Inherent, a London-based AI lab founded by former DeepMind researchers, raised $50 million in seed funding to build Faraday, a platform that automates research-question prioritization for scientists. The Next Web reported the round on May 29. At $50 million, this is one of the largest seed rounds for a UK-based AI lab working in the scientific-discovery vertical.
The size of the raise signals something beyond a proof-of-concept. Inherent now has enough capital to compete with US-based science-AI labs on both talent and compute, without relocating to San Francisco. That matters because the UK’s AI talent layer has been a sustained recruiting target for US frontier labs since DeepMind’s parent Google began concentrating headcount in California. A well-funded science-AI spinout that stays in London is a direct counter-signal in that retention battle.
Faraday is positioned as the layer upstream of execution tools. AlphaFold tells you the protein structure once you know what protein to fold. Materials Project gives you property data for materials you already know to study. Faraday’s stated function is deciding which proteins, materials, or mechanisms are worth investigating in the first place, automating part of what a senior principal investigator does when setting a lab’s research agenda. If that positioning holds, Faraday sits above an entire generation of experiment-running tools rather than competing with them.
This is the second significant ex-DeepMind science-AI spinout in recent months, and the pattern is now clear enough to name. FutureHouse, backed by Eric Schmidt and focused on automating the scientific literature review and hypothesis generation loop, is the closest US analog. OpenBio has staked a position in the wet-lab automation space. Inherent is doing neither: it is not running experiments and it is not building lab robotics. Its bet is narrower and more specific: question prioritization as a standalone product.
That narrowness is worth scrutinizing. Prioritizing research questions sounds like a discrete function, but the hard version of this problem requires deep domain specificity. In drug discovery, the question-selection problem looks nothing like it does in fundamental physics or climate materials science. The Next Web’s reporting does not specify which scientific domains Faraday is targeting at launch. The absence of named domains in the coverage makes it difficult to evaluate whether Inherent’s system is a broadly applicable reasoning engine or a series of domain-specific modules wrapped in a unified product interface. That distinction determines whether Faraday becomes infrastructure or a point solution.
Inherent’s differentiation from DeepMind’s internal Project Genesis (which focuses on experiment execution rather than question selection) is clean on paper. The market test will come when Faraday has to prove that question prioritization produces measurably better research outcomes than the judgment of experienced scientists, and does it faster. That is a high bar. The company has not, based on current reporting, disclosed independent validation data or named institutional research partners.
For venture investors watching the science-AI stack assemble, Inherent’s round prices in a specific theory: that the most durable layer is not the one running experiments but the one deciding which experiments matter. If that theory is correct, Faraday could become a required input for any serious drug-discovery or materials-science program within the next two years. If the theory is wrong, $50 million is enough runway to pivot toward execution tooling before the burn becomes fatal.
Labs and research-intensive enterprises currently evaluating science-AI tools should track whether Inherent discloses domain coverage and validation partnerships in the next six months; that disclosure will determine whether Faraday is a strategic infrastructure bet or a well-funded hypothesis.
Reported by The Next Web on May 29, 2026.
