Reflecting on 2024: a category sweep of the Nobels, AI enabled clinical readouts, and the first trillion dollar pharma in sight. We stand on the shoulders of giants. Looking to 2025: This is just the beginning. Make no small plans.

Earendil 🤝 Genentech. Another one.

I started @enveda so mothers could fight their diagnoses with power mine never had. Sept 18 is her birthday. It’s also the day our first IND cleared, our first patient data came in, and now our Series E closed. More backup for the fight ahead. Happy Birthday, Mom!

Today we are announcing a $311M Series E to bring pharma into the 21st century. Enveda started with a thesis: a molecule that evolution has already refined is a better place to begin than one designed from scratch. This year, two first-in-class molecules discovered by Enveda (ENV-294 and ENV-308) had positive readouts in humans. Many more are behind them. We are grateful to @CatalioCapital for leading this round, to our new investors Durable Capital Partners, @ICONIQCapital, @lightspeedvp, Surveyor Capital (a Citadel company), accounts advised by @TRowePrice, @digitalisvc and Alderline Group, and to @BaillieGifford, @PremjiInvest, @trueventures, @KinnevikAB, @FPVventures, @_DimensionCap, @lifeforcecap and @Lux_Capital for continuing to back us. The best is yet to come… Read more: businesswire.com/news/home/2… #Biotech #AI #LearningfromLife

Molecular design tools are nowhere close to commoditization, have tremendous headroom left to improve, will generate substantial value for those who move the needle, and are clearly the bottleneck for a meaningful set of potentially high-value targets.

Zavain Dar on Backing Hugging Face Early & Why NVIDIA Wanted It piped.video/watch?v=M_pVjudx…

Couldn’t be more excited to announce our partnership with Ridge Bio and @WKightlinger, who are developing a computational enzyme and linker design platform to generate novel conjugation medicines — starting with ADCs. Insufficient linker design and payload conjugation methods have plagued ADCs since the beginning. Much of the payload falls off early in circulation due to non-specific cleavage, limiting efficacy when the ADC arrives at the tumor site and leading to considerable systemic tox. Ridge’s platform solves both of these issues while simultaneously enabling selection of novel tumor targets that were previously impossible to leverage without their tech. biospace.com/press-releases/…

Welcome to the team Darsh!

Welcome @darsh_cp!!

shared with our partners a month ago, 1H 2026. in light of Astra, navier-stokes, and clear signs of RSI, sharing below for all. buckle up. The most provocative statement we’ll write in this letter: the prior analysis of market disjuncts, the search for air pockets, and the conclusion that demand still massively outstrips supply holds even with diminishing returns to frontier model capability. If instead we continue to see linear or even super-linear returns, we are fast approaching a position of looking up at systems that far outpace our own capabilities, left with only broad and blunt instruments for intuition, let alone analysis of capability and performance. The Anthropic infrastructure responsible for training the models Levent Alpöge leveraged in refuting the Jacobian conjecture drew under 5 GW of power applied to compute. If we project largely public knowledge of OpenAI’s and Anthropic’s infrastructure contracting and buildout, the equivalent models entering Q3 2028 will have been trained on a combined 50 GW. Naively, we can intuit an IQ of 200 - can we do the same for 600, 700, or 800 IQ agents, digitally available, infinitely replicable, and forever on? In our conversations over the last two quarters we found surprisingly few wrangling with this question, or attempting to develop analytic scaffolds to reason meaningfully about what such a future could look like. Few beyond pockets of researchers, largely in the RSI space, seemed willing or able to traverse from the space of change in model performance over time, to changes in supply and demand over time. We might posit this as the central governor of technological, scientific, and economic growth for the next decade: the underlying rate of model improvement over time. We’d be remiss not to spend disproportionate time ourselves working internally on such tools to guide portfolio construction and investments as we lay the foundations for Dimension III.

excited to team up with people I’ve admired for a long time. humbled by the founders we get to work with, and those we’ve yet to meet. lots to build.

Excited to announce @_DimensionCap's partnership with @WKightlinger and the team at Ridge Bio to develop next-generation conjugate medicines! Ridge has built a computational enzyme design model to create custom conjugation and linker systems. Payload stability and programmatic release are two important levers in conjugate drugs, and Ridge has shown compelling data in both of these areas. Ridge's portfolio of enzymes and linkers are both compatible with existing ADCs and unlocks entirely new programs directed towards novel targets. Excited for the journey ahead and much more to come. ridgebio.com/

Today we’re announcing Darsh Patel has officially joined Dimension. @darsh_cp spent the last 4 years at Bessemer Venture Partners, and will help lead investments across the frontiers of science through to cutting edge compute and infrastructure. Welcome to the team Darsh!!

with AI the US and China are more intertwined today than ever been before. a few of us from @_DimensionCap recently spent a week on the ground in china. this past we shared the below with our LPs. information tends to want to be free. the letter was leaked. sharing here for all.

With all the excitement around AI accelerating science in the next few years, it can be easy to lose sight of the impact models are having _today_ on some of the hardest challenges in the world's leading pharmaceutical companies. Chai-3 was a step-change for us and our partners. Excited to work with @bmsnews to chart new frontiers in medicine and science.

*clean up on IL-6*

Frontier drug discovery programs require frontier capabilities. Today, we’re excited to share that BMS will be using Chai’s models to support the discovery of new therapeutics. Since rolling out Chai-3 to partners at the start of this year, we’ve partnered with @EliLillyandCo, @Novartis, @Pfizer, @argenxglobal, and now @BMSnews. Many of the world’s leading biopharma companies are now using our models to go after diseases. Scientific discovery is the domain where access to frontier AI matters the most. At @ChaiDiscovery, we’ve seen first hand how our latest model generations can make the difference between discovering a molecule and not. We’re excited to work with @BMSnews to break new ground in medicine and science.

Today's cancer vaccine results from @moderna_tx and @Merck are exciting, confirming the earlier Phase 2 results and also the news we got a few months ago at @ASCO. They're also a great excuse to discuss the lifecycle of AI in the drug development process. All emerging technologies have a maturation lifecycle. AI in the natural sciences is a perfect example. On the "boring end", you have old-school ML classifiers, decision trees, SVMs, and computer vision techniques that are silently baked into every sequencing machine and molecular diagnostic workflow on the planet. Would you consider this "AI in drug development". No, you wouldn't. (1) Because precision diagnostics aren't drug development and (2) because these techniques don't get clustered alongside the frontier/hypey areas like 'virtual cell'. That's the other end of the spectrum, the "hypey end". These are the AI implementations that attract disproportionate venture investment, make all the headlines, and that have attached a bunch of very smart sounding reasons why they'll never work. They become associated with the whole. Folks will often point to them as class representatives for 'AI in drug development' as a simple way to reject the entire class's effects on the industry. I have an allergic reaction to this because, to me, data-driven/ML methods are inseparable from and massively contribute to the life sciences. It's just that we don't celebrate the critical trailing edge because it's lost its sci-fi sheen, but it nonetheless serves as the foundation that all new advances become reality. So let's take the @moderna_tx and @Merck results. Would you consider this a success story for 'AI in drug development'? No, you probably wouldn't and I wouldn't blame you. It's true that cancer vaccines had/have many translational obstacles spanning delivery, supply chain, format, antigen selection, and more. It's foolish to try to decompose and assign credit to these pillars quantitatively. But let me zoom in to give some credit where it's due, because make no mistake, ML played a massive role. Beyond the fact that advanced Illumina machines have onboard NVIDIA accelerators and run various deep-learning algorithms in their analysis pipelines, which provide the essential information for personalized cancer vaccines, I can offer up a few things you may not know. @PersonalisInc, the diagnostic partner for @Merck and @moderna_tx throughout these vaccine trials, spent years training ML algorithms on proprietary immunopeptidomics datasets (paired sequencing and mass-spec) to help with neoantigen presentation/immunogenicity prediction. This isn't commodity sequencing. These and other Personalis technologies enhanced by data-driven methods (e.g., enhanced HLA typing) were indispensable to this process by supplying the vaccine developers with high-fidelity mutanomes. We don't know everything about the handoff to Moderna (e.g. which models Personalis used versus Moderna used), but we do knew they used another algorithm in the pipeline to select the exact neoantigens to include in the vaccine formulation. So, I think it's safe to say that without the steady improvement in data-driven/ML/algorithmic techniques, we would not be standing here today with these results. Of course, we can wax poetic that 'these aren't neural networks', and you'd be making my point for me. These weren't leading edge AI implementations, which is why the fuller class of technology doesn't earn its roses here. But that's how this is going to happen. Non-leading AI techniques will continue creating value in drug development. Today's frontier techniques will improve, lose their association with 'hype', and folks will still move goalposts around what is AI in drug development. But that's the nice thing about goalpost moving. You can start to build a nice trend-line about where things are going. ------- *I use the term AI loosely to refer to all data-driven techniques. Don't @ me.

a brief snippet from DIMENSION's most recent letter to partners

.@enveda’s two ph2 assets 1: novel target & pathway for I&I, showing early (small N, open label) signal for dupi-BETTER in a small molecule format 2: first in class leptin sensitizer, a workout in a pill @viswacolluru moving from strength to strength! statnews.com/2026/08/18/enve…

first-in-history data arrived @newlimit today small piece of a note to our team. we are so lucky to be living in this future.