A San Francisco startup that grows real neurons in petri dishes says it has found a way to make AI video generation cheaper and faster, and it just got Amazon Web Services to back the bet. The Biological Computing Co. (TBC) announced on September 22, 2026 that it is partnering with AWS to bring what the two companies call the world’s first neuron-derived AI video model to market, according to an official AWS press release.
The pitch sounds like science fiction: living brain cells, grown and studied in a lab, feeding data into a software layer that then makes conventional AI video generation run faster and cheaper. TBC says the layer, built from measurements of living neural activity, delivers up to 5x faster generation and cuts inference costs by 80% compared to the underlying open-source video model it modifies. No customer ever touches a living cell. The neurons stay in TBC’s lab; only the software patterns they inspired ship to production, running on ordinary AWS infrastructure.
It’s a striking claim, and one that arrives at an unusually competitive moment for AI video. OpenAI is winding down its Sora API on September 24, 2026, just two days after TBC’s announcement, according to reporting cited by CapCut’s Dreamina video AI roundup. Google’s Veo lineup and Runway’s Gen-4.5 are fighting for enterprise and creator budgets alike. Into that scrum steps a company whose core technology isn’t a bigger transformer or more training compute, but biology.
What TBC and AWS Actually Announced
The announcement, dated September 22, 2026 and made from San Francisco, describes TBC as an applied biological computing company that uses real neurons to improve conventional AI models. The product at the center of the news is a text-to-video model built on an existing open-source video-generation system. TBC has not named which base model it modified, and reporting from the-decoder.com notes that the company has kept that detail private, along with its benchmark methodology.
What TBC has disclosed is the shape of the product: a proprietary software layer, derived from data collected by growing and studying living neurons, that sits on top of the base video model. TBC says this layer adds less than 0.1% to the size of the underlying model, a strikingly small footprint for a claimed 5x speed gain. The company frames this as proof that biological systems encode efficiency patterns that decades of silicon-first AI research haven’t replicated, echoing the framing AWS itself used in its own announcement.
Alex Ksendzovsky, CEO and co-founder of The Biological Computing Co., described the milestone as commercial validation of a research program that has run for several years. “Our partnership with AWS takes neuron-derived AI optimization to commercial scale,” he said, in comments included in the AWS press release. In a separate interview with GamesBeat published the same day, he added that “the biological computing revolution is just beginning.”
How a Neuron-Derived Software Layer Is Supposed to Work
TBC’s core technique is not new to this announcement. The company has spent years growing neurons in culture dishes, stimulating them, and recording how they respond, then translating those response patterns into mathematical structures that can be applied to conventional neural networks. Coverage of TBC’s earlier research describes this as an attempt to borrow efficiency tricks that biological brains evolved over millions of years, rather than trying to discover them through brute-force machine learning experimentation.
For this AWS-backed product, TBC applied that same approach specifically to video generation, a task the company’s leadership says its biologically-derived algorithms happen to handle well. Ksendzovsky told GamesBeat that “one of the things that these algorithms are good at handling are video generation tasks,” pointing to why TBC chose video as its first commercial application rather than text or image generation, where dozens of competitors already crowd the field.
Crucially, none of this requires a customer to run biological hardware. The neurons stay in TBC’s laboratory. What ships to AWS customers is conventional code: a lightweight layer that modifies how the base video model allocates compute during generation. That distinction matters, because it means TBC’s product can be deployed on standard cloud infrastructure without new hardware procurement, a detail that likely made the AWS partnership possible in the first place.
The Performance Numbers TBC Is Publishing
TBC’s headline claims are a 5x improvement in generation speed and an 80% reduction in inference cost relative to the unmodified base model, alongside what the company describes as improved output quality. The table below summarizes the claims as published, alongside what independent parties have and haven’t confirmed.
| Metric | TBC’s Claim | Independent Verification |
|---|---|---|
| Generation speed | Up to 5x faster than base model | Not independently benchmarked |
| Inference cost | 80% lower than base model | Not independently benchmarked |
| Software layer size | Adds less than 0.1% to base model | Company-reported, unverified |
| Output quality | Improved vs. base model | No published side-by-side results |
| Hardware requirement | Runs on conventional AWS infrastructure | Consistent across all reporting |
| Earlier OASIS proof of concept | 2x quality gain, ~4.4x lower cost, 3x more coherent video | Reported under TBC’s own specified test conditions |
That last row matters for context. Before this AWS deal, TBC had already published results from an earlier research effort it called OASIS, where it reported a twofold improvement on a selected video-quality benchmark, roughly 4.4x lower inference cost, and more than three times as much coherent video output compared to a base model, under conditions TBC itself defined. The AWS-backed product appears to be a commercialized, higher-performing successor to that earlier proof of concept, though TBC has not published a direct technical comparison between the two.
Why Independent Verification Is Still Missing
Every outlet that covered the announcement flagged the same gap: TBC has not published independent benchmarks for its 5x speed and 80% cost figures, and neither has AWS. The-Decoder’s coverage, headlined around the idea that “a tiny software layer from lab-grown neurons promises faster, cheaper AI video,” was explicit that the numbers come from TBC itself, measured against a base model TBC has not named, using a methodology TBC has not disclosed.
That’s not unusual for a young startup’s launch announcement, but it does mean the 5x and 80% figures should be read as marketing claims rather than confirmed results until third-party labs, academic researchers, or customer case studies test them independently. Buyers evaluating TBC’s product will likely want to run their own workload comparisons before committing budget, particularly since the underlying model TBC modified remains unidentified, which makes apples-to-apples comparison against known baselines difficult.
There’s also the question of durability. A software layer that adds less than 0.1% to a model’s size and still delivers a 5x speedup is an unusually strong ratio of overhead to gain. If validated, it would be a genuinely notable result in efficient inference research, an area where companies pursuing techniques like custom silicon and quantization have historically had to trade off model size, speed, and output quality against each other rather than improving all three simultaneously.
AWS’s Role: Trainium, SageMaker AI, and the Marketplace
AWS’s involvement goes beyond a co-branded press release. According to the companies’ announcement, TBC plans to run its optimized model on AWS Trainium chips, offer deployment through Amazon SageMaker AI, and distribute the product through AWS Marketplace. That’s a fairly standard path for AWS’s startup partnerships: give an early-stage company access to compute and an enterprise sales channel in exchange for building on AWS’s stack rather than a rival cloud.
Jason Bennett, Vice President and Global Head of Startups and Venture Capital at AWS, framed the deal in evolutionary terms. “Nature solved the computing efficiency problem billions of years ago. TBC’s insight is that we can learn from the original computer, the human brain, to make AI faster, more efficient, and more economical,” he said, according to the AWS press release. Bennett also pointed to the practical mechanics of the partnership, noting that “by building on the AWS AI stack and leveraging go-to-market channels like AWS Marketplace, TBC can move from discovery to commercial scale at startup speed.”
For AWS, backing TBC fits a broader pattern of the company positioning itself as the cloud where unconventional AI research gets commercialized, not just where mainstream large language models get served. It’s the same logic behind AWS’s agentic benchmarking efforts and its push into specialized inference hardware: differentiate on more than raw GPU rental by hosting technology rivals don’t have.
Limited Preview, Not a Public Launch
Despite the announcement’s ambitious framing, TBC’s video model is not broadly available. Reporting from Superpower Daily describes the rollout as a limited preview for select AWS customers, with both companies expecting a wider release to the full AWS enterprise customer base at an unspecified later date. No pricing has been published, no product name distinct from “TBC’s neuron-derived video model” has surfaced, and no general sign-up process has been confirmed in the reporting available.
That gap between announcement and availability is worth underlining, because it shapes how the market should read this news. This is a partnership announcement and a technology reveal, not a product launch with a price list. Enterprises interested in TBC’s technology will need to go through AWS’s startup and enterprise sales channels rather than signing up on a self-serve page, at least for now.
Competitive Landscape: How TBC Stacks Up Against Sora, Veo, and Runway
TBC enters a video-generation market that has consolidated around a handful of well-funded players, each competing on a different axis. OpenAI’s Sora line has focused on narrative scene construction and world simulation, though the company is shutting down the standalone Sora API on September 24, 2026. Google’s Veo 3.1 family offers image-to-video generation with synchronized audio and, according to a comparison published by Teamday, pricing around $0.10 per second of 720p output. Runway’s Gen-4.5 continues to target creative production workflows with an established editing ecosystem.
| Model | Maker | Primary Positioning | Key Differentiator |
|---|---|---|---|
| Neuron-derived video model | The Biological Computing Co. + AWS | Enterprise inference optimization layer | Claimed 5x speed, 80% lower cost via bio-derived software layer |
| Sora 2 | OpenAI | Narrative scene generation | Strong world simulation; standalone API being discontinued Sept. 24, 2026 |
| Veo 3.1 | Image-to-video with audio | Synchronized audio, multiple quality tiers, ~$0.10/sec at 720p per teamday.ai | |
| Gen-4.5 | Runway | Creative production workflows | Established editing tools, image-to-video support |
The comparison isn’t a clean one. TBC’s disclosed metrics concern inference speed and cost against its own base model, while public comparisons of Sora, Veo, and Runway tend to emphasize visual quality, audio support, clip length, and ease of use. TBC hasn’t published resolution limits, maximum clip duration, or audio capabilities for its product, so buyers can’t yet weigh its efficiency claims against feature parity with the established players. What TBC is selling, at least for now, looks less like a creator tool and more like an enterprise inference layer aimed at companies that already generate large volumes of video through existing pipelines and want to cut the compute bill.
Historical Context: Biological Computing’s Long Road to Commercial Product
Using living neurons to inform computing isn’t a brand-new idea. Research labs and startups working under the banner of “organoid intelligence” or “wetware computing” have spent the past several years growing small clusters of neurons and studying how they process signals, chasing the long-standing observation that biological brains do far more computation per watt than silicon chips. Most of that research has stayed in academic papers and early-stage lab demonstrations, without a clear commercial product attached.
TBC’s approach differs in one important respect: it doesn’t ship biological material or ask customers to run living cells. Instead, it treats the neurons purely as a research instrument, a way to discover efficiency patterns that are then translated into conventional software running on conventional chips. That’s a meaningfully more pragmatic path to market than biological hardware, which faces enormous regulatory, logistical, and reliability hurdles before it could ever run in a commercial data center. It also means TBC’s claims live or die on ordinary software benchmarking, not on unresolved questions about biological hardware safety or scalability.
The AWS deal marks the clearest signal yet that a major cloud provider is willing to bet distribution and infrastructure resources on this category of research, following AWS’s history of partnering with unconventional AI approaches through programs like its startup and venture capital initiatives. It’s a similar dynamic to how other AI labs have begun exploring biology-adjacent research as a source of algorithmic ideas rather than pure scaling of existing architectures.
Market Impact: What This Means for AWS’s AI Cloud Business
For AWS, the TBC partnership is a low-risk, high-narrative bet. The company isn’t committing to build biological computing infrastructure; it’s hosting a software product on hardware it already sells, in exchange for an exclusive-sounding announcement that reinforces AWS’s positioning as the cloud where novel AI research gets commercialized fastest. Coverage from Zhitong Finance, cited in Amazon stock news trackers, framed the deal in similar terms: an efficiency story attached to AWS’s existing Trainium and SageMaker AI product lines, rather than a new hardware category.
If TBC’s cost claims hold up under real customer workloads, the bigger implication is for the video-generation cost curve broadly. An 80% reduction in inference cost, even if it only applies to a specific class of video-generation tasks, would meaningfully change the unit economics for companies generating video at scale, whether for advertising, product visualization, or synthetic media pipelines. That’s the kind of efficiency gain cloud providers have chased for years through new GPU generations and custom accelerators. TBC’s pitch is that a comparable gain can come from a software layer instead, which, if true, would be a cheaper and faster path to the same economics than waiting on new silicon.
The counterpoint is that “if true” is doing a lot of work in that sentence. Cloud providers and enterprise buyers have been burned before by efficiency claims that didn’t survive contact with production workloads at scale. Until TBC publishes a reproducible benchmark or a named enterprise customer confirms real-world savings, the market impact remains speculative rather than measured.
How This Fits the Broader AI Video and Generative Media Market
TBC’s announcement lands in a video-generation market that has grown crowded and price-competitive over the past year. OpenAI’s decision to sunset the standalone Sora API on September 24, 2026, just two days after TBC’s news, suggests the company is consolidating its video tools elsewhere in its product lineup rather than exiting the category. Meanwhile, image-generation tools have also been racing on speed and cost, echoed in recent moves like OpenAI’s own faster image generation update, which similarly emphasized speed gains as a selling point over raw output quality improvements.
Against that backdrop, TBC’s bet is that enterprise buyers care more about inference economics than about chasing the most cinematic output. That’s a reasonable wager for a company entering a market dominated by better-funded, better-known incumbents: compete on cost and throughput rather than trying to out-produce Sora or Veo on visual fidelity. Whether that strategy works depends entirely on whether TBC’s claimed numbers hold up once customers outside the limited preview start publishing their own results.
What Analysts and Reporters Are Watching Next
Coverage of the announcement has converged on a handful of open questions. First, which open-source video model did TBC actually modify? Without that detail, comparisons to the base model’s original benchmarks are impossible to verify. Second, when does the limited preview expand, and to how many customers? Superpower Daily’s reporting suggests both companies expect a broader AWS enterprise rollout, but neither has committed to a date. Third, will TBC publish an independently reproducible benchmark, or continue to rely on company-reported figures?
There’s also a longer-term question about whether TBC’s neuron-derived approach generalizes beyond video. If the technique genuinely captures something useful about biological efficiency, it’s reasonable to expect TBC or competitors to try applying similar methods to other generative tasks, from image synthesis to language model inference. Nothing in the current announcement confirms plans for that expansion, but the underlying research program was never described as video-specific; video was simply the first commercial application TBC’s leadership chose to pursue.
Predictions: Where This Goes From Here
- Independent benchmarks will surface within months. Given the scale of TBC’s claims and AWS’s involvement, expect third-party researchers or early customers to publish their own speed and cost comparisons well before TBC does so voluntarily.
- The base model identity will leak. Reporters and competitors have strong incentives to identify which open-source video model TBC modified, and that detail rarely stays hidden for long once a product moves toward general availability.
- AWS will use this to court more unconventional AI startups. Expect AWS’s startup and venture arm to reference the TBC deal as a template for attracting research-stage companies working on efficiency techniques outside mainstream scaling approaches.
- Pricing will land closer to existing video APIs than to a dramatic undercut. Cost claims often compress once real infrastructure, support, and margin requirements enter the picture; expect TBC’s eventual public pricing to be competitive but not radically cheaper than Veo or Runway.
- Rivals will publicly question the neuron framing. Because TBC’s marketing leans heavily on the novelty of biological computing, expect competitors and skeptical researchers to push back on how much of the performance gain is genuinely attributable to neuron-derived insights versus conventional software optimization techniques repackaged with new branding.
Table: Key Facts at a Glance
| Detail | Confirmed Information |
|---|---|
| Company | The Biological Computing Co. (TBC), San Francisco |
| Partner | Amazon Web Services (AWS) |
| Announcement date | September 22, 2026 |
| Product | Neuron-derived text-to-video AI model |
| Base model | Unnamed open-source video-generation model |
| AWS infrastructure used | AWS Trainium, Amazon SageMaker AI, AWS Marketplace |
| Availability | Limited preview for select AWS customers |
| CEO | Alex Ksendzovsky, CEO and co-founder of TBC |
The Bottom Line
TBC’s neuron-derived video model is a genuinely novel pitch: use living brain cells as a research tool to discover efficiency patterns, then ship those patterns as ordinary software running on ordinary cloud infrastructure. AWS’s backing gives the idea real distribution muscle through Trainium, SageMaker AI, and the AWS Marketplace, and the company’s own executives are framing the deal as proof that biological insight can beat brute-force scaling on cost and speed.
But the headline numbers, 5x faster, 80% cheaper, remain TBC’s own figures, measured against a base model it hasn’t named, using a methodology it hasn’t published. That doesn’t make the claims false. It does mean the AI video market, and the enterprises TBC hopes to sell to, should treat September 22’s announcement as the opening chapter of a story that won’t be fully written until independent benchmarks and named customers weigh in.
Frequently Asked Questions
What is The Biological Computing Co. (TBC)?
TBC is a San Francisco startup described in its AWS announcement as an applied biological computing company that grows and studies real neurons to improve conventional AI models, then translates the resulting patterns into software.
Does TBC’s AI video model actually use living neurons to generate video?
No. The neurons remain in TBC’s laboratory. Customer video generation runs entirely on software derived from measurements of that neural activity, using conventional AWS infrastructure with no biological hardware involved.
How much faster and cheaper does TBC claim its model is?
TBC says its neuron-derived software layer delivers up to 5x faster video generation and 80% lower inference costs compared to the underlying open-source base model, along with improved output quality. These figures are company-reported and have not been independently verified.
What role does AWS play in this partnership?
AWS is the infrastructure and go-to-market partner. TBC’s model is set to run on AWS Trainium chips, deploy through Amazon SageMaker AI, and be distributed via AWS Marketplace, giving TBC access to AWS’s enterprise sales channels.
Can I sign up for TBC’s neuron-derived video model today?
Not broadly. Reporting describes the product as being in a limited preview for select AWS customers, with a wider rollout to AWS’s enterprise customer base expected but not yet dated or priced publicly.
Which open-source model did TBC build its video generator on?
TBC has not publicly identified the base model. This remains one of the most-asked open questions in coverage of the announcement, since it affects how the company’s performance claims can be independently evaluated.
How does TBC compare to Sora, Veo, or Runway?
Direct comparison is difficult because TBC has published efficiency metrics (speed and cost) rather than the quality, resolution, and feature metrics typically used to compare Sora, Veo, and Runway. TBC’s product currently looks positioned as an enterprise inference-cost play rather than a creator-facing video tool.
Is this the first time neurons have been used in AI research?
Growing neurons to study biological computing efficiency isn’t new; research in this area, sometimes called organoid or wetware intelligence, has existed for years. What’s new here is a commercial partnership with a major cloud provider to ship software derived from that research as a paid product.
