Autonomous Labs
Abstract
Unlocking laboratory capacity and expanding testing capability through dexterous robotics and physical intelligence
Description
The foundation of my investment thesis in robotics and automation is the ability to demonstrate a throughput unlock. Nowhere is this more apparent to me than in third-party labs and CROs. The level of economic and scientific activity being implicitly held back by the manual nature of lab testing is incalculable—but ask any founder and I’d bet they’d agree.
This is not due to incompetence or complacency. Laboratory processes are slow, delicate, and skill-dependent. Safety and accuracy are paramount, which necessitates a high degree of care and control. Still, my sense is that most lab systems are at or near capacity, and lab technicians, scientists, etc., are not a labor pool that can be quickly expanded. There are over 300,000 CLIA labs in the US, and this year the CRO market is valued at ~$85 billion with no sign of slowing down (9.6% CAGR). Yet it’s still commonplace for complex tests (e.g., tumor screening, genetic analysis) to take weeks, and for more niche analyses, even months. This isn’t limited to health and life sciences, though they overwhelmingly represent the market activity. Chemical and material synthesis show similar inefficiencies
At the same time, the cost of even routine tests is absurdly high. This is a macro issue across healthcare in the West, driven by a multitude of factors, but increased volume would almost certainly drive down costs. Lab operations—with their standardized processes and robust logistics networks—offer a unique opportunity to (regionally) centralize and scale.
We’re nearing a functional intelligence inflection point where robotics can move beyond single-device automation. Robotic systems likely have one or two more manipulation capability breakthroughs to go, but soon entire lab tests and workflows will be streamlined and processed 24/7. The “why now” is classically technological—fine-skilled dexterous endpoints, transfer reinforcement learning, VLAs, cheaper COTS sensors, etc.—all coming together for true systems engineering. And unlike most robotics applications, I believe this will be a true Jevons paradox, with vastly more testing taking place in a post-automation world.
Working Thoughts
This obviously has enormous overlap with my Automated Research hypothesis that I’m exploring in parallel. The primary differentiator is that Autonomous Labs focus on the means of production and are thus, in my opinion, more interesting for investment.
I’m quite unclear on the best positioning in this field. I could see making a bet on a semi-autonomous, on-demand CRO cloud lab—recognizing that this would require the right investor base, and that revenue would almost certainly be non-recurring and transactional.
Equally, a colocation lab cell model (RaaS) seems like a viable entry point. I’m increasingly of the view that going fully vertical could be the superior path, but that places one in a more traditional life sciences inflection point strategy.