For the better part of a decade, the dominant recipe in artificial intelligence has been deceptively simple: build a bigger model, feed it more data, and plug it into more chips. That formula produced the large language models that now draft emails, write code, and answer questions. It also produced an energy bill that is starting to look unsustainable. A new startup called Flourish has just raised a half-billion dollars on the bet that the way out of that bind is not another data center, but a closer reading of the only system we know that already runs general intelligence on a trickle of power — the human brain.
What Happened
Flourish Inc. has raised $500 million in funding at a $2.5 billion valuation, the company confirmed in early June. The round was first reported by Wired, which noted that Amazon founder Jeff Bezos supplied roughly a fifth of the total. According to people familiar with the deal, Bezos initially committed around $50 million and then nearly doubled his stake after other prominent investors joined. The remaining capital came from a consortium that included Alphabet's venture arm GV, the deep-tech fund Lux Capital, and the healthcare-focused investor Catalio Capital.
The company is the work of two founders with unusually relevant resumes. Thomas Reardon is a neuroscientist who, in an earlier life, led the team that built Internet Explorer at Microsoft. He went on to co-found CTRL-labs, a brain-computer interface company that Meta acquired in 2019 for an estimated $1 billion; that technology now underpins the Meta Neural Band, a wristband that lets people control smart glasses with subtle hand gestures. His co-founder, Rob Williams, is a former senior Amazon executive. Together they are building what Flourish calls Cortex AI, a system meant to emulate how biological brains actually process information rather than simply scaling up today's neural networks.
The headline pitch is efficiency. Flourish says a server-grade graphics card uses roughly 30 times more energy than the human brain to process a comparable amount of information, and the company wants to close that gap by more than an order of magnitude. Its stated target is a model that runs in the range of 20 to 50 watts — about what a laptop draws — instead of the kilowatts that power a rack of AI accelerators. To get there, the team plans to lean on an in-house neuroscience lab rather than a purely software approach.
Why It Matters
The timing is not an accident. The AI boom has quietly become an energy story. Training and running frontier models now consumes so much electricity that hyperscalers are signing power-purchase agreements for nuclear plants and racing to secure grid capacity years in advance. Every incremental gain in model capability has tended to arrive with a matching increase in compute, and compute means watts. That trajectory is exactly what Flourish is trying to break.
The intellectual argument behind the company is that biological brains are a proof of concept the industry has largely ignored. Real neural networks are sparse — most neurons are not wired to most other neurons — and they fire asynchronously, doing local processing only when needed instead of pushing every input through every parameter. They lean on hierarchical abstraction to avoid the brute-force computation that defines current architectures. If even part of that design philosophy can be ported into silicon, the payoff is not a marginally cheaper model but a fundamentally different cost structure for intelligence itself.
That framing also helps explain the investor list. A bet like this is less about capturing next quarter's enterprise software revenue and more about owning a foundational shift if it arrives. For Bezos, GV, and Lux Capital, a relatively modest check against a $2.5 billion valuation buys exposure to an outcome that, if it works, would reshape who controls the most valuable layer of the AI stack. As SiliconANGLE noted, Flourish joins a small cluster of companies — including state-space-model startup Cartesia and Meta's own JEPA research line — pursuing architectures that depart from the transformer orthodoxy.
Reaction
The response across the technology and investing community has been a mix of genuine intrigue and measured skepticism, and both reactions are well earned. On the enthusiastic side, observers point to Reardon's track record: he has already built a hard-science company at the intersection of neuroscience and computing and sold it to one of the largest technology firms on earth. The premise that current AI is wildly inefficient compared with biology is also not controversial; it is something most researchers will concede readily.
The skepticism is just as substantive. "Brain-inspired" computing has a long and humbling history, from earlier waves of neuromorphic chips to research programs that promised to reverse-engineer cognition and delivered narrower results. Connectomics — the painstaking mapping of neural connections — has produced beautiful, detailed wiring diagrams of fruit fly and mouse brains, but translating a static map of connections into a working, energy-sipping algorithm is a leap no one has yet made. Critics also note that Flourish is, for now, a research bet wrapped in a commercial valuation, with the most ambitious claims sitting several years out.
What has kept the conversation from tipping into dismissal is the company's willingness to commit to physical infrastructure rather than slideware. Reardon has said the plan is to build a neuroscience lab equipped with electron microscopes — instruments that can cost millions of dollars apiece and resolve structures far smaller than any optical microscope, because they image with electrons whose wavelength is orders of magnitude shorter than visible light. That is an expensive, unglamorous way to spend a Series A, and to many readers it signals that the founders intend to do the slow empirical work rather than simply rebrand existing techniques.
What's Next
Flourish has laid out a roadmap with both near-term and long-horizon milestones. In the immediate future, the company says it plans to release a model "that can learn continuously" and make it available on consumer devices — a meaningful contrast with today's models, which are largely frozen after training and re-learn nothing from the people who use them. It is also reportedly in talks with an unnamed chipmaker to ship a processor capable of running that model efficiently, and it is developing an AI memory-management system designed to cut the amount of data required for training.
The deeper scientific work will run through that planned neuroscience lab. One of its stated focus areas is the cortical column, the repeating structure of neurons that many researchers believe does much of the brain's heavy lifting in perception and reasoning. The idea is to study these columns in fine detail, extract the principles that make them so efficient, and translate those principles into a computational architecture. Reardon has framed the timeline candidly, telling reporters the team expects a genuine breakthrough within roughly five years rather than promising one next quarter. Wired's original report described the effort as a hunt for the brain's "core algorithm," a phrase that captures both its ambition and its uncertainty.
For the broader industry, the milestones to watch are concrete. Can Flourish demonstrate a model that meaningfully learns after deployment? Can it show real efficiency gains on a working task, not just a benchmark? And will the chip partnership materialize into hardware that anyone can buy? Those answers, more than the funding headline, will determine whether this is a new architecture or an expensive detour.
Closing Thoughts
There is a useful tension at the heart of the Flourish story. On one hand, the company is making a bet that is genuinely contrarian in a field that has spent years rewarding scale above almost everything else. The willingness of seasoned investors like Bezos, GV, and Lux to back an approach that explicitly rejects the bigger-is-better playbook suggests at least some of the smart money believes the current trajectory has a ceiling. On the other hand, the history of brain-inspired computing counsels patience and caution; the gap between a gorgeous connectome and a deployable, 50-watt general model is enormous, and intentions do not close it.
What makes the company worth watching is less the specific promise of a laptop-power AI and more the question it forces the industry to confront. If intelligence can run on 20 watts in a human skull, the staggering energy demands of today's models look less like an immutable law of nature and more like an artifact of a particular engineering choice. Flourish may or may not be the company that proves the point. But the half-billion dollars now riding on that question is a signal that the assumption underpinning the entire AI build-out — that capability and power consumption must rise together — is no longer something the field is willing to take for granted.
한글 요약
뇌 모방 인공지능을 개발하는 스타트업 플로리시(Flourish)가 25억 달러 기업가치를 인정받으며 5억 달러를 유치했다. 투자금의 약 5분의 1은 아마존 창업자 제프 베이조스가 댔고, 알파벳의 벤처 부문 GV, 딥테크 펀드 럭스 캐피털, 헬스케어 전문 펀드 카탈리오가 함께 참여했다. 회사는 인터넷 익스플로러 개발을 이끌었고 이후 메타에 인수된 뇌-컴퓨터 인터페이스 기업 CTRL-labs를 공동 창업한 신경과학자 토머스 리어든과, 전직 아마존 임원 롭 윌리엄스가 세웠다.
플로리시의 핵심 주장은 효율성이다. 서버급 그래픽카드가 인간의 뇌보다 약 30배 많은 에너지를 쓴다고 보고, 회사는 그 격차를 한 자릿수 배율 이상 줄여 노트북 수준인 20~50와트로 작동하는 모델을 목표로 한다. 이를 위해 전자현미경을 갖춘 사내 신경과학 연구소를 세워, 뇌가 정보를 처리하는 핵심 단위로 꼽히는 피질 기둥(cortical column)을 정밀하게 연구할 계획이다. 단기적으로는 소비자 기기에서 '지속적으로 학습하는' 모델을 선보이고, 미공개 반도체 기업과 전용 칩 생산을 논의 중인 것으로 전해진다.
업계 반응은 기대와 신중함이 뒤섞여 있다. 현재의 AI가 생물학적 뇌에 비해 비효율적이라는 전제 자체에는 이견이 적지만, 뇌의 연결 지도(커넥토믹스)를 실제로 작동하는 저전력 알고리즘으로 옮기는 일은 아직 누구도 해내지 못한 과제다. 리어든은 약 5년 내 돌파구를 기대한다고 밝혔다. 플로리시가 그 답이 될지는 미지수지만, 능력과 전력 소비가 함께 늘어야 한다는 AI 업계의 오랜 가정에 5억 달러가 의문을 던졌다는 점만은 분명하다. (참고: Wired, SiliconANGLE, The Next Web)