‘We need the bravery to change the process’: How AI is impacting physical environments
Sifted
AI is moving from chatbots to materials science and drug discovery, and money's following fast. European advanced materials startups already raised €3bn this year, up from €1.6bn for all of last year.
Based on reporting by Sifted — read the original for the full story.
Summary, retelling and take written by AI under human oversight; images are AI-generated illustrations. How we work · Report an error
For a few years the AI conversation was mostly about language: chatbots, code assistants, digital workflow tools. That's shifting. Experts speaking to Sifted describe a pivot toward physical and scientific problems — materials for clean energy, molecules for medicine, the kind of stuff that doesn't live neatly inside a browser tab.
The money backs this up. European advanced materials startups have pulled in €3bn so far this year, according to Sifted data, dwarfing the €1.6bn raised across all of last year. Drug discovery startups have raised €4bn this year and are on track to top the €4.7bn raised in the whole of last year. Investors clearly think AI applied to atoms and molecules is where the next wave sits.
But physical problems don't behave like software problems. Chad Edwards, CEO of CuspAI, points out that code runs instantly and gives instant feedback; materials science doesn't. You have to actually make something and test it, and that loop is slow and expensive. Anthony Bradley of DaltonTx says surviving that loop means AI models can't just chase statistical patterns — they need physical constraints baked in, so a model generating antibody designs actually understands what's biologically expressible rather than spitting out nonsense that looks plausible on paper.
Getting from lab demo to real deployment is its own separate fight. Iraia Ibarzabal of Multiverse Computing says the real blocker right now is infrastructure — something that works in a controlled lab setting often falls apart outside it. Her fix is designing toward the end deployment from day one, rather than building a proof of concept and hoping it scales later. Bradley makes a similar point about process: sprinkling AI onto existing workflows won't transform anything, and organisations need the nerve to actually rebuild how they work around it.
None of this, they stress, means cutting scientists out. Edwards says the goal is freeing researchers from grinding through literature searches and manual data curation, not replacing their judgment. And Bradley offers a useful check on hype: AI is good at recombining what's already been measured, much shakier once you step outside that space. Looking ahead, Ibarzabal expects breakthroughs from pushing AI off cloud servers onto edge devices in places like defence or industrial settings where constant connectivity isn't an option, while Edwards imagines a future where success shows up not in a dashboard but in the physical world — solar panels, chips, materials you can point to and say AI helped build that.
My take — AI-written commentary, not fact-checked reporting
The line about needing 'bravery to change the process' is the most honest thing said here, because most companies bolting AI onto old workflows are doing exactly the lazy version everyone warns against. Bradley's warning about overestimating AI's ability to do genuinely new things without data deserves more attention than the funding numbers do — €3bn and €4bn raised this year won't matter much if the underlying models are just recombining old measurements dressed up as discovery. The real test isn't the cash pouring into materials and drug discovery startups; it's whether any of it survives contact with a lab bench.
Read more about this at: Sifted
