Jeff Bezos Backs Flourish: The $500 Million Startup Trying to Copy the Brain to Fix AI’s Power Crisis

A startup called Flourish has raised $500 million at a $2.5 billion valuation from a consortium that includes Jeff Bezos, Alphabet’s GV, and Lux Capital — and its ambition is to solve one of the most fundamental problems in modern AI: the power consumption crisis. While the world’s leading AI companies are paying SpaceX hundreds of millions of dollars per month for compute capacity and routing data through each other’s server farms to keep up with demand, Flourish is building something different. Its Cortex AI system is designed to emulate how the brain actually processes information — at 20 to 50 watts of power, roughly the draw of a laptop, rather than the megawatts consumed by conventional AI data centres.

Why the AI Power Crisis Is Real

The scale of AI’s energy problem is not fully appreciated outside the industry. Training a single frontier language model like GPT-4 or Claude 3 Opus requires approximately 50 to 100 megawatt-hours of electricity — equivalent to the annual power consumption of several thousand UK homes. Running those models in production, answering queries from millions of users simultaneously, requires even more. Google, Anthropic, and OpenAI between them now consume power at the scale of small cities.

The UK is feeling this directly. National Grid has warned that AI data centre demand could add 50 gigawatts of new electricity requirements to the UK grid by 2035 — equivalent to adding another large country’s worth of demand. Planning permission for new data centres in London and the South East is under pressure, and energy prices for large commercial consumers have risen partly due to AI-driven demand increases.

The reason conventional AI is so power-hungry is architectural. Large language models work by performing billions of mathematical operations — multiplications and additions across matrices of numbers — for every token of text they process. The hardware doing those calculations (Nvidia H100 GPUs) is extraordinarily power-dense. The calculation itself is energy-efficient at the chip level but the sheer volume of calculations required for a single query means the total energy adds up fast.

What Flourish Is Building

Flourish’s Cortex AI takes a different approach. Instead of scaling up matrix mathematics on conventional silicon, the company is mapping real biological neurons and their connections — a field called connectomics — and using those maps to design AI architectures that process information the way the brain does.

The human brain performs remarkable feats of pattern recognition, language understanding, and reasoning on approximately 20 watts of continuous power. It does this through a fundamentally different computational strategy: sparse activation (most neurons are silent most of the time), asynchronous processing (not everything runs on a clock cycle), and physical co-location of memory and processing (unlike conventional computers that shuttle data back and forth between chips and memory). Flourish is attempting to capture these properties in artificial systems.

The company’s target of 20 to 50 watts for its Cortex AI models would represent a reduction of three to four orders of magnitude compared to current frontier model inference. If achievable at comparable capability levels, this would be one of the most significant technological advances in the history of computing — not just for AI, but for how we think about energy-efficient information processing.

The Founders: A Track Record in Brain-Computer Interfaces

Flourish was founded by Rob Williams, a former Amazon executive, and Thomas Reardon, a neuroscientist whose career trajectory reads like a tour of transformative technology. Reardon created Internet Explorer at Microsoft in the 1990s — the browser that introduced hundreds of millions of people to the web. He then founded CTRL-labs, a company building brain-computer interface technology that captures neural signals through a wristband rather than a brain implant. Meta acquired CTRL-labs in 2019 for an estimated $1 billion.

Reardon’s background matters for Flourish’s credibility. Brain-inspired AI has a long history of ambitious promises that failed to materialise at scale. Previous approaches, including early neural network research and later neuromorphic computing chips from companies like Intel (Loihi) and IBM (TrueNorth), produced interesting research but have not challenged conventional AI architectures commercially. Reardon’s grounding in actual neuroscience and brain-machine interface engineering, combined with recent advances in connectomics technology, provides a more credible foundation than previous attempts.

Jeff Bezos’s Strategic Logic

Bezos’s involvement in Flourish — his initial $50 million commitment reportedly grew after other high-profile investors joined — is consistent with his investment pattern of backing technologies that address fundamental resource constraints. Amazon Web Services is one of the world’s largest consumers of electricity, and energy costs are a significant component of AWS’s operating expenses. AI inference on Flourish’s architecture would, if it works, dramatically reduce those costs.

Google Ventures (GV) joining the round is also strategically legible: Alphabet is facing the same compute and energy pressures as every frontier AI company, and having a stake in a technology that could offer a fundamentally different architecture gives it optionality regardless of whether Cortex AI becomes the dominant approach.

Fable 5 Still Offline: Eight Days and Counting

In a reminder that the AI industry’s most pressing near-term problem is regulatory rather than architectural, Anthropic’s Fable 5 model remained offline on 14 June — entering its eighth day under the hold imposed by the Trump administration’s jailbreak ultimatum. UK subscribers who had purchased Anthropic API credits for Fable 5 access remained unable to use those credits for the model’s most advanced capabilities.

Anthropic has been routing Fable 5 subscribers to its Claude Opus 4.8 model as a substitute, but the capability gap between the two is meaningful for professional and research use cases. UK AI developers who had built applications specifically optimised for Fable 5’s capabilities face disruption until the hold is lifted.

The Fable 5 situation and the Flourish funding round sit on opposite ends of the AI policy timeline. Flourish’s work will take years to produce commercially deployable systems. The Fable 5 hold is a problem that exists right now, today. Both underscore that the AI industry in June 2026 is simultaneously grappling with immediate regulatory crises and long-horizon technical bets about what the future of AI actually looks like.

What This Means for UK AI Users and Businesses

Flourish’s technology is too early-stage to affect UK AI users in the near term. The company’s $2.5 billion valuation reflects the potential of the technology if it works at scale, not a delivered product. UK businesses relying on AI should continue planning around conventional LLM architectures for their AI strategies through at least the end of the decade.

However, the investment is worth tracking as a signal of where the industry believes the structural bottlenecks lie. If Flourish or a competitor demonstrates brain-inspired AI at a fraction of the power cost of conventional models, it would dramatically change the economics of running AI — making sophisticated AI accessible to smaller organisations and reducing the environmental footprint of the industry as a whole.

For UK policymakers and the National Grid planners thinking about energy demand forecasting, Flourish’s approach is one of the more credible potential disruptors to the “AI = ever-increasing electricity demand” trajectory. It deserves to be taken seriously in energy planning scenarios, even at this early stage.

This article is for educational purposes only.