Biology's Recombination Moment: AI Is Becoming the Operating System for Life Sciences
In cosmology, "recombination" was the moment when the early universe cooled enough for protons and electrons to form neutral atoms. The universe became transparent. Light could finally travel. Structure could begin.
Biology may be entering its own recombination moment.
For decades, life sciences has been drowning in data: genomes, transcriptomes, proteomes, pathology slides, clinical records, real-world evidence, wearables, imaging, electronic health records, and increasingly multimodal patient journeys. The problem was not that we lacked signal. The problem was that the signal was trapped inside systems too complex for humans — or traditional software — to fully interpret.
AI is beginning to change that. Not magically. Not instantly. And certainly not without failures. But materially.
The story has also evolved. A few years ago, the conversation was mostly about whether AI could design a molecule, predict a protein structure, or read an image. Today, the more strategic question is broader: who will build the infrastructure layer for biology?
Recent announcements make the shift clear. Isomorphic Labs, the Google DeepMind spinout, announced a $2.1 billion Series B in May 2026 to scale its AI drug design engine and advance its candidate pipeline. Lilly and NVIDIA announced a co-innovation lab with up to $1 billion of investment over five years, focused on AI-enabled drug discovery, agentic wet labs, computational dry labs, and continuous learning systems. Thermo Fisher and NVIDIA announced a strategic collaboration to bring AI into scientific instrumentation, laboratory infrastructure, and automation. QIAGEN and NVIDIA announced a collaboration to combine curated biomedical knowledge bases with graph-based AI for target discovery, biomarker research, and hypothesis generation.
This is no longer just "AI for biology." It is biology being reorganized around AI-native infrastructure.
Prediction 1: Drug discovery becomes a data-and-compute arms race — but clinical proof still matters
AI will almost certainly make parts of drug discovery faster. It can generate molecules, search chemical space, model protein interactions, prioritize targets, and design experiments. But the more important point is that AI is changing the competitive architecture of R&D.
In the old model, pharma advantage came from chemistry expertise, development experience, capital, clinical operations, and privileged biological insight. Those still matter. But a new layer is being added: proprietary datasets, foundation models, compute infrastructure, automated labs, and the ability to close the loop between prediction and experiment.
That is why the Lilly-NVIDIA announcement is so strategically important. The stated goal is not simply to "use AI." It is to create a continuous learning system that connects wet labs and dry labs, so experiments, data generation, and model development reinforce one another 24/7. In other words, the lab itself becomes a data engine.
This is also why companies like Isomorphic Labs are attracting such extraordinary capital. The prize is not one drug. The prize is a reusable drug design engine. Isomorphic says its funding will support its AI drug design engine, global scale-up, and drug candidate pipeline across multiple therapeutic areas and modalities.
But this is where we need to be intellectually honest. AI has not repealed biology. To date, AI-designed or AI-discovered drugs have not yet produced the kind of broad late-stage clinical validation that would settle the debate. A 2025 Nature Medicine paper on Insilico's rentosertib noted that few AI-discovered or AI-designed drugs have reached clinical trials, that none had yet progressed through Phase 3, and that whether AI can meaningfully disrupt drug development remained unanswered.
That is not a bearish statement. It is a clarifying one.
AI may compress discovery. It may improve candidate quality. It may help identify better targets and better biomarkers. It may reduce the number of failed experiments. But clinical development still runs through humans, biology, regulators, payers, physicians, and patients. The bottleneck is moving downstream.
So the winning companies will not be the ones with the best AI demo. They will be the ones that can connect AI-generated hypotheses to validated biology, patient selection, trial design, regulatory-grade evidence, and ultimately clinical adoption.
Prediction 2: Diagnostics and tools become the sensory layer for AI-native medicine
If therapeutics are where AI gets the headlines, diagnostics and life science tools may be where AI gets its eyes and hands.
AI models are only as good as the data they can learn from. In life sciences, that data comes from instruments, assays, images, sequencing runs, pathology slides, lab workflows, and clinical systems. That means tools and diagnostics companies are not peripheral to the AI-biology story. They are central.
The Thermo Fisher-NVIDIA collaboration is a perfect example. Thermo Fisher described the goal as connecting scientific instruments, laboratory infrastructure, and data to AI solutions, with the aim of increasing automation, accuracy, and laboratory speed. That is a fundamentally different vision from selling boxes and reagents. It is a vision where instruments become AI-connected nodes in a broader discovery network.
QIAGEN's announcement points in the same direction from the informatics side. Drug discovery does not only require more models. It requires credible biological context: genes, pathways, diseases, compounds, evidence, and clinical associations. QIAGEN and NVIDIA are working on graph-based AI over biomedical knowledge graphs so researchers can explore disease mechanisms, therapeutic targets, biomarkers, and hypotheses with greater speed and context.
This is why I believe tools and diagnostics are strategically underappreciated in the AI conversation.
Everyone wants the AI-generated drug. But before you get the drug, you need the data. Before you get the data, you need the instrument, the assay, the workflow, the sample, the quality control, the annotation, and the interpretation layer. In other words: tools and diagnostics are the sensory system of AI-enabled biology.
In clinical medicine, the same logic applies. The FDA now maintains a public list of AI-enabled medical devices authorized for marketing in the U.S., while also emphasizing that these devices have met applicable premarket requirements and that the list is meant to improve transparency for patients, providers, and innovators. The FDA has also issued AI-related guidance for medical device software and has acknowledged that adaptive AI/ML does not fit neatly into the traditional device regulatory paradigm.
The signal is clear: AI is moving from novelty to regulated clinical infrastructure.
Prediction 3: Biology becomes more programmable, but not fully predictable
The most exciting advances are no longer limited to classic drug discovery. AI is beginning to attack the deeper problem: understanding biological function itself.
Google DeepMind's AlphaGenome, introduced in 2025, predicts how single variants or mutations in human DNA sequences may affect gene regulation across molecular properties. It can process sequences up to one million base pairs and predict features such as transcription, splicing, chromatin accessibility, and protein binding. That matters because so much disease biology sits outside the protein-coding genome. For years, the non-coding genome has been a kind of dark matter: clearly important, but hard to interpret at scale.
Biohub's May 2026 protein biology "world model" is another example of the direction of travel. Biohub described the release as an open engine for prediction, design, and discovery that can map proteins across the tree of life, predict structures, and design new protein binders that function in laboratory experiments. Its ESM Atlas covers 6.8 billion proteins, giving researchers a massive map of protein structure and function.
Google's broader Gemini for Science initiative points to the same pattern: AI systems that help generate hypotheses, run computational experiments, search literature, and integrate tools such as AlphaFold, AlphaGenome, UniProt, and InterPro into scientific workflows.
This is where the "recombination" metaphor still works. Biology is becoming more visible. Not simple. Not solved. But more transparent.
We are moving from an era where we could measure biology faster than we could understand it, to an era where AI may help turn measurement into models, and models into experiments.
But we should resist the temptation to say biology is becoming fully programmable. Cells are not software. Patients are not simulations. Biology is path-dependent, redundant, adaptive, and messy. The better analogy may not be "AI writes the code of life." It may be "AI gives us a better debugger for biology."
That is still revolutionary.
Prediction 4: Regulators will become AI users, not just AI referees
A major update since the original version of this article is that regulators are not just reacting to AI. They are starting to use it.
The FDA launched Elsa, an agency-wide generative AI tool, in June 2025 to help employees such as scientific reviewers and investigators read, write, summarize, compare labels, summarize adverse events, and generate code for internal databases. In May 2026, the FDA announced Elsa 4.0 features including custom agents, document generation, quantitative data analysis and visualization, OCR, secure web search, and optimized search across large document repositories.
At the same time, CDER has acknowledged a significant increase in drug submissions incorporating AI components and said its 2025 draft guidance was informed partly by experience with more than 500 submissions with AI components from 2016 to 2023.
This matters because the adoption curve for AI in life sciences will not be determined by algorithms alone. It will be determined by trust.
Can a sponsor explain how the model was trained? Can the model be validated in a specific context of use? Can the evidence be reproduced? Can regulators understand what role AI played in the decision? Can physicians explain the result to patients? Can payers assess clinical utility?
In life sciences, trust is not a soft concept. It is a commercial requirement.
Strategic implications: build, buy, partner — and above all, own the workflow
For executives and investors, the strategic question is no longer whether AI matters. It is where to place the bet.
Some companies will build. Large pharma, scaled diagnostics companies, and major tools companies will increasingly need internal AI capability. Not because every company should become OpenAI, but because AI strategy cannot be fully outsourced when the underlying data, workflows, and customer relationships are core to the business.
Some companies will buy. AstraZeneca's acquisition of Modella AI in January 2026 — focused on oncology R&D, quantitative pathology, biomarker discovery, and patient selection — is a good example of AI capability being pulled directly into pharma infrastructure. Recursion's completed combination with Exscientia in November 2024 is another example of consolidation among AI-native discovery platforms.
Many companies will partner. In fact, partnership may be the default model because no one company has all the pieces: proprietary biological data, clinical samples, wet-lab infrastructure, regulatory expertise, foundation models, compute, commercial channels, and customer trust.
But the deeper strategic principle is this: own the workflow.
AI does not create value in isolation. It creates value when embedded into a workflow that matters. Drug discovery workflows. Trial enrollment workflows. Variant interpretation workflows. Pathology workflows. Lab automation workflows. Manufacturing workflows. Reimbursement workflows. Physician decision workflows.
The companies that win will not simply "add AI." They will redesign workflows around AI — while preserving the validation, compliance, and trust that healthcare requires.
Conclusion: the lights are coming on, but biology still has the final word
The original recombination event did not create the universe. It made the universe visible.
That is the right way to think about AI in biology. AI will not instantly solve disease. It will not remove the need for experiments. It will not eliminate clinical trials. It will not make regulators, physicians, or patients irrelevant.
But it may make biology more legible.
It may help us see patterns we could not see, generate hypotheses we would not have imagined, design experiments more intelligently, and connect data across modalities that previously lived in separate silos.
For life science tools and diagnostics, this is a particularly important moment. AI needs data, and data needs instruments, assays, workflows, quality systems, and interpretation. That means the companies building the measurement layer of biology may become even more strategic, not less.
The future of precision medicine will not be driven by therapeutics alone. It will be driven by the convergence of therapeutics, diagnostics, tools, data, and AI-enabled workflows.
Biology's recombination moment is not the end of uncertainty. It is the beginning of a clearer sky. The lights are coming on. The question now is who will build the telescopes.