Autonomous Labs & AI Data
Ginkgo has spent five years and $6B failing at synthetic-biology services. It's now pivoting to sell the robots and data behind that work to AI-driven drug discovery. The potential has never been the problem; execution has.
Personal notes published for reference. Figures are sourced from company filings, releases, and press coverage as of the date above — the interpretation is mine and may be wrong.
Jason Kelly has been CEO since co-founding Ginkgo out of MIT in 2008. He's a genuine visionary and a relentless promoter, and that's the problem as much as the strength. The 2021 SPAC pitch at a $20B valuation promised growth that never came, and revenue has now fallen three years running. Credit to him for admitting the old model didn't work and cutting hard rather than slowly bleeding out. But this management team has to earn back credibility on results, not narrative.
Large, and better defined than it used to be. The new pitch is to replace the lab bench, so that most of the experimental data generated at a Merck or Pfizer comes from robots. That puts Ginkgo in lab automation, contract research (it's pricing directly against WuXi), and training data for biology AI models. Every one of those is a real, paying market today, unlike "programming cells" in 2021.
Middling. $302M of cash and securities at June 30, 2026, plus $87M restricted, against 2026 burn guidance of $125–150M. That's roughly two years of runway without new money. Burn actually rose to $45M in Q2 as the lab build-out accelerated. It isn't a crisis, but there's no margin for the pivot taking longer than planned.
The failing criterion, and the reason the stock is down ~95% from its peak. Q2 2026 revenue was $20M, down 48% YoY. The net loss from continuing operations widened to $57M and adjusted EBITDA worsened to −$36M. The new businesses are signing customers, but they aren't yet large enough to offset the legacy decline.
In early 2026 Ginkgo started closing traditional labs and turned its internal automation into the product. There are three ways to buy it.
Reconfigurable Automation Carts: wheeled robotic units that dock together into a lab that runs 24/7. Ginkgo's own Nebula lab in Boston has grown to 105 RACs running 30+ protocols a day. Enterprises and universities buy whole systems. The largest disclosed deal is a $47M, 97-instrument autonomous lab at the DOE's Pacific Northwest National Lab, the kickoff project for the administration's Genesis Mission. Caltech, MIT, Northwestern, and Maryland are also customers.
On Sept 15, 2026, Novo Nordisk bought a RAC-based autonomous lab for its new US R&D site in Waltham. Terms weren't disclosed. It's the most important customer signal yet, because a top-10 pharma is putting Ginkgo's robots inside its own labs rather than just outsourcing work to Ginkgo.
Ginkgo runs standardized assays at scale and sells the data. ADME-One is priced at $199 a panel versus $1,000–5,000 at Western and Chinese CROs, and it signed ~17 customers in its first six weeks. In mid-September Lilly's TuneLab added Datapoints as a data provider to the biotechs using its AI models.
Scientists can order individual experiments online starting around $39. It's the long-tail entry point into the same robot fleet.
This is the real question. My read: AI is the first demand driver that fits what Ginkgo actually built.
Biology AI models are limited by data, not compute. Public datasets are small and inconsistent, and pharma's proprietary data is locked away. Better models need large volumes of standardized, reproducible experimental data, which is exactly what a warehouse of 24/7 robots produces. The old Ginkgo sold bespoke cell-engineering projects that rarely paid off. The new Ginkgo sells the input that every AI-for-biology effort needs more of.
There's evidence the loop works. In a 2026 study with OpenAI, GPT-5 ran Ginkgo's autonomous lab through six closed-loop design-build-test cycles covering 580+ plates and ~150,000 data points. It cut the cost of cell-free protein production by 40% versus the previous state of the art ($422/g vs $698/g). That's an AI model doing science on Ginkgo hardware, not a slide.
Ginkgo Datapoints and Apheris (a Berlin federated-learning company) launched the Antibody Developability Consortium with founding members AbbVie, Takeda, argenx, and Lundbeck. Each contributes proprietary antibody sequences; Ginkgo runs the assays and fills the rest from public sources to reach 10,000 antibodies, the largest standardized developability dataset in the field. Apheris lets members train AI models on the pooled data without seeing each other's sequences. Data access is expected by early 2027. It's the clearest example yet of the model: rivals co-funding a dataset that only Ginkgo's labs can generate. The stock rose ~12% on the day.
Ginkgo has always been excellent at announcements. The 2021–2024 era was full of partnerships with big names that never became big revenue. Most of the new deals have undisclosed terms, and consortia, university labs, and $199 panels are small tickets. Selling capital equipment is also lumpier and lower-multiple than the platform story implies. The AI thesis is the most coherent version of Ginkgo yet, but it hasn't reached the income statement.
Total revenue bottoming and turning up, which needs tools plus Datapoints growing faster than the legacy services unit shrinks. Next would be a second and third pharma RAC purchase with disclosed dollar values, and burn falling toward the low end of guidance. Revenue that grows for two consecutive quarters would do more for this thesis than any partnership press release.
The stock went from about $7 in mid-September to about $14.70 on Oct 5 (intraday high ~$15.12), roughly doubling in three weeks and taking the market cap from ~$480M to ~$990M. Every step of the move came on a press release, and none on a revenue number. In order:
My answer: almost entirely AI narrative and partnerships, not revenue. The only hard financials are still Q2's: $20M of revenue (−48% YoY), a $57M loss, and burn rising to $45M. The one disclosed dollar figure in the whole run is the $17.5M ARPA-H ceiling, spread over several years. Nobody has revised estimates up; consensus still models ~$20M of Q3 revenue. The pattern also looks like momentum feeding itself: two of the biggest up days had no news at all, ~10.6% of shares are sold short (6+ days to cover), and the float is mostly retail. A director sold at $7.08 on Sept 15, the day it all started, and there have been no insider purchases in the past year against roughly $3.6M of sales. The week of Oct 5 made the point clearly: an $18 high, an $11 close, and a $13 finish, on two minor press releases. The partnerships are real and they fit the AI thesis better than anything Ginkgo has had before, but the stock has now priced in a revenue turn that hasn't shown up. Q3 earnings on Nov 5 are the first chance to see whether any of this reaches the income statement.
An unusually low institutional share for a Nasdaq/NYSE name this size. Most of the float is retail, left over from the 2021 hype cycle.
BlackRock, Viking Global, and Vanguard lead the list, while insiders hold ~4.6% and retail ~72%. The telling detail is that Cathie Wood's ARK once owned ~10% through ARKK and sold out completely during 2024. The most famous bull on this story gave up on it before the pivot. If the AI thesis works, it isn't priced into institutional ownership yet, which is both the opportunity and the warning.
Trackers disagree badly on this one. One puts institutions at ~24%, another at ~79%. The gap is probably index funds and how pre-reverse-split holdings are counted. The holder names line up across sources even when the percentages don't.
Whether specialist or active funds start building positions after the Novo and consortium deals. Viking is the one active holder worth tracking.
A company with a better story than ever and worse revenue than ever. Which one wins will be decided in the next four to six quarters.
If this plays out: ~$5
Legacy revenue keeps shrinking faster than tools and Datapoints grow, as it did through Q2's −48%. Burn creeps up as more labs get built, and in 2027 the company raises equity from a weak position. The deals stay small or undisclosed, much as the 2021–2024 partnerships did. The 3 covering analysts are already at Sell with a $7 target, and the stock just traded near $5.37 within the past year.
If this plays out: ~$12
Revenue bottoms in late 2026 or early 2027 as RAC sales (PNNL, Novo, universities) and Datapoints make up for the legacy decline. Burn stays within guidance and the runway stretches without a big raise. The momentum-driven part of the October run fades, but the stock holds above where it started as the partnerships turn into some disclosed revenue. At a ~$990M market cap with $300M+ in cash, the market is already pricing a lot of success.
If this plays out: ~$25+
AI-for-biology turns data generation into the bottleneck, and Ginkgo owns the largest autonomous lab fleet to supply it. Pharma buys labs (Novo is the template), co-funds datasets through consortia (the Apheris model), and routes ADME and developability work to $199 panels in place of CROs. Revenue turns into growth in 2027 on a much leaner cost base. A re-rating toward tools and AI-infrastructure multiples would pass the prior $17.58 52-week high on the way.
$13.00 at the Oct 9, 2026 close, after swinging between an $18.22 high and an ~$11 low in the same week (52-week range $5.37–$18.22, market cap ~$875M). It's still up about 85% from the ~$7 it traded at on Sept 15. Coverage is thin and bearish: 3 analysts, a consensus Sell, and a $7 average target (BTIG among the Sells; TD Cowen at Hold/$9), a little over half the share price. Q3 results are due Nov 5. That's an unusual setup, with the sell side well below a stock that retail and news flow keep pushing up. The scenario targets above are my own rough, directional estimates, not formal price targets. There's too little coverage to anchor them to a sell-side range.
Five years of missed projections and $6B of cumulative losses. The base rate for this management team delivering on a new plan is not good, however sensible the plan is.
Legacy cell-engineering revenue is being deliberately wound down. If the new lines don't ramp quickly, reported revenue can keep falling for several more quarters and test investors' patience.
About two years of cash at current burn. A raise before the pivot proves out would come at a poor price for a stock that has already done a 1-for-40 reverse split (2024).
Lab automation (Emerald Cloud Lab, established instrument makers), low-cost Chinese CROs on price, and AI companies that could build their own data factories. Ginkgo's edge is scale and a head start, which aren't permanent moats.
PNNL/Genesis Mission and ARPA-H work are meaningful early customers, and they're exposed to budget and policy swings.