Key Points
- Flourish raised $500M at a $2.5B valuation, closing around June 4, 2026
- Jeff Bezos anchored roughly $100M; GV, Lux Capital and Catalio joined
- Co-founder Thomas Reardon created Internet Explorer and later sold CTRL-labs to Meta
- Flourish targets AI chips drawing 20–50 watts, near the brain's ~20-watt draw
The most expensive problem in artificial intelligence right now is not intelligence — it is electricity. Every frontier model runs on warehouses of GPUs that draw orders of magnitude more power than the three-pound organ they are trying to imitate. Flourish, a New York startup, just raised $500 million at a $2.5 billion valuation to attack that gap directly, and it convinced Jeff Bezos to anchor the round.
The round
Flourish closed roughly $500 million around June 4, with Bezos personally committing on the order of $100 million — close to a fifth of the raise — after Alphabet's GV, Lux Capital, and the healthcare-focused fund Catalio piled in. Bezos reportedly started smaller and nearly doubled his stake as the round filled. For a company that has shipped no product, a $2.5 billion valuation is a statement about who is building it more than what it has built.
Who is building it
Co-founder Thomas Reardon is one of the more improbable résumés in computing: he led the team that created Internet Explorer at Microsoft, then earned a neuroscience PhD and founded CTRL-labs, the neural-interface company Meta acquired in 2019 in a deal reported at $500 million to $1 billion, and whose technology now powers Meta's Neural Band wristband. His partner, Rob Williams, came out of senior roles at Amazon. The pitch is connectomics — mapping the brain's wiring at the level of cortical columns, using in-house electron microscopes — to find what they call the brain's "core algorithm" and rebuild it in silicon.
The bet
A modern GPU cluster burns something like 30 times more energy than a human brain doing comparable work. Flourish says its approach can cut that draw by more than an order of magnitude, targeting chips that run on 20 to 50 watts — roughly the brain's 20-watt envelope. The near-term goal is narrower than artificial general intelligence: a model that can learn continuously, cheaply enough to live on a consumer device rather than phone home to a data center. The company is reportedly in talks with an unnamed chipmaker and expects a breakthrough within five years.
The read
The dominant AI trade is still brute force — more GPUs, more power, more capital. Apple is renting Nvidia Blackwell GPUs through Google Cloud just to keep its private-cloud inference running. Flourish is the contrarian wager: that the path past AI's power crisis runs through neuroscience, not another data center, and that the investors who backed the last era of compute are quietly hedging against it. Whether biology actually has a transferable "core algorithm" is the multi-billion-dollar open question — but the cap table says serious money is no longer willing to assume the answer is no.
Source: SiliconANGLE, Tech Times
Comments