Flourish

Flourish is a New York neuro-AI startup, founded in 2024, betting that AI’s power and data-efficiency problems are architectural rather than a hardware-scaling problem; its founding thesis is that reverse-engineering how biological brains compute can produce AI that learns continuously and runs on a tiny fraction of current compute.¹ Founder Thomas Reardon frames the failure of large language models bluntly: “There’s something fundamentally wrong with saying, ‘I need to basically read every book ever written 20 times over in order to learn English.’ A human baby does it with a couple hundred thousand utterances.”²
Co-founder Thomas Reardon built the project that became Internet Explorer at Microsoft in the 1990s (delivering the first browser implementation of CSS and becoming a founding board member of the W3C), then left tech, earned a PhD in neuroscience and behavior from Columbia (with an M.S. from Duke), and in 2015 co-founded CTRL-labs, which built a non-invasive wrist-worn neural interface using surface electromyography (sEMG) to decode motor-neuron activity through the skin; Meta acquired CTRL-labs in 2019 for an estimated $500 million to $1 billion, and Reardon subsequently led neuromotor interface research (the Neural Band, now shipping in Meta’s smart glasses) inside Meta Reality Labs for six years before departing to found Flourish.³ ⁴
Co-founder Rob Williams is a former Amazon “S-team” senior executive who had previously worked with Reardon at Microsoft; he pitched Bezos in December 2025 using Amazon’s “working-backwards” practice of writing a press release for a product that doesn’t yet exist, opening with the line that Flourish “is a neuro AI company that is solving the two most difficult problems facing AI today: power efficiency and continuous learning.”² ⁵
Cortex AI is Flourish’s proposed system: rather than building new chips, the company is working at the software/architecture layer, aiming to identify the brain’s “canonical computational unit” — the cortical column — through original wet-lab neuroscience (electron-microscopy-based connectomics, i.e., mapping individual neurons and synapses) and then encode the resulting principles (sparsity, asynchronous/event-driven activation, hierarchical local processing, continuous on-device learning without a separate training phase) into new machine-learning architectures deployable on existing silicon.² ⁶
Flourish’s approach differs from prior neuromorphic hardware efforts like Intel Loihi, IBM TrueNorth, and SpiNNaker — those built custom spiking-neuron chips that never displaced GPU-based deep learning commercially; Flourish is explicitly not building specialized chips but instead targeting the algorithmic/architectural layer so that any efficiency gains could, in principle, run on commodity hardware, sidestepping the fabrication and software-ecosystem barriers that stalled earlier neuromorphic silicon.⁷ UC Berkeley’s Ben Recht, a Flourish adviser, notes that brain-inspired computing predates the LLM boom and is skeptical of the “neural net” label applied to transformers: “They call those neural nets, but there’s nothing brain-like happening there.”⁷
The company closed a $500 million round at a $2.5 billion valuation around June 4, 2026, roughly five weeks after fundraising conversations began; Jeff Bezos personally committed close to $100 million, having initially put in $50 million after reading Williams’s two-page pitch and then nearly doubling his stake, telling Reardon he would have put in more had they asked; other backers include Lux Capital and GV (Alphabet’s venture arm) — the same two firms that led CTRL-labs’ Series A a decade earlier — plus Catalio Capital.² ¹ ⁵
The Flourish round is Bezos’s third visible brain-related move within roughly a 12-month span: he holds a position in Synchron (via Bezos Expeditions, dating to its $75 million Series C in December 2022 and extended through a $200 million Series D on November 6, 2025, alongside Bill Gates’s Gates Frontier and ARCH Venture Partners) building implantable, non-surgical brain-computer interfaces; his family separately committed $100 million to the Allen Institute’s $400 million Brain Health Accelerator, announced June 2, 2026, targeting neurodegenerative disease; and now Flourish’s brain-inspired AI architecture — making him, per industry trackers, the only tech billionaire with capital spanning invasive/BCI hardware, basic brain-disease research, and brain-inspired AI algorithms simultaneously.⁸ ⁹ ¹⁰
Catalio Capital, a New York life-sciences investment firm co-founded in 2020 by George Petrocheilos and Dr. R. Jacob Vogelstein, is credited not just as an investor but as a company co-founder alongside Reardon and Williams; Jacob Vogelstein previously designed and ran MICrONS (Machine Intelligence from Cortical Networks), the IARPA-funded program that spent roughly $100 million over seven years to densely map a cubic millimeter of mouse visual cortex (200,000+ cells, 523 million synapses) specifically to “reverse-engineer” cortical algorithms for machine learning — direct intellectual lineage for Cortex AI’s connectomics research.¹¹ ¹² His brother, Dr. Joshua Vogelstein, a Johns Hopkins associate professor running the NeuroData/connectomics lab and a Catalio venture partner, is also listed as a Flourish co-founder, giving the company deep ties to the Johns Hopkins connectomics research network.¹³ ¹¹
Team composition blends AI research and computational neuroscience: early recruits include Greg Wayne, a longtime DeepMind researcher who led Google’s Project Astra, who joined with DeepMind CEO Demis Hassabis’s blessing to split his time (roughly 20%) between DeepMind and Flourish; UC Berkeley computer scientist Ben Recht serves as an adviser; and the company is actively recruiting computational neuroscientists to run original wet-lab experiments alongside machine-learning engineers, describing itself publicly as “a startup full of neuroscientists and AI researchers working together.”² ⁵
The stated power target for Cortex AI is 20–50 watts — roughly the draw of a high-end laptop — versus the human brain’s approximately 20-watt budget and the megawatt-scale power draw of frontier transformer training runs; Flourish cites a roughly 30x energy gap between a server-grade GPU and the brain for equivalent information processing, and aims to close more than an order of magnitude of that gap, which matters because AI’s growth is increasingly bottlenecked by electricity and data-center buildout rather than by algorithmic ideas alone.¹ ⁶
Reardon has stated the ultimate goal is “a synthetic artificial intelligence brain that runs on 50 watts or less” that adapts continuously like a human mind rather than being frozen after training; Flourish plans to ship nearer-term, more modest products derived from its research on the way to the larger Cortex AI ambition, though as of mid-2026 no launch date, customer, benchmark, or shipped product has been disclosed, and one report cites a founder estimate of 5–10 years to reach the full breakthrough while the company pursues parallel low-power model releases in the interim.² ¹⁴
Key risks and open questions: Reardon and Williams have publicly acknowledged they “haven’t figured out yet how to build a system that matches” the brain; connectomics has not yet produced a commercially viable computing paradigm despite decades of neuromorphic hardware attempts (Loihi, TrueNorth, SpiNNaker); critics note the cortical-column-as-universal-computational-unit hypothesis is not settled neuroscience, and that incremental transformer-efficiency gains (quantization, sparsity, mixture-of-experts, better chips) could shrink the efficiency gap Flourish is targeting before its architecture is ready; the $2.5 billion valuation for a pre-product, pre-revenue company prices Reardon’s track record and team pedigree rather than any demonstrated result.² ¹⁵ ⁷
Flourish’s core neuroscience infrastructure centers on an in-house electron-microscopy wet lab studying cortical columns and synaptic connectivity, directly building on techniques pioneered by the MICrONS consortium (Allen Institute for Brain Science, Baylor College of Medicine, Princeton University), which in April 2025 published the largest functional connectomics dataset to date — 2.5 miles of axons and 523 million synapses across 200,000 cells in mouse visual cortex — establishing the technical playbook (dense EM reconstruction paired with functional imaging) that Flourish’s research program appears to extend toward human-relevant, AI-oriented cortical circuits.¹² ¹⁶
Reardon’s CTRL-labs sEMG research was itself peer-reviewed and published in Nature in July 2025 (“A generic non-invasive neuromotor interface for human-computer interaction”), describing a wrist-worn device achieving greater than 90% classification accuracy for gesture and handwriting decoding in novel users without per-person calibration — underscoring that Flourish’s founding team has a demonstrated record of turning motor/computational neuroscience into shipped, peer-reviewed, and commercially deployed technology (now embedded in Meta’s smart-glasses Neural Band).⁴ ¹³

Neural Noise

Browse

Graph View

Neural Noise

Browse

Flourish

Footnotes

Graph View