Flourish (AI lab)

Not to be confused with Flourish, the data visualization tool acquired by Canva.

Flourish is a New York-based artificial intelligence research company that pairs an in-house neuroscience laboratory with an AI engineering team, with the aim of finding what Wired called the brain's "core algorithm" and using it to build AI that keeps learning after deployment and runs on a small power budget.^[1][2] It was founded in 2025 by Thomas Reardon, a neuroscientist who earlier co-founded the neural-interface startup CTRL-labs, and Rob Williams, a former Amazon devices executive.^[2][6] Its founding pitch called it "a neuro AI company that is solving the two most difficult problems facing AI today: power efficiency and continuous learning," and named the system it intends to build Cortex AI.^[1] Flourish came out of stealth on 4 June 2026, when Wired reported that it had raised $500 million at a reported valuation of $2.5 billion from investors including Jeff Bezos, Lux Capital, GV and Catalio Capital Management.^[1][2] Coverage of the fundraising described the company as having no commercial product,^[17] and as of October 2026 its website carried a one-line description ("Flourish is a Neuro AI company.") and a link to the Wired story.^[9]

Overview

History

Founding

Williams spent 12 years at Amazon, latterly as vice president of device software and services, and was a member of the S-team, the senior leadership group that advises the company's chief executive.^[7] In October 2025 he said he would retire at the end of the year; an internal memo from devices chief Panos Panay said Williams would stay on as an adviser and S-team member until then.^[7] Wired described him as having been "in charge of software products such as Alexa" before his departure that fall.^[1] Reardon, who had known Williams since they both worked at Microsoft, persuaded him to join the new company.^[1]

In December 2025 Williams pitched Bezos, this time approaching him as a funder rather than as a boss. He used the Amazon format of a press release written as if the product had already been built, and watched over Zoom while Bezos read it.^[1] After reading the two-page proposal Bezos put in $50 million, and he later almost doubled his initial stake. According to Williams, the one thing Bezos needed to know was whether the founders were committed to spending years on the project. Bezos did not respond to Wired's requests for comment.^[1]

The New York venture firm Catalio Capital Management says it co-founded Flourish and co-led its seed financing. Catalio co-founder George Petrocheilos told citybiz that "the concept behind Flourish was derived from the U.S. intelligence community's flagship NeuroAI research program called MICrONS, which Catalio co-founder and managing partner, Dr. Jacob Vogelstein, conceived and led prior to launching Catalio."^[4] Jacob Vogelstein's brother Joshua T. Vogelstein, a Johns Hopkins professor and Catalio venture partner, is a Flourish co-founder.^[4][11] Wired describes Jacob Vogelstein as an investor in and adviser to the company.^[1]

Building the lab

By the end of March 2026 Reardon had hired around two dozen neuroscientists and AI researchers. Wired visited on the day the company moved into offices in New York's West SoHo area, in a 10-story building with a built-in data center; on moving day the lab equipment, including electron microscopes, had not yet arrived.^[1] An early recruit was Greg Wayne, a longtime researcher at Google DeepMind who, according to Wired, heads Project Astra. Wired reported that DeepMind chief executive Demis Hassabis fought to keep him, and that the two agreed Wayne would keep his DeepMind job while spending 20 percent of his time at Flourish as a senior advisor.^[1]

In early May 2026 the scientists held an all-hands meeting in the New York office to debate six candidate experiments, each requiring multimillion-dollar microscopy equipment and years of work. Topics ranged from how rabies spreads through the cortex to the neurobiology of birdsong, and from whether to study molecules and synapses or larger cells and circuits to whether mouse connectomes would suffice or human tissue was needed.^[1] Sean Bittner, a computational neuroscientist who had worked with Reardon at Meta, summarized the outcome: "The takeaway is that we want to do data collection across the nano, micro, and meso scales to support the discovery of the core algorithm."^[1]

Funding

On 30 April 2026 Bloomberg reported that Reardon was raising money for Flourish and that the company was in talks with investors at a $2.5 billion valuation, adding that the terms could change. It said Lux Capital, where Reardon is a venture partner, and GV had agreed to be major investors.^[3] The Next Web reported the target as about $500 million and, citing Bloomberg, said Lux and GV were expected to lead.^[17]

Wired disclosed the completed financing on 4 June 2026: $500 million in total and a reported valuation of $2.5 billion, with Bezos, Lux Capital, GV and Catalio among the investors.^[1] SiliconANGLE, summarizing the Wired report, said about a fifth of the money came from Bezos and the rest from a consortium including GV, Lux Capital and Catalio, which it described as a healthcare-focused fund.^[2] AlleyWatch listed the round as led by Lux Capital and GV with participation from Catalio and Bezos,^[6] while Catalio said it co-led Flourish's seed financing.^[4] Crunchbase News called it "initial funding" and counted it among three $500 million rounds tied for second-largest among announced rounds by US-based companies from 30 May to 5 June 2026.^[5]

Research approach

The efficiency and learning problem

Flourish's argument is that current AI has hit two limits at once. Reardon told Wired that large language models are greedy consumers of computing power and data, and Wired added that once trained, such models stop learning.^[1] Wired noted that a person uses about 20 watts to process information while a single chip in an AI training cluster uses more than 30 times that amount, and that hyperscale AI requires thousands of chips and gigawatts of energy;^[1] SiliconANGLE reported the company's own comparison as a server-grade graphics card using about 30 times more energy than the brain, and said Flourish hopes to cut that power draw by more than an order of magnitude.^[2] Reardon described the goal as "a synthetic artificial intelligence brain that runs on 50 watts or less." On data efficiency he said: "There's something fundamentally wrong with saying, 'I need to basically read every book ever written 20 times over in order to learn English,'" adding that "a human baby does it with a couple hundred thousand utterances."^[1] The company's interest in systems that adapt after training places it in the research area usually called continual learning.

Cortical columns and wet-lab neuroscience

Flourish's scientists are focusing on cortical columns, which one Flourish scientist called "the canonical computational unit" of the brain.^[1] The plan is for the neuroscience team to run original wet-lab experiments, with electron microscopes among the planned equipment, while an algorithms team builds models informed by the results; Wired reported that the algorithms team might in turn find clues that help the neuroscientists.^[1][2] "You don't really know if you understand something until you can build it, implement it in silicon," Flourish neuroscientist Josh Morgan told Wired. Staff said they are open to publishing some of their original research.^[1] Wayne put the premise as: "The brain has a secret we haven't found yet."^[1]

Timeline and near-term products

The founders frame the project as long-range. Williams told Wired that "the way you make big differences is to plan for things that have value seven to 10 years out," while Reardon said he hopes Flourish will have "its big solution" in five years.^[1]

Reardon also described nearer-term revenue paths. He told Wired that the team is developing a hippocampus-inspired way of handling memory that would let its models learn without extensive training data, that the algorithms team has built a model that can learn continuously and is working to run it on "the kinds of devices you carry in your pocket", and that he was negotiating with a major chip manufacturer to put the model on silicon.^[1] SiliconANGLE reported the chipmaker as unnamed and described the memory work as a system meant to reduce the data needed for training.^[2] As of October 2026 the company's website listed no products.^[9]

Research lineage

MICrONS

MICrONS (Machine Intelligence from Cortical Networks) was a research program of the Intelligence Advanced Research Projects Activity (IARPA) that, in IARPA's words, "aimed to close the performance gap between human analysts and automated pattern recognition systems by reverse-engineering the algorithms of the brain." IARPA says the program assembled a multi-petabyte dataset of co-registered neurophysiological and neuroanatomical data spanning one cubic millimeter of mammalian brain, and that its performers delivered neuroscience-informed pattern-recognition algorithms with improved robustness to noise.^[12] The project's data portal thanks David A. Markowitz and Jacob Vogelstein as the program managers.^[13] Catalio says Jacob Vogelstein "conceived and led" MICrONS;^[4] the acknowledgments of the 2025 Nature paper credit him and Markowitz with co-developing the program and Markowitz with coordinating the work through all three of its phases.^[14] The main MICrONS dataset was published in Nature in April 2025: an electron-microscopy reconstruction of mouse visual cortex containing more than 200,000 cells and 0.5 billion synapses, co-registered with calcium imaging of around 75,000 neurons, with accompanying studies that included a synapse-level connectivity diagram of a cortical column.^[14] Jacob Vogelstein, his brother Joshua and others also started the Open Connectome Project, which Jacob described to Wired as an effort to "collect all these images of the brain and start to do data processing on them to try to interpret the circuits."^[1]

Connectome-derived networks

Joshua Vogelstein is an associate professor of biomedical engineering at Johns Hopkins University.^[11] Wired reported that he had recently co-authored a paper on the neural network of a fruit fly that found it to be 10 times more efficient than the transformer.^[1] Both Vogelsteins are co-authors of a July 2025 preprint, "Biological Processing Units," that converts the complete connectome of the Drosophila larva brain (about 3,000 neurons) into a fixed recurrent network. The authors report 98% accuracy on MNIST and 58% on CIFAR-10, above size-matched multilayer perceptrons, and, on the ChessBench chess dataset, a model trained on 10,000 games reaching 60% move accuracy, which they describe as "nearly 10x better than any size transformer."^[15]

Memory in neural networks

Wayne's earlier research includes the 2014 paper "Neural Turing Machines," written with Alex Graves and Ivo Danihelka, which coupled neural networks to external memory that they read and write through attention, trained end to end with gradient descent.^[16]

People

Facebook did not disclose what it paid for CTRL-labs; TechCrunch reported at the time that Bloomberg put the price between $500 million and $1 billion.^[8] SiliconANGLE notes that CTRL-labs' technology underpins the Meta Neural Band, a wristband for controlling Meta's smart glasses with hand gestures.^[2]

Related efforts

Wired placed Flourish among several attempts to draw on biology for AI. It noted that IBM and Intel have both released neuromorphic chips inspired by the brain's architecture, that Cortical Labs combines lab-grown neurons with silicon chips, that Sam Altman backs Merge Labs, that Meta's superintelligence group says its TRIBE v2 model "acts as a digital twin of human neural activity," and that Unconventional AI is creating a grant program for research on AI that replicates biological efficiency.^[1] Reardon's view, as Wired reported it, is that Flourish's edge lies in the strength of its neuroscience staff.^[1] Recht, the Berkeley adviser, said of today's models: "They call those neural nets, but there's nothing brain-like happening there."^[1]

Reception

Wired called the company's plan "a risky, long-range bet."^[1] Recht told Wired: "I'm not convinced that it's going to work. But if it does, it would be amazing."^[1] Wayne told Wired: "I didn't know if they could achieve their goal, but I thought it would lead to interestingness, which probably will be useful."^[1] Writing on 1 May 2026, while the round was still being negotiated, The Next Web's Alina Maria Stan described the risk as Reardon "selling a thesis at a frontier-lab valuation," noting that $2.5 billion was being asked for a company that had not demonstrated a working model, a benchmark or a product.^[17]

Debarghya Das's "The Ultimate Neolab List" (updated 24 October 2026) includes Flourish, filed as a "Frontier lab" with the sub-label "Brain-inspired AI". On the list's four-step scale for whether a company fits his neolab definition, Das rates it "Mostly," calling it "researcher-led, no revenue, huge early raise" but noting that it "runs a wet neuroscience lab and may build chips, so not pure software."^[18]

See also

• List of neolabs
• Neolab
• Neuromorphic computing
• Continual learning
• Meta Neural Band
• Cortical Labs
• Merge Labs
• Unconventional AI
• AI energy consumption

References

1. ^1 ^2 ^3 ^4 ^5 ^6 ^7 ^8 ^9 ^10 ^11 ^12 ^13 ^14 ^15 ^16 ^17 ^18 ^19 ^20 ^21 ^22 ^23 ^24 ^25 ^26 ^27 ^28 ^29 ^30 ^31 ^32 ^33 ^34 ^35 ^36 ^37 ^38 ^39 ^40 ^41 ^42 ^43 ^44 ^45 ^46 ^47Levy, Steven. "Jeff Bezos Is Funding a Wild Hunt for the Brain's 'Core Algorithm'". Wired, 4 June 2026.
2. ^1 ^2 ^3 ^4 ^5 ^6 ^7 ^8 ^9 ^10Deutscher, Maria. "AI startup Flourish reportedly raises $500M round backed by Jeff Bezos". SiliconANGLE, 4 June 2026.
3. ^1 ^2 ^3Swetlitz, Ike, and Natasha Mascarenhas. "Startup Bringing Brains to AI Aims for $2.5 Billion Valuation". Bloomberg (via Bloomberg Law), 30 April 2026.
4. ^1 ^2 ^3 ^4 ^5 ^6"Catalio's Neuroscience Startup Flourish Emerges With Funding from Bezos, Google Ventures". citybiz, 5 June 2026.
5. ^1 ^2Glasner, Joanna. "The Week's 10 Biggest Funding Rounds: Megarounds Proliferate, Led By Enterprise Software, AI, And Space Tech". Crunchbase News, 5 June 2026.
6. ^1 ^2 ^3 ^4"The AlleyWatch Startup Daily Funding Report: 6/5/2026". AlleyWatch, 5 June 2026.
7. ^1 ^2 ^3"Amazon Devices Executive Robert Williams to Retire". PYMNTS, 3 October 2025.
8. ^1 ^2Matney, Lucas. "Facebook buys startup building neural monitoring armband". TechCrunch, 23 September 2019.
9. ^1 ^2 ^3Flourish (company website). Flourish, retrieved 27 October 2026.
10. ^1 ^2 ^3Flourish company page. LinkedIn, retrieved 27 October 2026.
11. ^1 ^2 ^3 ^4"Joshua T. Vogelstein, Ph.D.". Catalio Capital Management team page, retrieved 27 October 2026.
12. ^"MICrONS: Machine Intelligence from Cortical Networks". IARPA, archived 1 January 2025.
13. ^1 ^2"Acknowledgments". MICrONS Explorer, retrieved 27 October 2026.
14. ^1 ^2The MICrONS Consortium. "Functional connectomics spanning multiple areas of mouse visual cortex". Nature 640, 435-447, April 2025. doi:10.1038/s41586-025-08790-w.
15. ^Yu, Siyu; Qin, Zihan; Liu, Tingshan; Xu, Beiya; Vogelstein, R. Jacob; Brown, Jason; Vogelstein, Joshua T. "Biological Processing Units: Leveraging an Insect Connectome to Pioneer Biofidelic Neural Architectures". arXiv:2507.10951, 15 July 2025.
16. ^Graves, Alex; Wayne, Greg; Danihelka, Ivo. "Neural Turing Machines". arXiv:1410.5401, 20 October 2014.
17. ^1 ^2 ^3 ^4Stan, Alina Maria. "The man who built Internet Explorer wants to teach AI to think on 20 watts". The Next Web, 1 May 2026.
18. ^Das, Debarghya. "The Ultimate Neolab List". Updated 24 October 2026.