Jeff Bezos Backs Flourish With $500M to Build Brain-Inspired AI That Runs on 50 Watts

The scale of energy consumption required to run modern AI has become one of the industry’s most uncomfortable open secrets. A single server rack of NVIDIA H100 GPUs can draw 80 kilowatts. A large AI training run can consume as much electricity as thousands of homes use in a year. Against that backdrop, a startup called Flourish is pursuing a different path—one that starts not with chips, but with neurons.

Flourish raised $500 million at a $2.5 billion valuation in a round that closed in early June, with Jeff Bezos committing what sources describe as roughly $100 million, or approximately one-fifth of the total. Other investors include Alphabet’s GV (formerly Google Ventures), Lux Capital, and Catalio, a healthcare-focused fund. The round is among the largest for any AI company operating without a commercial product.

The Mission: Chase the Brain’s Algorithm

Flourish’s central thesis is that the human brain represents a solved existence proof for efficient general intelligence. The brain operates on roughly 20 watts—less than a single LED light bulb—while performing perception, reasoning, planning, memory consolidation, and creative synthesis simultaneously. The brain’s power efficiency is not merely impressive; it is approximately 30 times better per unit of information processed than a modern server-grade GPU, by Flourish’s own calculation.

The company’s flagship technology, Cortex AI, attempts to bridge this gap by doing something most AI companies do not: study actual neurons. Flourish operates an in-house neuroscience laboratory that uses electron microscopes to map the three-dimensional structure of biological neural tissue at nanometer resolution—a field called connectomics. The goal is to reverse-engineer the organizational principles that make biological neural networks so efficient, then instantiate those principles in artificial systems.

This is not the first time researchers have tried to learn from biology to build better AI. Decades of connectionist AI research drew inspiration from neuroscience. But Flourish’s bet is that modern connectomics—which can now map thousands of neurons and millions of synaptic connections in far greater detail than was possible even five years ago—will reveal architectural principles that have so far remained hidden from earlier generations of researchers.

The Founders

Flourish was co-founded by Thomas Reardon and Rob Williams, a pairing that combines unusual technical and operational depth.

Reardon’s claim to fame is building Internet Explorer at Microsoft in the 1990s, but his more recent work is arguably more relevant. He founded CTRL-labs, a brain-computer interface company that spent years studying the neural signals that control hand movements with extraordinary precision. Meta acquired CTRL-labs in 2019 for an estimated $1 billion. That acquisition gave Meta access to EMG-based neural interface technology; Reardon left and eventually co-founded Flourish.

Williams is a former Amazon S-team executive—the small group of senior leaders who report directly to Amazon’s CEO and shape the company’s strategy. His operational background gives Flourish the kind of large-organization credibility that pure research labs often lack when they attempt to bring technologies to market.

The Target: Consumer Devices, Not Data Centers

Flourish’s ambition is not to build a better GPU. The company is explicitly targeting a regime that current AI hardware cannot reach: models that run on 20 to 50 watts, suitable for deployment on consumer devices—laptops, smartphones, wearables, edge computing hardware—rather than requiring server racks.

If Flourish can deliver models with meaningful general capability in that power envelope, the implications extend far beyond energy savings. Always-on AI assistants that run locally without sending data to the cloud would address privacy concerns that have made many enterprises and individuals cautious about AI adoption. AI capabilities on devices in environments with poor or no connectivity—remote locations, aircraft, manufacturing floors—would unlock application categories that cloud-dependent AI cannot serve.

The company is also working on an AI memory management system that would reduce the amount of training data required to achieve given performance levels—attacking AI’s ravenous appetite for data at the same time it attacks its energy consumption. Flourish expects a technological breakthrough within approximately five years, with near-term consumer device deployment through an undisclosed chipmaker partnership.

Why Bezos Is Betting Big

Jeff Bezos’s investment pattern in recent years has followed a clear thread: he backs companies pursuing what he calls “gifts”—technological breakthroughs that seem implausible but have enormous leverage if they succeed. He has backed Blue Origin in space access, committed to nuclear fusion startups, and invested in longevity research. Flourish fits this pattern: the probability-weighted upside of cracking efficient general intelligence is enormous, even if the probability of success within any given timeframe is uncertain.

Bezos also has an institutional lens on AI power consumption through Amazon Web Services. AWS operates some of the largest GPU clusters in the world for AI inference workloads. The energy cost of running those clusters is a significant and growing line item. Any technology that reduces the compute power required to run inference—whether at the device level or at the data center level—has direct financial relevance to Amazon’s core cloud business.

Sources indicate that Bezos initially committed approximately $50 million to Flourish, then nearly doubled his stake when other high-profile investors joined the round. That pattern—initial bet, then reinforcement as signal confirmation arrives—is consistent with how Bezos has approached previous deep-tech investments.

The Skeptical View

The history of neuromorphic and brain-inspired computing is littered with ambitious projects that produced interesting science but failed to compete with conventional deep learning. Intel’s Loihi neuromorphic chip has been available for years with limited commercial traction. IBM’s TrueNorth chip, announced with significant fanfare in 2014, never achieved mainstream adoption. The fundamental challenge is that modern GPU-based deep learning, while energy-intensive, has proven remarkably flexible and improvable. Each generation of hardware has brought efficiency gains that postponed the inflection point where alternative approaches become competitive.

Flourish’s bet is that connectomics represents qualitatively new information that previous generations of brain-inspired AI researchers did not have access to. That may be true—or it may be that the organizational principles of biological neural tissue, evolved over hundreds of millions of years for biological survival rather than artificial intelligence, do not transfer cleanly to artificial systems.

The five-year horizon Flourish has set for a breakthrough is long by startup standards and short by research standards. Whether $500 million and two exceptional co-founders can close the gap between neuroscience and practical AI systems in that window is a question that will not have a clear answer for years. For now, the company represents the most serious capital commitment to the neuromorphic computing thesis in the current AI era—and a clear signal that at least some of the world’s most sophisticated investors think the human brain still has lessons to teach.