Seventeen days. That's how long it took Berkeley Lab's A-Lab to synthesize 41 new materials compounds last year—work that ordinarily drags on for months, sometimes years. No grad student pipetting late into the night. No principal investigator hunched over a bench. An AI system designed the experiments, a robotic platform executed them, and the whole apparatus closed the loop on itself: hypothesis, synthesis, validation, repeat.
If that sounds like a proof of concept, you're behind the curve.
What happened at Berkeley isn't an isolated demo anymore. It's the advance guard of something broader—a fundamental reordering of how scientific research actually gets done. A new generation of AI systems, variously called co-scientists, autonomous labs, or agentic research platforms, is starting to orchestrate entire discovery cycles from end to end. And judging by the billion-dollar pharma deals, the flurry of infrastructure startups, and the enterprise platform launches now hitting the market, the smart money thinks this shift is happening faster than most research organizations have prepared for.
The Money Is Moving
Start with the numbers, which only tell part of the story. Analysts project the autonomous AI market will balloon from $6.86 billion in 2024 to $170.6 billion by 2034—a 37.9% compound annual growth rate, with North America claiming roughly 42% of that pie. Lab automation itself, the unglamorous substrate of robotics and liquid handlers that forms the physical layer, is expected to grow from $8.36 billion in 2025 to nearly $15 billion by 2034.
But the more revealing signal? How life sciences companies are actually allocating capital. McKinsey reports that the share of organizations spending $5 million or more on generative AI is set to leap from 20% in 2024 to 32% in 2025.
That's not pilot money. That's infrastructure.
The ecosystem now spans several distinct layers, each maturing at its own pace. At the top, foundation model companies like Isomorphic Labs—Alphabet's DeepMind spin-out—are building AI engines specifically for drug design. Isomorphic inked partnerships with Eli Lilly and Novartis worth up to $3 billion in potential milestone payments, then raised another $600 million in 2025. Periodic Labs, founded by alumni from OpenAI and DeepMind's GNoME project, pulled in a $300 million seed round from Andreessen Horowitz to build what it's calling an "AI scientist" for materials discovery.
Below that tier, a new crop of platforms is weaving AI directly into the daily grind of research. Benchling, which provides electronic lab notebooks and LIMS software to more than 1,500 scientists at Sanofi alone, launched Benchling AI in 2025—agents and models sitting alongside governed experimental data, not bolted on as an afterthought. Revvity rolled out Signals Xynthetica last December, a "models-as-a-service" platform that lets researchers toggle between in-silico molecular design and wet-lab validation inside a governed ELN/LIMS environment.
And at the physical layer? Robotic orchestration platforms are finally getting serious. Automata raised a $45 million Series C in 2026 with backing from Danaher Ventures, positioning its LINQ benches as an "AI-ready lab OS." Emerald Cloud Lab operates fully instrumented, remotely accessible facilities with software-defined protocols and built-in analytics—4,500 functions you can call via a unified Command Center interface, no on-site technician required.
Here's the kicker: all ten of the top pharma companies have now partnered with AI drug discovery startups since 2023, according to CB Insights. Funding to AI-driven drug discovery companies hit $1.6 billion for biologics and roughly $1 billion for small-molecule platforms in 2024 alone.
Why Now?
Three forces are converging to make autonomous research not just viable, but inevitable.
First, the AI layer has matured past mere prediction into planning and execution. DeepMind's GNoME predicted 2.2 million new crystal structures in 2023—roughly 380,000 deemed stable enough to matter—expanding the Materials Project database nearly tenfold. Impressive, sure. But prediction without validation is just computational guesswork. When those predictions fed into Berkeley's A-Lab, the autonomous synthesis system successfully created several hundred of them in a matter of weeks.
The closed loop works.
ChemCrow, a large language model agent published in Nature Machine Intelligence in 2024, pushed the concept further. Give it a synthesis challenge, and it autonomously planned reactions, selected tools, and executed chemistry via IBM's RoboRXN platform. More recent preprints—SciToolAgent, ChemGraph, ChemHAS—show increasingly sophisticated multi-agent orchestration patterns, often backed by knowledge graphs and domain-specific tool libraries. The academic literature is moving fast here, maybe faster than the market realizes.
Second, lab infrastructure has reached a tipping point. It's not just better robots. It's integrated stacks. Automata announced a partnership with CellVoyant in 2026 to deliver AI-powered, closed-loop cell culture workflows. Benchling now connects directly to HighRes Biosolutions workcells and NVIDIA NIM for model inference, letting researchers trigger automated experiments from the same interface where they log data. Strateos and Emerald Cloud Lab offer entirely remote, software-programmable labs.
Nature named self-driving labs a "technology to watch" in its 2025 outlook—not exactly a fringe prediction anymore. McKinsey positioned AI as shifting from a research tool to a "coworker," capable of orchestrating design-make-test-analyze cycles end to end. BCG X launched an AI Science Institute in April 2025 to partner with R&D organizations on deployment programs. On the academic side, the University of Toronto's Acceleration Consortium secured $200 million in Canadian federal funding and attracted another $300 million in partner commitments to build self-driving lab capacity. Their stated goal? Compress materials discovery from 20 years and $100 million down to one year and $1 million.
Ambitious? Certainly. Impossible? Probably not.
Third, there's genuine market pull. Pharma executives are placing real bets that AI-designed molecules will reach the clinic in 2025 or 2026. Isomorphic's Demis Hassabis has publicly framed the industry as entering clinical validation—not exploratory research, but actual trials. Recursion acquired Valence and Cyclica to beef up its ML research capabilities. Schrödinger signed a collaboration with Novartis worth up to $2.5 billion.
These aren't research grants. They're product bets.
What's Actually Working
The best way to understand this shift is to look at specific implementations, not just the hype cycle.
Materials Discovery at Scale
DeepMind's GNoME project wasn't just a paper. It generated 2.2 million candidate crystal structures using graph neural networks trained on existing materials databases. When those predictions were made available to the Materials Project and linked to Berkeley Lab's A-Lab—an autonomous synthesis platform that integrates computational planning, robotic execution, and real-time characterization—the system synthesized 41 new compounds in 17 days. Several hundred of GNoME's predictions have now been validated experimentally.
That's the closed loop functioning in the wild: AI generates hypotheses, autonomous labs test them, results refine the model. Rinse, repeat.
Pharma Partnerships
Isomorphic Labs kicked off 2024 with multi-target discovery collaborations with Lilly and Novartis, collectively worth up to $3 billion in milestone payments. The company raised an additional $600 million in 2025 to scale its AI drug design engine—serious capital for a company that didn't exist five years ago. Sanofi is using Benchling's platform across more than 1,500 scientists to underpin what it describes as AI-driven R&D, embedding agents directly into the workflow where experimental data lives.
These aren't pilots tucked into a skunkworks team. They're core infrastructure decisions.
Autonomous Biologics
Automata's partnership with CellVoyant, announced in 2026, aims to deliver closed-loop cell culture workflows where AI monitors culture conditions, predicts optimal media changes or passages, and directs robotic execution—all without manual intervention. It's the biologics equivalent of A-Lab: continuous optimization cycles running faster than any human-in-the-loop system could manage. Whether it scales beyond a few carefully controlled use cases remains to be seen, but the technical foundation is there.
Integrated Platforms
Revvity's Signals Xynthetica represents the data-platform approach. Instead of building a standalone AI tool—easy to demo, hard to embed—Revvity put model-as-a-service capabilities directly into its ELN and LIMS software. Researchers can generate in-silico designs, export protocols to the lab, capture results, and feed them back to the model—all within a governed, auditable environment. That matters in regulated industries where data lineage and compliance carry as much weight as discovery speed.
What Comes Next
The next two to three years will determine whether this becomes a niche capability or the default mode for research-intensive industries. A few things to watch.
Convergence
Multi-agent orchestration frameworks are proliferating in academic preprints, but they need to harden into production-grade platforms. The companies that figure out how to marry agentic planning, governed data, and robotic execution in a single coherent stack—without requiring customers to stitch together five vendors—will capture disproportionate value. Benchling and Revvity are positioned well here. So are the cloud lab providers like Emerald Cloud Lab and Strateos, which already operate fully integrated, software-defined infrastructure.
Regulation
It's coming, whether the industry wants it or not. The EU AI Act entered force in August 2024, with obligations for general-purpose AI models starting in August 2025 and high-risk system rules phasing in through 2027. For AI-powered lab systems embedded in health or medical device workflows, providers may need to meet transparency, documentation, and data governance requirements by 2026.
In the U.S., the regulatory posture shifted when Biden's Executive Order 14110 was rescinded in January 2025, but sectoral frameworks—NIST's AI Risk Management Framework, FDA guidance—still apply. Then there are the dual-use biosecurity concerns. OpenAI and Anthropic have published early evaluations on LLM-assisted biological threat creation, and several model providers now offer "trusted access" programs for life sciences use cases. It's a delicate balance: enable innovation, but don't accidentally turbocharge bad actors.
Consolidation
Expect it. Periodic Labs' $300 million seed, Isomorphic's $600 million raise, and enterprise launches from Benchling and Revvity signal that the race to standardize the "AI co-scientist stack" is on. Startups building single-point solutions—just the agent layer, or just the robot orchestration—will need to integrate or get acquired. The winners will be platforms that span the full loop, from computational design to physical synthesis to data feedback.
For R&D leaders, the implications are tactical and immediate. If you're in biopharma, ask whether your ELN and LIMS can interface with both AI model endpoints and robotic workcells. If you're in materials science, evaluate whether your computational pipeline can feed autonomous synthesis platforms or if you're stuck generating predictions with nowhere to validate them at scale. If you're investing, note that all the mega-rounds are going to teams with deep technical pedigrees—GNoME alumni, frontier-model researchers, former DeepMind staff—and partnerships with established pharma or materials companies.
The bar for credibility is high. As it should be.
The Trajectory Is Clear
Research that once took years is compressing into months, sometimes weeks. Labs that once required teams of postdocs are running overnight with minimal human oversight. The companies that built the first functional closed loops—DeepMind with GNoME and A-Lab, Isomorphic with its pharma deals, Automata with its integrated benches—are now in execution mode, not R&D mode.
The infrastructure is real. The funding is committed. The market is moving.
Whether your organization is ready or not, the autonomous lab era has started. And it's not waiting for permission.
