Watch the pattern: AI optimizes Yamanaka factors and the manufacturing timeline for cell therapy compresses from 3 weeks to 7 days. AlphaFold 3 predicts drug candidates and hit rates in small-molecule discovery jump from single digits to 60%. DeepSeek-R1 open-sources 671B parameters and the barrier to training custom biological models drops by orders of magnitude.
This isn’t coincidence. It’s a flywheel. And it’s redefining what’s possible in biotech fundamentally and irreversibly.
The History of Technology: Commoditization Drives Convergence
The history of the internet is best understood as a progression of commoditization. Compute became cheap. Storage became cheap. Bandwidth became cheap. Each time a foundational layer commoditized, new layers could build on it, and the overall velocity of innovation accelerated.
We’re watching the exact same pattern in AI and biotech right now, except the timescale is compressed to months instead of years.
2020: AlphaFold solves protein folding. Structural prediction becomes a solved problem.
2023-2024: Frontier AI models (Claude Opus 4.6 at 96.3% on AIME, DeepSeek-R1 reasoning at open-source scale) democratize reasoning about complex systems.
2024-2025: Domain-specific AI (generative models for protein engineering, AlphaFold 3, ADMET prediction, sequence generation) starts solving category-specific problems faster than humans can formulate the questions.
2025-onwards: Biological design becomes programmable. Not in theory. In practice. Companies are doing this today.
The Loop: More Data, Better Models, Faster Discovery
Here’s the flywheel concretely. Foundation models improve—better language models, larger parameters, improved reasoning enable better biological reasoning. Domain-specific models get trained on protein sequences, crystal structures, drug-target interactions, genetic data. Discovery accelerates as these models predict faster, cheaper, better candidates. Every experiment run generates new data that becomes training data for the next generation of models. And the loop tightens: what took 5 years in 2015 takes 6 months in 2025 and will take 2 weeks in 2030.
This is why we wrote about programmable reality and the Triple Convergence. Biology stops being something you study and starts being something you engineer. The investment thesis changes overnight when that inflection happens.
The Concrete Examples
Protein Engineering (OpenAI + Retro Biosciences)
Traditional protein engineering: identify target protein, design variants based on structure, test hundreds, find winners. Months of work. AI-enabled protein engineering: use generative models to design variants, predict function computationally, test the top 5-10. Weeks of work. Result: Yamanaka factors redesigned. iPSC generation efficiency up 50x. Timeline to iPSC therapy compressed by years. See more detail in our piece on generative models for protein engineering.
Small-Molecule Discovery (AlphaFold 3 + Scripps)
Traditional small-molecule discovery: identify target, screen 5,000-10,000 compounds, find 1-5 that work, optimize. Years of work. $100M+ per drug. AI-enabled discovery: identify target, predict binding with AlphaFold 3, synthesize top 20-50 candidates, validate. Months. Dramatically lower cost. Result: 22 compounds predicted for aging targets. 70% hit rate. Top compound shows 74% lifespan extension in C. elegans.
Gene Editing (AI + CRISPR)
We’ve analyzed how AI is optimizing CRISPR efficiency and safety. The same pattern: AI predicts optimal guide RNAs, predicts off-target effects, reduces experimental iterations from hundreds to tens, improves editing efficiency dramatically.
Drug Safety Prediction (ADMET Modeling)
AI-enabled ADMET means fewer compounds fail in clinical trials due to bad pharmacology. Faster clinical translation. Lower cost per approved drug. Read more in our ADMET modeling guide.
Why This Matters: The Venture Lens
From a VC perspective, this flywheel changes the risk calculus of biotech investing fundamentally. Biotech has historically been a “massive upfront capital, decade-long timeline, binary outcome” asset class. High variance. High time-to-return. Requires deep scientific conviction to hold through failures.
AI changes that. Not perfectly. But meaningfully. Computational prediction means you validate hypotheses faster, earlier, cheaper. Design-make-test cycles compress from months to weeks. Lower capital intensity means more shots on goal. A team with one world-class biologist plus computational tools can outpace a traditional 10-person lab. AI multiplies human capability.
This is why companies like Retro Biosciences are raising at $5B valuations. Why longevity biotech is suddenly a category. Why I’ve invested in 180+ biotech companies and keep finding new ones worth backing. The flywheel is real, it’s accelerating, and we’re early in it.
The Compounding Problem: Multiple Loops Running in Parallel
Here’s what makes this truly explosive: these flywheels aren’t sequential. They’re parallel and interdependent.
Better AI models lead to better protein prediction, which leads to better iPSC manufacturing, better cell therapies, more biological data, and better AI models. Better AI models lead to better small-molecule discovery, more drug candidates, more clinical data, better ADMET models, and more drug candidates. Better AI models lead to better gene editing, more genome engineering, more genetic data, and better generative models.
All three loops are running simultaneously. They’re producing data that feeds into each other. The velocity doesn’t compound linearly. It compounds exponentially. And the compounding started roughly 18 months ago. We’re still in the early phase.
Open Source Acceleration
There’s an additional layer: open sourcing. When DeepSeek releases a 671B parameter reasoning model as open-source, the barrier to entry for training custom biological AI models drops catastrophically. Any team with cloud compute can train models for their specific problem.
This democratizes AI-enabled biotech. You don’t need to be OpenAI to have access to frontier AI for biology. You can build custom models on top of open-source foundations. This is how the Neuromorphic Web and decentralized intelligence emerge—see our deep dive on the neuromorphic web for the broader context, but the point here is: open-source AI models are democratizing biological discovery just like they democratized software.
The Historical Parallel
In the 1990s, the internet commoditized communication. This enabled e-commerce, which created data, which created better targeting, which created better e-commerce. The loop tightened over 20 years.
In the 2010s, cloud compute commoditized processing. This enabled machine learning at scale, which created better models, which enabled smarter applications, which created more data.
In the 2020s, foundation models are commoditizing intelligence. This enables biological engineering at scale. Which creates better medicines. Which creates more biological data. Which trains better models. Which accelerates the cycle.
Each phase was thought to be impossible by people inside the previous one. We’re not going to make the same mistake and say AI will never design therapies that measurably extend human lifespan. Because we’re already watching it happen.
What We’re Watching
The key inflection points: First AI-discovered drug in Phase 2 human trials. First AI-optimized cell therapy approval. First gene-therapy using AI-designed guides in humans. First biomarker-defined longevity improvement in humans. Consolidation of computational biology tools as the winners in AI for bio start emerging.
Each of these is probably 2-4 years away, maybe closer. But they’re not theoretical. They’re pipeline. Companies are building them. And once one hits, the flywheel doesn’t slow down. It accelerates further.
The Bet
We’re betting that this flywheel doesn’t slow down. That biology becomes increasingly programmable. That longevity science moves from theoretical to applied to therapeutic within the next 3-5 years. And that the companies building the infrastructure for AI-driven biotech will capture enormous value.
If you want to understand where this is heading next—and which companies are actually moving fastest—subscribe to Accelerated for analysis informed by 180+ biotech investments and the pattern recognition that comes from building this ecosystem.
Related Reading
- AI for Science: How Artificial Intelligence is Revolutionizing Drug Discovery, Biology & Research
- AlphaFold and Drug Discovery: How AI Is Finding Molecules That Extend Lifespan
- The Evolution of Layer-1 Blockchains and Smart Contracts: Ethereum, Solana, and Beyond
- Programmable Reality: AI, Biotech, and the Future of Abundance