-electrode figure for TBC's cortical cultures.[15][16] Ksendzovsky told GamesBeat that the team can keep cultures alive for six months to more than a year. The cells sit in an incubator and their nutrient medium is changed every few days.[14] On its About page, TBC says its discovery work uses neural cultures, not whole brains, and that "there is no possibility for consciousness or sentience."[8]

### From recordings to adapters

TBC describes a three-step process: encode an AI problem as electrical stimulation patterns, record how the network responds, and turn useful patterns into algorithms. Researchers then test each modified model against the unmodified original.[6][7] Fortune summarized the same pipeline as growing neurons on a grid of electrodes, converting an image or video sequence into stimulation patterns, recording the resulting activity and turning that data into software.[10] Ksendzovsky told GamesBeat that "the tangible thing that comes out of the dish is a mathematical model that recapitulates how they process pixels," added to a [transformer](https://aiwiki.ai/wiki/diffusion_transformer) as an extra set of trainable parameters that transforms the latent representation used to predict the next frame.[13]

According to Ksendzovsky, the adapter lets TBC combine the model with "very aggressive optimization strategies," at the CUDA, numerical and model levels, without the loss of quality those methods usually cause. Pomeraniec said the company has many adapters, that "many of them fail," and that many others move onto an internal leaderboard.[13] TBC says its software is applied after training and wraps each diffusion transformer block.[7] The company argues that common efficiency techniques such as [quantization](https://aiwiki.ai/wiki/quantization), [distillation](https://aiwiki.ai/wiki/knowledge_distillation) and caching are shared across the industry, whereas its optimizations are "derived from biology, not just inspired by it."[2]

### Published research

The results TBC has reported so far appear as blog posts, which are not peer reviewed. The posts are a series it calls "a living document." All results below are self-reported.[23][24][25][26][27]

| Post | Base system | Reported result (TBC) |
|---|---|---|
| Biological Neural Dynamics for Computer Vision | MNIST and CIFAR images encoded as stimulation; spike-rate readout fed to a small CNN | On [MNIST](https://aiwiki.ai/wiki/mnist), 4.7% average accuracy gain over binarized images after ten epochs; on CIFAR, similar final accuracy (about 43.5%) but about three times faster convergence[23] |
| Using Neural Connectivity Patterns To Improve Image Reconstruction Quality | ViT-based [VAE](https://aiwiki.ai/wiki/variational_autoencoder) of the Oasis Minecraft model | 27% lower reconstruction loss with 2.36 million added parameters (about 1% of the 229-million-parameter autoencoder)[24] |
| Bridging Biology and Silicon: The Neural Dynamics Adapter | Oasis world model ([DiT](https://aiwiki.ai/wiki/diffusion_transformer) frame predictor) | About 156,000-parameter adapter; about 19% better long-rollout image quality than base Oasis, about 15% better than a parameter-matched fine-tune and about 5% better than [LoRA](https://aiwiki.ai/wiki/lora)[25] |
| A Sense of Space and Time: Optimizing Interactive Video Generation | Oasis, served live and interactively | "Hypercolumn Neural Optimizer" adapter of about 14 million parameters (about 3% of the base); about 2 to nearly 10 frames per second, 4.4 times the original serving path's throughput, with quality above the unoptimized base[26] |
| Beyond Weight Recycling: Algorithm Discovery for Sustained Plasticity | [Continual learning](https://aiwiki.ai/wiki/continual_learning) benchmark | A local plasticity rule reaching about 97% accuracy on new tasks, against 96% for Continual Backpropagation from [Richard Sutton](https://aiwiki.ai/wiki/richard_sutton) and colleagues[27] |

The Oasis experiments use the open interactive Minecraft world model released by [Decart](https://aiwiki.ai/wiki/decart_ai) and Etched (the public checkpoint is named Oasis 500M; TBC describes the model as having 600 million parameters).[25][32] According to The Decoder, an outside test by Bluesky Compute in July 2026 found about 4.4 times lower inference cost and more than three times as much coherent video as the base model.[16]

## TBC mischief 1

**TBC mischief 1** is an image-to-video model. TBC calls it "the world's first publicly available neuron-derived AI video model."[2] The launch release says it is built on Wan 2.2, an open-weight video model, and "matches its quality while generating video 5x faster and at a fraction of the cost."[2] Wan 2.2's weights and inference code were released by the Wan team (Wan-AI) on 28 July 2025 under the Apache 2.0 license (see [Wan 2.1](https://aiwiki.ai/wiki/wan_2_1) and [Wan 2.5](https://aiwiki.ai/wiki/wan_2_5) for the model family).[31] The fal API documentation describes the prompt field as steering "WAN."[33]

### Specifications and pricing

| Item | Detail (as listed on 1 November 2026) |
|---|---|
| Endpoints | fal: `tbc/mischief-1/image-to-video`; OpenRouter: `tbc/mischief-1`[3][4] |
| Published | 30 October 2026 on both marketplaces[3][4] |
| Inputs | Text prompt (up to 2,000 characters) plus a first-frame image[33][4] |
| Resolution | 480p or 720p, 16:9 landscape[33][4] |
| Clip length | 3 or 5 seconds, rendered at 16 fps (49 or 81 frames)[33] |
| Audio | None on OpenRouter (generate_audio false)[4] |
| Price | $0.005 per second at 480p; $0.01 per second at 720p (a 5-second clip costs $0.025 or $0.05)[3][4] |
| OpenRouter provider | Ambient[4] |

The fal endpoint can rewrite the prompt against the input image before rendering ("enhanced" mode, the default), and the documentation says TBC's policy service screens every prompt and image.[33]

### Claims

TBC's product and technology pages claim 5x faster generation and 80% lower inference cost compared with the original open-weight model. They also claim the same quality as the base model.[5][6] The product page shows an animated comparison in which Wan 2.2 takes 142 seconds per take and mischief 1 takes 12 seconds. The page labels these as average times from TBC's GPU benchmarks and notes that actual times vary.[5] The launch release quotes Justin Hackney, co-founder of the AI studio Wonder Studios, which tested the model in production work.[2] TBC says it is applying the same process to newer open-weight video models, and to language models and [physical AI](https://aiwiki.ai/wiki/physical_ai).[2]

## Reception and analysis

The company's technical claims have drawn caveats in coverage. Writing about the AWS announcement, The Decoder noted that TBC had not named the base model, hardware, resolution, clip length or quality metric behind its headline figures. It also pointed out that "five times faster and 80 percent cheaper are mathematically the same thing," probably one effect stated in two units, and said TBC had not published a controlled comparison against purely digital methods such as distillation or caching on the same base model.[16] Deap Ubhi, global director of technology for startups at AWS, told Wired that TBC takes a "pragmatic approach" by improving existing visual models instead of replacing the transformer. He also said it remained to be seen whether the gains hold for much longer videos.[12] In SiliconANGLE, analyst Holger Mueller of Constellation Research called the idea "compelling" but said the company must show that it scales.[11]

## Background: biological computing

TBC belongs to a small field of companies and labs that use living neurons for computation. In 2022, Brett J. Kagan and colleagues at [Cortical Labs](https://aiwiki.ai/wiki/cortical_labs) reported in *Neuron* that human or rodent neurons on a high-density multi-electrode array (the DishBrain system) showed apparent learning within five minutes of playing a simulated version of Pong.[28] In 2023, Lena Smirnova and colleagues coined the term "organoid intelligence" in *Frontiers in Science* for biocomputing with human stem-cell-derived brain organoids.[29] In 2024, the Swiss company FinalSpark described a remotely accessible "Neuroplatform" that had run experiments on more than 1,000 brain organoids.[30] Wired noted that Cortical Labs also builds its technology on AWS.[12]

TBC differs from these efforts mainly in its product strategy. Instead of selling access to living cells, it sells software for existing models and treats the cultures as a research instrument. Its stated long-term goal is "real-time biological computing," in which neurons and silicon compute together.[7][8] Ksendzovsky told Fortune that within five to ten years the company wants biological networks to be part of the compute circuit.[10] Unlike [neuromorphic computing](https://aiwiki.ai/wiki/neuromorphic_computing) chips, TBC's current products need no special hardware; they run on standard AI accelerators.[7]

## See also

- [Cortical Labs](https://aiwiki.ai/wiki/cortical_labs)
- [Neuromorphic computing](https://aiwiki.ai/wiki/neuromorphic_computing)
- [AI video generation](https://aiwiki.ai/wiki/ai_video_generation)
- [Text-to-video generation](https://aiwiki.ai/wiki/text_to_video)
- [Inference optimization](https://aiwiki.ai/wiki/inference_optimization)
- [fal.ai](https://aiwiki.ai/wiki/fal_ai)
- [OpenRouter](https://aiwiki.ai/wiki/openrouter)

## References

1. Ksendzovsky, Alex. Post on X announcing mischief 1, 30 October 2026. https://x.com/AlexKsendzovsky/status/2116153410101944352
2. The Biological Computing Co. "The Biological Computing Co. Launches World's First Publicly Available Neuron-Derived AI Video Model." Press release, 30 October 2026. https://www.tbc.co/press-releases/neuron-derived-ai-video-model
3. fal. "tbc/mischief-1/image-to-video" model page and model registry entry, accessed 1 November 2026. https://fal.ai/models/TBC/mischief-1/image-to-video
4. OpenRouter. "TBC: mischief 1." Model page, accessed 1 November 2026. https://openrouter.ai/tbc/mischief-1
5. The Biological Computing Co. "Product: TBC mischief 1." Accessed 1 November 2026. https://www.tbc.co/product
6. The Biological Computing Co. "Technology." Accessed 1 November 2026. https://www.tbc.co/technology
7. The Biological Computing Co. Homepage and FAQ. Accessed 1 November 2026. https://www.tbc.co/
8. The Biological Computing Co. "About." Accessed 1 November 2026. https://www.tbc.co/about
9. The Biological Computing Co. "The Biological Computing Co. (TBC) Secures $25 Million Seed Round, First to Deploy Neuron-Based Alternative to Silicon AI." PR Newswire, 12 February 2026. https://www.prnewswire.com/news-releases/the-biological-computing-co-tbc-secures-25-million-seed-round-first-to-deploy-neuron-based-alternative-to-silicon-ai-302685429.html
10. Garfinkle, Allie. "Two neurosurgeons just raised $25 million betting brain cells can (someday) outcompute silicon." Fortune, 11 February 2026. https://fortune.com/2026/02/11/two-neurosurgeons-raised-25-million-betting-brain-cells-can-someday-outcompute-silicon/
11. Wheatley, Mike. "AI startup The Biological Computing Co. raises $25M to swap out silicon for lab-grown brains." SiliconANGLE, 12 February 2026. https://siliconangle.com/2026/02/12/ai-startup-biological-computing-co-raises-21m-swap-silicon-living-lab-grown-neurons/
12. Goode, Lauren. "AI Models Built From Rat Brains Just Got Closer to Reality." Wired, 22 September 2026 (updated 22 and 28 September 2026). https://www.wired.com/story/ai-models-built-from-rat-brains-are-about-to-become-a-reality/
13. Takahashi, Dean. "The Biological Computing Co. teams up with AWS to improve data center AI with brain-cell research." GamesBeat, 22 September 2026. https://gamesbeat.com/the-biological-computing-co-teams-up-with-aws-to-improve-data-center-ai-with-brain-cell-research/
14. Takahashi, Dean. "How The Biological Computing Co. will speed AI processing with human brain cells." GamesBeat, 22 April 2026. https://gamesbeat.com/the-biological-computing-company-will-speed-ai-with-human-brain-cell-research/
15. Yu, Yifan. "Amazon taps TBC to make AI models faster, cheaper with human neurons." Nikkei Asia, 22 September 2026. https://asia.nikkei.com/business/technology/artificial-intelligence/amazon-taps-tbc-to-make-ai-models-faster-cheaper-with-human-neurons
16. Schreiner, Maximilian. "A tiny software layer from lab-grown neurons promises faster, cheaper AI video." The Decoder, 22 September 2026. https://the-decoder.com/a-tiny-software-layer-from-lab-grown-neurons-promises-faster-cheaper-ai-video/
17. The Biological Computing Co. "TBC partners with AWS to bring the world's first neuron-derived AI video model to market." Press release, 23 September 2026. https://www.tbc.co/post/aws-partnership
18. The Biological Computing Co. "Former Meta and NBC Universal CIO Atish Banerjea joins TBC as Strategic Advisor." Press release (TBC dateline September 2026; PR Newswire, 5 October 2026). https://www.tbc.co/press-releases/atish-banerjea-joins-tbc
19. The Biological Computing Co. "The Biological Computing Co. (TBC) Named to TIME's Best Inventions of 2026." Press release, 15 October 2026. https://www.tbc.co/press-releases/tbc-named-to-times-best-inventions-of-2026
20. Konrad, Alex. "Using Living Brain Cells To Train AI, With TBC's Alex Ksendzovsky." Upstarts Media, 12 June 2026. https://www.upstartsmedia.com/p/podcast-tbc-alex-ksendzovsky
21. Primary. "The Biological Computing Company raises $25M Seed." https://www.primary.vc/articles/the-biological-computing-company-raises-25m-seed
22. Dalton, Amelia. "Bio-Inspired AI: How The Biological Computing Co. is Using Living Neurons to Supercharge Generative Models." EE Journal Fish Fry, 4 September 2026. https://www.eejournal.com/fish_fry/bio-inspired-ai-how-the-biological-computing-co-is-using-living-neurons-to-supercharge-generative-models/
23. The Biological Computing Co. "Biological Neural Dynamics for Computer Vision." Blog. https://www.tbc.co/post/biological-neural-dynamics-for-computer-vision
24. The Biological Computing Co. "Using Neural Connectivity Patterns To Improve Image Reconstruction Quality." Blog. https://www.tbc.co/post/using-neural-connectivity-patterns-to-improve-image-reconstruction-quality
25. The Biological Computing Co. "Bridging Biology and Silicon: The Neural Dynamics Adapter." Blog. https://www.tbc.co/post/the-neural-dynamics-adapter
26. The Biological Computing Co. "A Sense of Space and Time: Optimizing Interactive Video Generation." Blog. https://www.tbc.co/post/a-sense-of-space-and-time-optimizing-interactive-video-generation
27. The Biological Computing Co. "Beyond Weight Recycling: Algorithm Discovery for Sustained Plasticity in Neural Networks." Blog. https://www.tbc.co/post/beyond-weight-recycling-algorithm-discovery-for-sustained-plasticity-in-neural-networks
28. Kagan, B. J., Kitchen, A. C., Tran, N. T., Habibollahi, F., et al. "In vitro neurons learn and exhibit sentience when embodied in a simulated game-world." Neuron 110(23): 3952-3969.e8, December 2022. https://doi.org/10.1016/j.neuron.2022.09.001
29. Smirnova, L., Caffo, B. S., Gracias, D. H., Huang, Q., et al. "Organoid intelligence (OI): the new frontier in biocomputing and intelligence-in-a-dish." Frontiers in Science 1, 28 February 2023. https://doi.org/10.3389/fsci.2023.1017235
30. Jordan, F. D., Kutter, M., Comby, J.-M., Brozzi, F., Kurtys, E. "Open and remotely accessible Neuroplatform for research in wetware computing." Frontiers in Artificial Intelligence 7, 2 May 2024. https://doi.org/10.3389/frai.2024.1376042
31. Wan-Video. "Wan2.2" GitHub repository and Hugging Face model card (Wan-AI/Wan2.2-I2V-A14B, created 24 July 2025, Apache 2.0). https://github.com/Wan-Video/Wan2.2
32. Decart and Etched. "Oasis 500M" (open-oasis) GitHub repository. https://github.com/etched-ai/open-oasis
33. fal. OpenAPI schema for tbc/mischief-1/image-to-video, accessed 1 November 2026. https://fal.ai/api/openapi/queue/openapi.json?endpoint_id=tbc/mischief-1/image-to-video