Here is a fact that sounds made up but isn't: roughly half of all the medicines a doctor can prescribe started life as a molecule made by something that was alive - a plant, a fungus, a bacterium quietly doing chemistry in the dirt. Aspirin came from willow bark. Penicillin came from mold. The statin in a lot of people's medicine cabinets came from a fungus. Nature, it turns out, has been running an enormous, unsupervised drug-discovery program for about four billion years, and it does not file patents.
The catch is that reading nature's output is miserable work. A single leaf can contain thousands of distinct molecules, most of them present in vanishing amounts, most of them structurally weird in ways no human chemist would think to draw. To figure out what any one of them is, a scientist historically had to purify it, run it through instruments, and puzzle over the results - a process that is slow, expensive, and does not scale to the size of the problem. So the pharmaceutical industry, being rational, largely gave up. It decided that natural chemistry was a data-legibility problem it could not afford to solve, and moved on to designing molecules from scratch.
Enveda's entire premise is that the industry gave up on the wrong problem. The company, founded in Boulder in 2019, does not think nature ran out of good drugs. It thinks we ran out of ability to read them. And reading, as it happens, is exactly the kind of thing modern machine learning has gotten unreasonably good at.
Spectra as a language
The core trick is to treat data from a mass spectrometer - the instrument that measures the fragments a molecule breaks into - the way a language model treats text. Enveda built a foundation model called PRISM, short for Pretrained Representations Informed by Spectral Masking, and trained it on more than a billion mass spectra. The idea is that spectra, like sentences, have a grammar: patterns that recur, structures that imply other structures. Feed a model enough of them and it starts to learn the syntax of the physical world, and can reconstruct the molecule behind a spectrum it has never seen before.
This is a subtly different bet than the one most AI-drug-discovery companies are making. The fashionable move is generative - have the model invent novel molecules and hope they're real. Enveda's move is comprehension - have the model read molecules that already exist, that evolution already stress-tested, and tell you what they are and what they might do. It is less glamorous and, arguably, more useful, because a drug eventually has to be a real substance that a real person can swallow.
"Nature has been running the most sophisticated R&D program on Earth for billions of years, yet nearly all of its chemistry remains unexplored."
The pharmacist's son
Enveda's founder and CEO, Viswa Colluru, grew up in Visakhapatnam, India, around his father's pharmacy and near a large hospital - close enough to watch medicine change lives and, in his family's case, close enough to feel its limits. He earned a PhD in immuno-oncology from the University of Wisconsin-Madison, reportedly the youngest in his graduating class, then worked at Recursion, one of the first companies to industrialize computational drug discovery. The combination - wet-lab biology plus computational scale - is more or less the exact skill set Enveda requires.
He started the company with $55,000 of his own savings. That detail gets repeated a lot, and it's worth pausing on why: it's not a rags story so much as a signal about conviction. Six years later the company has raised more than $700 million and, as of its September 2025 Series D, carries a valuation north of a billion dollars. That is a roughly five-orders-of-magnitude change in the size of the bet, and the bet itself never changed.
From a model to a molecule
A platform is only as interesting as what it produces, and this is where Enveda stops being a nice story about AI and starts being a drug company. Its lead candidate, ENV-294, is a first-in-class oral anti-inflammatory - a small molecule derived from natural chemistry that the company describes as a completely new chemical class, with both kinase-inhibitor and steroid-like properties. In a Phase 1b trial in patients with moderate-to-severe atopic dermatitis - eczema, the itchy, miserable, surprisingly large market kind - it reported a mean 85% reduction in disease severity by day 42, with every patient in the readout hitting the standard 50% improvement bar and more than half hitting 90%. It has since moved into Phase 2 trials in both eczema and asthma.
What makes that mechanistically interesting - and Enveda leans on this - is that ENV-294 doesn't work the way the blockbuster biologics do. Modern immunology drugs tend to block a single cytokine, one lever in a very complicated machine. Enveda says its molecule instead reconfigures the immune response more broadly, which is the kind of thing a molecule invented by evolution, rather than by a committee optimizing one target, might plausibly do. Nobody at Enveda sat down and designed it. A model found it in what the company likes to call the "dark chemical space."
The pipeline behind the headline
ENV-294 is the one with the data, but it's one of about 18 candidates. Behind it sit programs aimed at inflammatory bowel disease, at fibrotic conditions, and - the one that will get the most attention - a "hormone mimetic" for obesity, the hottest category in pharma. Roughly a dozen more programs sit in earlier stages. The strategic point is that if the platform genuinely works, the pipeline should keep refilling itself, which is the difference between a biotech that has one drug and a biotech that has a factory.
The people writing the checks appear to believe it. The Series D was led by Premji Invest, the family office of the Indian billionaire Azim Premji, with Kinnevik, Lux Capital, Dimension, Baillie Gifford and a long tail of others. Sanofi, an actual pharmaceutical company, is an investor - which matters, because pharma money is a form of due diligence that venture money isn't. And Mikael Dolsten, who until recently ran science at Pfizer, joined the board. When the person who oversaw Big Pharma's R&D engine goes to work for the outsiders reading plants with a neural net, the outsiders are no longer exactly outsiders.
What you can actually do with it
It's worth being concrete about who benefits, because "AI drug discovery platform" is the kind of phrase that can mean almost nothing. Enveda is not selling access to PRISM the way a software company sells seats; it is using the platform to build its own medicines. So the near-term beneficiaries are the patients its drugs are aimed at - the roughly one in ten adults who deal with atopic dermatitis, the asthma sufferers, the enormous and growing population that obesity drugs are chasing. If ENV-294 clears the trials ahead of it, the tangible thing a person gets is a pill for a condition currently treated mostly by injectable biologics that cost a great deal of money.
For the drug-discovery field more broadly, the useful export is the idea itself: that a whole category of data everyone had written off as unreadable can be read. Enveda's collaboration with Microsoft on PRISM, and the participation of Sanofi as an investor, both hint at a world in which large players want in on natural chemistry again - not by reviving the old, slow purification pipelines, but by pointing models at spectra at scale. If the approach generalizes, the winners aren't only Enveda's shareholders; they're anyone who was quietly sitting on decades of un-interpreted natural-sample data.
The competitive frame
Enveda is not alone in arguing that machine learning changes drug discovery. Recursion - Colluru's former employer - built its business on imaging cells at scale; Isomorphic Labs spun out of DeepMind's protein-folding work; Insilico, Genesis Therapeutics and Terray each have their own computational angle. What distinguishes Enveda is less the "AI" and more the substrate. Most of these companies are trying to predict or generate; Enveda is trying to read a specific, physical, under-explored dataset - the chemistry of living things - that its rivals largely aren't touching. Whether that's a durable moat or a niche is one of the genuinely open questions about the company.
The honest caveat
None of this means Enveda has won. It is a clinical-stage biotech, which is a polite way of saying most of its value is a promise, and biology has a long history of humbling promising Phase 1b data once it meets the larger, more skeptical trials that follow. The obesity program is early. The platform's core claim - that AI can systematically turn natural chemistry into approved drugs faster than anyone else - is exactly the kind of claim that takes a decade to actually prove. But the shape of the bet is unusually clean: not that nature has better drugs, which is basically known, but that the bottleneck was always reading, and reading is now cheap. If that's right, the interesting question isn't whether Enveda finds more molecules. It's how many were sitting there, legible all along, waiting for a machine patient enough to read them.
