Ginkgo’s CEO says biotech needs its Waymo moment

With late-stage R&D costs averaging over $2 billion per asset for large biopharma companies, according to Deloitte’s ongoing analysis, and more pressure to reduce those costs, something has got to give.

Jason Kelly, CEO of Ginkgo Bioworks, observes that one common answer has been outsourcing, especially to lower-cost markets. But he warns that this is not a durable fix for an industry still trapped by rising development costs and labor-intensive science. “You’ve got to kill the bench, or else we lose the industry,” Kelly said. He argues that the U.S. cannot stay competitive in biotech if discovery remains grounded in expensive manual lab work while China gains ground on cost, scale and scientific capacity.

One recent industry analysis cited Phase 3 non-small cell lung cancer trial costs of about $69,000 per patient in the U.S. versus $25,000 in China, and another report, citing GlobalData, said clinical trials can be 50% to 60% cheaper in China. McKinsey, in a  report titled “The emerging epicenter: Asia’s role in biopharma’s future,” also says China’s early discovery-to-IND cycles are 50% to 70% faster.  McKinsey also found that Chinese biopharma firms run “discovery programs at roughly one-third to one-half of global costs.”

Why outsourcing won’t save Western biotech

Western pharma and biotech sector has responded to the challenges with a mix of cost-cutting, outsourcing and selective modernization. The sector has leaned on CROs, overseas trial sites and lower-cost development hubs to keep programs moving, while also putting more money into AI and other tools meant to make R&D more efficient. The EY Biotech Beyond Borders Report 2025 notes that the VC funding picture in the sector remains mixed in the wake of the pandemic. Meanwhile, outsourcing to Asia is on the rise. While M&A deals between the West and China are in something of a holding pattern, EY notes: “There were 40 alliances with China in 2024, totaling $31.5 billion in value, and as of the end of 1Q2025, deals with Chinese biotechs reached $18 billion.”

The trade organization PhRMA is also sounding the alarm, saying: “Today, China is not just copying American products, it has the fastest growing biopharmaceutical pipeline in the world, driving novel drug development and completing clinical trials over 50% faster and over 40% cheaper than in the United States.” But for Kelly, these shifts only underscore the bigger theme: biotech needs to automate more benchwork to really move the needle. “We have to turn it into something automated, or else we don’t get out of this trap,” he said.

To explain what he means, Kelly reaches for a self-driving-car analogy. High-throughput screening and other fixed lab workflows are the subway: automated, but locked into the same route every time. “You run an experiment that might involve nine different devices, put a robotic arm in the middle, take a 384-well plate, put your different chemicals in there, your different drug candidates, put some cells in there, run a set of pretty complicated steps, then put it on an assay device and tell you, ‘find the drug that has the impact you’re looking for in a cancer cell,’” he said. “You can do even more sophisticated things, like look not just at whether a cell lives or dies but at the transcriptome and how it responds to your drug asset.”

That push to automate more flexible, variable experiments is already surfacing in deals such as Ginkgo’s recent ProQR partnership, in which ProQR is using the company’s Nebula autonomous lab to generate high-throughput data for AI-enabled RNA editing discovery.

The ProQR deal also lands as Ginkgo sharpens its focus. In February, Ginkgo said it would make autonomous labs a 2026 investment priority and commercialize that infrastructure through cloud lab services and customer-built systems. The same announcement recapped 2025 revenue of $170 million, down from $227 million a year earlier. A restructuring launched in 2024 cut more than 50% of the workforce.

From subway to self-driving: automating the bench

The lab bench is, essentially, the car: flexible, messy and still overwhelmingly human-driven. The next step, he argues, is the biology equivalent of Waymo. “We automated a lot of the things that were the same, and we’re going to start moving across the axis of variability,” he said.

Fellow Ginkgo co-founder Reshma Shetty also uses the self-driving vehicle analogy, presenting lab automation as a product ladder, from walk-up automation to integrated “subway” workflows to autonomous-lab “Waymos.” In the analogy, Ginkgo’s Reconfigurable Automation Carts serve as the hardware layer that makes the last step possible. In Kelly’s framing, the analogy becomes an economic argument, too: fixed-route automation is a stepping stone towards automating variable bench work, and biotech’s cost curve will not bend much until that happens.

By comparison, the self-driving car market is in some ways tougher. “The thing about self-driving cars: it’s a big market, but a really hard one to go after, because until the things are really safe, they’re basically zero value,” he said. By contrast, lab benches have a clearer path to ROI, he argued. “Eventually, I want to do any experiment automated, but along the way, even if I just take 40% of experiments and automate them, that’s a market,” he said. Automation does not need to go from 0 to 60 all at once. At a large research organization running thousands of protocols a year, that math starts to pencil out.

The self-driving comparison is useful in that the work of making a system, in either domain, operate safely within a specific operating domain varies depending on context. A driverless truck on a defined Texas freight corridor has one set of problems. A robotaxi handling dense urban pickups has another. Lab automation looks broadly similar. A 2024 paper in Digital Discovery, for instance, referred in its title to “Optimization of liquid handling parameters for viscous liquid transfers with pipetting robots” as a “sticky situation.”

On the other hand, the barriers to entry for lab automation are gradually falling away. Kelly points to generative AI coding tools as giving scientists a usable interface to robotic hardware. “The big unlock in our corner of the world with autonomous labs that AI models are unlocking is the user interface for scientists,” he said. “You don’t have to code to make the robots do the protocol you want.” Many scientists can code, but simply speeding up the process is helpful. It’s “huge,” Kelly said, pointing to the relative ease with which scientists can now program robots.

GenAI can help in ways besides coding as well. A bench scientist can describe an experiment in natural language to an LLM, the model translates it into executable robot instructions, and the run starts.

While Kelly speaks of killing the bench, Ginkgo sees the role of research scientists evolving rather than being displaced at scale. A Tufts analysis released this spring placed bench scientists among the occupations relatively insulated from AI displacement.

Instead, he points to historical precedents where automation actually expanded technical fields. Jevons paradox, for instance, describes a phenomenon in which, as a resource becomes cheaper and more efficient to use, demand for it increases. While there has been a recent period of job disruption for software developers, Kelly looks at the broader arc. “The history we saw with computers, where we moved from manual to automated, led to a lot more software developers,” he said. “Even now, as we’re automating software development, you’re seeing more growth in software developer careers, because they’re the people who understand how to use this stuff. There’s a paradox there.”

Eroom’s law and the limits of manual science

In Kelly’s view, the more urgent question is whether biotech can keep accepting Eroom’s law as normal. Kelly points to manufacturing. While there is progress in reversing decades of offshoring, it is “impossible to bring [manufacturing] back to the United States without automation,” he said.

Similarly, the research ecosystem in pharma and biotech can’t afford to preserve manual benchwork as a default. If biotech remains dependent on high-cost, high-skill labor for variable experimental work, the United States will keep losing ground to lower-cost geographies. Kelly argues that automation is the only durable way out. “If you get the human out, now we’re talking capex,” he said. “The United States is fabulous at investing capex. Look at us with the data centers.”