Ginkgo Bioworks (DNA) 2026年第三季法說會：自主實驗室斬獲與現金消耗指引

重點摘要

• 2026 年第三季營收為 2,800 萬美元，年減 6%。前九個月營收為 6,700 萬美元，年減 37%；若不計去年同期認列的 750 萬美元非現金營收，則約年減 32%。
• 持續營業部門淨虧損縮小 44% 至 4,100 萬美元。調整後 EBITDA 改善 58% 至負 2,200 萬美元，主因組織重組降低了研發（R&D）及一般與行政（G&A）費用。
• Ginkgo 截至本季末擁有 2.85 億美元的現金、約當現金及有價證券，無銀行負債，另有 8,600 萬美元的受限制現金、約當現金及有價證券。
• 管理階層重申 2026 全年現金消耗預估為 1.25 億至 1.5 億美元。前九個月現金消耗總計為 1.2 億美元。
• 諾和諾德選定 Ginkgo 為其位於沃爾瑟姆的新研發據點建造自主實驗室，這也是 Ginkgo 首度宣布與大型製藥公司簽署自主實驗室硬體合約。
• Ginkgo 亦強調了與麻省理工學院（MIT）、加州理工學院、馬里蘭大學、西北大學、ARPA-H 及太平洋西北國家實驗室合作的自主實驗室專案，以及來自生物製藥與前沿 AI 公司對雲端實驗室日益增長的需求。

重要財務數據

Ginkgo 將其生物安全業務列為停業單位。以下業績均與持續營業部門相關。

2026 年前九個月營收下降 37% 至 6,700 萬美元。2025 年同期包含與雙方合意終止 Biomet 協議相關的 750 萬美元非現金營收。若扣除該項目，營收下滑約 32%。

前九個月現金消耗為 1.2 億美元，低於去年同期的 1.24 億美元。2026 年的數據包含因先前重組之合約而向 Google Cloud 支付的 1,400 萬美元款項。

Ginkgo 本季透過其場內股票發行計畫（ATM）募集 700 萬美元淨收益。該筆款項未計入公司的現金消耗計算中。

業務與營運表現

Ginkgo 的自主實驗室策略圍繞兩項核心業務展開：已建置的實驗室系統，以及對其自有基礎設施的雲端存取服務。

在硬體方面，諾和諾德選定 Ginkgo 在其沃爾瑟姆研發據點建造自主實驗室。管理階層稱這是其首度與大型製藥公司簽訂自主實驗室硬體合約，並預期該協議將涵蓋設備、服務及經常性軟體組成部分。

Ginkgo 亦獲選參與麻省理工學院、加州理工學院、馬里蘭大學與西北大學的自主實驗室建置。加州理工學院、馬里蘭大學與西北大學是透過美國國家科學基金會（NSF）計畫參與，該計畫資助總額接近 5,000 萬美元。

該公司作為分包商參與了一項獲得高達 1,750 萬美元資助的 ARPA-H 專案。該專案將把機器人技術應用於基因藥物製造，旨在展示 Ginkgo 系統在 GLP 及 GMP 環境下的運作成效。

太平洋西北國家實驗室已為容納 97 台 RAC 自主系統的建築動工。Ginkgo 表示，在安裝之前，公司正著手驗證新儀器並籌備各子系統。

Ginkgo 位於波士頓的 Nebula 雲端實驗室目前包含 105 台 RAC 及超過 100 種實驗流程。管理階層表示，該系統單日曾支援超過 30 種不同的流程，包含約 160 台實驗室設備且全天候持續運作。

在雲端實驗室服務方面，Ginkgo 本季新增了 20 家生物製藥與生物技術新客戶。管理階層亦表示，公司正在為數家前沿 AI 實驗室生成生物與化學數據。

Ginkgo 與創始成員艾伯維（AbbVie）、argenx、靈北製藥（Lundbeck）及武田製藥（Takeda）共同發起抗體可開發性聯盟。參與者提供抗體設計，並由 Ginkgo 進行建構與測試，以在聯邦學習環境中建立共享資料集。

管理階層財測

管理階層重申 2026 全年現金消耗預估為 1.25 億至 1.5 億美元。該區間反映了在維持支出紀律的同時，持續對 AI、機器人、軟體以及波士頓 Frontier 自主實驗室擴建進行投資。

該公司未提供營收財測或 2027 年展望。管理階層表示，預計將在 2 月重新調整 2027 年預期並提供財測。

在自主實驗室銷售方面，設備營收預計於系統安裝完成時認列；服務與軟體授權則預計在系統的運作生命週期內產生經常性營收。

風險與關注事項

• 營收仍低於去年同期，2026 年第三季下滑 6%，前九個月則下滑 37%。
• 第三季營收受惠於一項大型長期合約在修改工作計畫後的累計補認列（cumulative catch-up），加速了部分營收認列。
• 調整後 EBITDA 仍為負 2,200 萬美元，而前九個月的現金消耗則達 1.2 億美元。
• 單季調整後 EBITDA 成本中，約有 1,400 萬美元與未使用的租賃空間相關。管理階層表示，這些成本或可透過轉租緩解，但該空間目前並未產生任何營收。
• 自主實驗室訂單通常需要 6 至 12 個月才能完成安裝，導致客戶承諾與設備營收認列之間存在時間差。
• Ginkgo 作為較新的廠商進軍實驗室自動化市場，必須與已和大型製藥公司建立深厚關係的老牌機器人供應商競爭。

分析師問答重點

雲端實驗室變現模式：管理階層預期最大商機將來自難以從傳統委託研究機構（CRO）獲得的複雜檢測與資料集。範例包括抗體可開發性、功能基因組學、CRISPR 干擾及小分子干擾數據。

自主實驗室營收認列時程：設備營收通常在安裝完成時認列。目前建置約需 6 至 12 個月，具體取決於實驗室規模、場地準備情況以及是否需要無塵環境。Ginkgo 預期，隨著製造日趨成熟，其標準化 RAC 設計將縮短整合時程。

典型系統規模：管理階層表示，初期自主實驗室可能包含約 10 至 15 台 RAC，而相當於數個傳統高通量工作單元的系統則可能使用約 30 台 RAC。

AI 與實驗室自動化：Ginkgo 看到 AI 相關需求的兩大潛在來源：生成用於訓練科學模型的生物數據，以及提供機器人實驗室環境，讓推理模型能在其中學習設計與執行實驗流程。

ADME 服務定價：管理階層將 Ginkgo 的定價優勢歸功於自動化降低了技術人員與實驗室空間成本。公司亦發現製藥客戶希望將分子合成與檢測分析分開進行的需求，以避免敏感化合物與效能數據過度集中於單一供應商手中。

完整電話會議逐字稿

完整財報電話會議逐字稿

管理層陳述

Unknown Executive

I'm joined by Jason Kelly, our Co-Founder and CEO; and Steve Coen, our CFO. Thanks as always for joining us. We're looking forward to updating you on our progress. As a reminder, during the presentation today, we will be making forward-looking statements, which involve risks and uncertainties. Please refer to our filings with the SEC to learn more about these risks and uncertainties, including our most recent 10-K.

Today, in addition to updating you on the quarter's results, we're going to show how Ginkgo is now 2 businesses built on 1 autonomous lab platform at a moment when the U.S. government is starting to treat autonomous labs as national infrastructure. We'll share how our Autonomous Labs business is growing, including our first autonomous lab sale to a pharma company, Novo alongside new government wins with ARPH and NSF. We'll also share how our cloud lab business is growing with new pharma logos, including 22 customers this quarter and with AI data generation for Frontier labs.

As usual, we'll end with a Q&A session, and I'll take questions from analysts, investors and the public. You can submit those questions to us in advance via X, hashtag ginkgoresults or e-mail, investors@ginkgobioworks.com.

All right. Over to you, Jason.

Jason Kelly

Thanks, Daniel. As we always say, our mission here at Ginkgo is to make biology easier to engineer. Over the last few years, we've been going through a transformation with this focus on autonomous labs, and that was really our goal in 2026. I would say I think that succeeded very well. And this quarter is really good evidence of that. We've been focusing our efforts on moving more and more of our internal lab work on to autonomous labs that continues and improving the product. But real breakthrough in the last quarter in terms of external customers adopting our systems. And so we'll be happy to talk about some of the new customers we have for autonomous labs, both in universities and the biopharma sector, and we've been expanding the services that we're offering on top of our big system in Boston here in Nebula. So really happy about this. I think we made the right choice in banking on this technology. So you're going to hear more from me on that in the strategic section.

Also, our financials have been going in the right direction. We made a big push to cut back on spending while still making major investments in research and development around autonomous labs, AI and robotics in the laboratory, and that has also paid off. You can see our year-over-year change here, pretty substantial improvements in net loss from continuing operations as well as our total adjusted EBITDA. So this is what we were aiming to achieve. And really congrats to the team on pulling this off. It's hard to do these 2 things that wants cost cutting and being really the technology leader in an emerging tech area, especially one next to AI.

All right. Last thing I'll mention, we have still quite a nice cash cushion. And this is, again, thanks to those cost-cutting efforts. $285 million in cash and cash equivalents, with no bank debt another $86 million on top of that, that we keep as restricted cash for some of our government projects and leases. So overall, I like our position for a cash position. I like where we are with cash spending. And as you're going to hear from me in the strategic section, I think we're doing a nice job externally selling and improving really a new category of product in autonomous labs. So back to me in a moment, but I'm going to hand it to Steve to a little more depth on the financials.

Steven Coen

Thanks, Jason. Before I walk through our financials, I want to remind everyone that our biosecurity results are classified as discontinued operations within our financial statements, where we have and will retrospectively recast all prior periods presenting to come forward to this presentation. All the financial information I will cover today relates exclusively to continuing operations where we operate as a single segment.

With that, I will now discuss our Q3 results. As Jason mentioned, revenue was $28 million in the third quarter of 2026 down 6% compared to the third quarter of 2025. For the 9 months of 2026, revenue was $67 million, down 37% compared to the same period last year. As previously disclosed, revenue in the first 9 months of 2025 included $7.5 million of noncash revenue relating to the mutual termination of the Biomet agreement. Excluding this, revenue for the first 9 months of 2026 was down approximately 32% from the prior year period.

As noted in our release, revenue in the third quarter was higher than last quarter, and this was primarily from a cumulative catch-up on one of our larger long-term contracts, which was modified to realign the work plan, resulting in us recognizing some revenue soon. It is important to note that our net loss includes a number of noncash and other nonrecurring items as detailed more fully in our financial statements. Because of these noncash and nonrecurring items, we believe adjusted EBITDA is a more indicative measure of our profitability. A full reconciliation between adjusted EBITDA and GAAP net loss from continuing operations can be found in the event.

In addition, in discussing our R&D and G&A expenses, we exclude stock-based compensation and other separately captured items. In the third quarter of 2026, R&D expense decreased 48% from $51 million in the third quarter of 2025 to $27 million in the third quarter of 2026. G&A expense decreased 32% from $14 million in the third quarter of 2025 to $9 million in the third quarter of 2026. These decreases were primarily driven by our restructuring efforts, which was potentially completed by the end of 2025.

Moving further down the page. Net loss from continuing operations was $41 million in the third quarter of 2026 compared to a loss of $73 million in the prior year end, a 44% improvement. You will note that adjusted EBITDA in the third quarter 2026 was negative $22 million compared to negative $53 million in the third quarter of 2025, a 58% improvement. It is important to note that adjusted EBITDA includes the carrying cost of excess lease space which you can see was approximately $14 million in both periods. This cash operating cost represents the base rent and other charges relating to unused leased space, net of sublease income. That is not drive any revenue right now and can potentially be mitigated through subleasing.

And finally, cash burn in the third quarter of 2026 was $27 million compared to $28 million in the third quarter of 2025. For the 9 months of 2026, cash burn was $120 million, down from $124 million in the same period last year, a 4% decrease. As previously reported, we paid Google Cloud $14 million earlier this year as part of the favorable restructuring of that contract late last year. Excluding this payment, cash burn reflects a significant decrease from the first 9 months of 2026 compared to the first 9 months of 2025, which was directly a result from the restructuring, partially offset by the investments we are making.

During the third quarter, we raised an additional net $7 million through our at-the-market equity program. Consistent with our methodology, these related proceeds are excluded from cash burn for all periods presented. As previously reported, cash burn excludes the restrictions we made last quarter to secure a bond on our P&L contract.

We ended the quarter with $285 million in cash, cash equivalents and marketable securities and no bank debt, plus $86 million of restricted cash, cash equivalents and marketable securities for customers and operating activities.

Now turning to guidance. As we discussed earlier this year, 2026 is about continuing to be cost efficient, while investing in our AI robotics and software to bring autonomous labs to our Bioscience customers including the build-out of our Frontier Autonomous lab in Boston. For these reasons, we believe cash burn best reflects our continuing services and tools and further investments in autonomous Labs.

In terms of the outlook for the full year, we are reaffirming our overall cash burn guidance for 2026 totaling $125 million to $150 million. This range reflects a firm balance amongst cost efficiency, continuing services and tools and further investments we are making.

In conclusion, we are pleased with the continued improvements in cash burn efficiency in our business pursuits for 2026.

And with that, I'll hand it back over to you, Jason.

Jason Kelly

Thanks, Steve. Our mission here at Ginkgo is to make biology easier to engineer. As I mentioned, we've been doing that by expanding our technology in autonomous labs and bringing those to market. I will say they are currently having a movement. Big push on the government side, a lot more attention from private sector biopharma. And as I'll talk about it even frontier AI labs are getting excited about the potential of robotic laboratories. We sell those in 2 different ways, both hardware. We'll install an autonomous lab for you or you can leverage our autonomous lab as a cloud. And so those are going to be my 2 sections. You're going to hear about sort of the blue chip early adopters we're seeing on selling our systems as well as the growth in the services.

All right. But it is a moment. So I do want to just highlight a few of the good things that have been moving in our favor over the last quarter. Just a few weeks ago, I was at President Trump's Golden Age of Science Summit. This was to announce a report out of the Office of Science and Technology policy. I've mentioned this before, but there's a little clip in there from the report about accelerating autonomous experimentation. And we've seen real money coming behind this $400 million going to the National Science Foundation to invest in this.

The Genesis mission, which is partially what funded our big project at the Pacific Northwest National Lab. I think you will continue to see more of this coming from the White House as we expand AI and science in the United States. But importantly, our rigor customer base, ultimately, I think, is the biopharma companies, Dave Ricks, the CEO of Lilly, posted on X just a couple of weeks ago, AI's going to help us conquer disease, but this is going to take time, effort and lots and lots of data that doesn't exist yet. And I would add that robotic labs will begin to make scaling data more tractable. That's a big deal.

Robots, just if you don't know, it have been sort of relegated to the basements of biopharma companies. They run high-throughput screens, some diagnostic testing at the diagnostic labs runs on robotics. But by and large, robots have not been a major part of the overwhelming laboratory investment at pharma companies, which is really manual labs. And so it's exciting to hear that from the CEO of Lilly, I think Dave is great. We just sold as is of de novo, which I'll talk about, but Dave should probably come by and visit our site in Boston.

Anthropic also, if you can believe it, announced that they had set up a biology lab in San Francisco as part of their AI drug discovery efforts. Super exciting. That is not a robotic lab yet, but I do think -- if you look at the way that those companies think about the world, it's very natural to go for models that can ultimately run the lab experiments themselves, which would be robotics.

So okay. As I mentioned, we have 2 ways to buy the labs, either by -- you can either get your own installed or you can use ours in the cloud. Alongside that, we had just had 2 great events in September, one for our automation group and people that wanted to buy the systems and then also for our Data Point Summit just the next day. And the reason I mentioned this, we did these back-to-back, there was maybe about a 10% overlap in the crowd between the days. It's really -- there's a real nice additive effect of Ginkgo running services using our robotics infrastructure because we are able to learn faster than we would if we were just building systems for external customers. And so I think that was a wise choice of setting the business up that way. We will continue to do that. And I actually want to migrate more and more and more of our services on to an ever bigger Nebula here in Boston, and you'll see more of that in the next year.

Okay. Let's talk about those new customers because this is the thing I'm most excited about, a number of blue chip early adopters. I'll start with the universities. So we are very fortunate to be selected by MIT, Caltech, Maryland and Northwestern, Caltech Maryland Northwestern is part of this NSF program I mentioned to build out. In the case of the NSF, there's close to $50 million flowing through this program, autonomous labs at those schools. We also have a couple of other schools that we're talking to where it's not coming from a grant. So I think there's sort of 2 ways we're going to sell to the universities either as part being a subcontractor equipment provider on a grant. Or directly from the infrastructure budget, kind of the overhead budget, the department budgets in, say, a major biology R&D department at a large school. So I see both of those roads working out. I think we will continue to sell systems into academic environment. That's super exciting.

A lot of the most exciting new protocols are developed in academia. I'd love to see that work happening directly on autonomous labs from the beginning. I want to see grad students do their whole bioengineering PhD without touching a pipette. These are dreams that we've had for a long time at Ginkgo. It's going to be really fun to see that taking off and what an awesome set of first customers for us in the university space.

The other thing I'm very excited to announce, Novo Nordisk selected Ginkgo to build an autonomous lab at their new R&D site in Waltham. Very excited about this. This is our first sort of big pharma announcement of an autonomous lab. That sends a signal to all the other companies in the industry. This is hardware. This is infrastructure. There's a big switching cost for people to go with new vendors or to try new players. There are a number of existing kind of laboratory robotics companies, and I won't give my usual comparison of the Waymos and subways. But there's a lot of subway sellers who sell robotic systems that do the same thing over and over again that have a lot of existing relationships with the large pharma and a lot of trust.

We're coming in as a new vendor. We got to have something new to offer if we're going to unseat some of those existing vendors. And what we're offering is a system that can do variable work, something closer to what you get from your lab ventures. That's why I call it am wamo. And so it's really exciting to see Novo Nordisk valuing that enough to bring us in as a new automation vendor. We obviously have a great relationship with them on the services side, too. But this is the first hardware contract we have with them.

The other thing I was really excited to see ARPA H selected us as part of, again, a subcontract on a $17.5 million up to $17.5 million award. This is very cool because this is using robotics for manufacturing of genetic medicines and these sort of if you're familiar with like the Moderna, Merck partnership around cancer vaccines. These are a vaccine, but it's being built just for you, so it's sort of a custom manufacturing process. That's a great fit for robotics and especially Ginkgo's robotics. And so what I think is great about this is we're going to, through this grant, start to demonstrate our robotics in GLP, GMP environments. And so that could be very exciting for ultimately having these as part of people's manufacturing processes in biopharma. So stay tuned on that, but a very exciting project.

Last but not least, I've mentioned this before. We were selected to build out a 97 RAC system for Pacific Northwest National Labs. I just want to say congratulations to our partners at PNNL, they broke ground just a month or so ago. On the new building that's going to contain this big system. So we're working away building and validating some of the new instruments that they want on there that we haven't put on RACs before and building some of the subsystems for the big install so that the minute that building is the door is open, we can roll our RACs in. So very excited that they're doing that.

Okay. And maybe on that note of sort of building these systems. This is just a video from our lab here in Boston. And the thing I'd like to convey is we want the experience of adding new equipment to an existing robotic lab like so we put in a robotic lab say, Caltech, it's going to initially be 5 or 10 RACs or something, right? Within in this case, less than a day, you can add a new device onto that RAC system. It's very similar to if you just a new piece of equipment for your manual lab and put it on a bench. It's not a big project. It's not a giant 3D modeling exercise or anything else. You can just add it quite quickly, turn it back on, and now you have a new capability with all of the other equipment you already have on the system. That's again, unique to our setup. Nobody else's stuff does that with anywhere near our robustness or scale.

And so I'm really excited that as we enter all these places, we will -- customers and we've already seen this, will want to grow those systems and they can do that incrementally. It's not like another giant project to bring them in, they can just order another 1 or 2 pieces of equipment that they might want to add. So that's very exciting for how we'll grow once we're in at these customers.

We, as I mentioned earlier, because we run our services, we actually have the largest autonomous lab in the world, 105 of our RACs here in Boston. Customers can use it as a cloud lab, but very importantly, we are able to learn a lot faster because we are developing on this system. So when PNNL bought 97 RACs, we expanded this from 50 to 105 because I wanted 18 months in advance of them getting a big system, I wanted us to have tried everything, found all the bugs, squeeze them out, right? Nobody's really pushed autonomous labs and robotic automation and labs to the scale Ginkgo has. So we want to make sure we get there before our customers do. That's the purpose of Nebula.

And frankly, it is amazing what we've been able to show on the system. We're doing more than 100 protocol. These numbers are actually kind of dated now, but more than 100 protocols, more than 30 unique in a single day. 100-plus down, I think it's 160 lab devices on that setup and it's running 24/7, right? So that is -- again, that scale of devices, that range of variability, especially has not been shown on a laboratory robotic system previously. So really excited about this.

And maybe the last thing I'll make. We started to look at like how many equivalent technicians would it take to like run around and move the plates onto the the different devices like a plate moves across the room, a person would have had to kind of walk between those 2 pieces of equipment in a manual lab. And if it spins at a centrifuge for 5 minutes, the person would have 5 minutes kind of staring at that centrifuge.

So we add all that up. And our peak day so far has been 500 technician hour equivalents there. But like an average, they were up more like 200 to 300. So that's pretty good. I mean, if you think of it as a 5-hour in the lab sort of per technician that's something like 100 -- 50 to 100 technicians capability, right? And then the question is like, how do you design experiments for 50 to 100 technicians to be running on automated worth of work happening across these devices, and that's where AI comes in.

I'll talk about that later. But I think there's a big overlap between AI meeting autonomous labs. And in fact, autonomous labs needing AI. So I'll touch on that. Okay. But either way, if you come visit, don't take my word for it. There's a sign-up link here. We've now probably more than 1,000 people come visit the lab now. So you should definitely sign up, it's great. We do tours every week. Seeing is believing on this. It's really wild. Okay.

Lastly, I want to talk about our cloud lab services. These have been growing really well with biopharma customers where we're operating really like a CRO and large data gen. And then we're also seeing a little bit from the frontier model companies, which I'll talk about.

Okay. So love this post as opposed to 1 of our customers up on X. One of our agents flagged that Ginkgo data points has been outperforming others for protein biophysical data production, ahead on cost and quality as well as tax. So cheaper, faster and higher quality. And they use their AI and just sort of compare us to some of the other vendors they've been using. I think this is going to surprise people.

When you do things more robotically, they can be competitive with work. We've been offshoring to low-cost scientific labor in China, frankly, over the last 30 years in the pharmaceutical industry. And so if you look at, for example, our ADME services, I can go data points, for Tier 1 ADMEassays. These are very common assays being used as part of a small molecule, just drug discovery every week. If you want those same assays from Wuxi, it's like $1,000 or more depending on what you're -- how you're doing it.

And so that's really exciting. I think this is the potential of laboratory automation. And by the way, I think it applies for automation around a lot of sort of small batch, I'll call it, manufacturing in the United States. If we want to bring it back. This is how you're going to do it. But it's working. I mean, we're -- we added 20 new logos to Ginkgo. Like these are biopharma biotech customers that have never engaged us, whether on the equipment side or our old services solutions type services that we did in the past, just new customers in the last quarter.

So this is really exciting and people should try it out, and it's cheap to try. So give it a shot. The other thing I'll mention on the data point side is we launched our antibody developability consortium. So there's been a lot of work in antibody binding. So there's a few different companies, Swift's been doing this adaptive, others, we offer this as well where you can synthesize an antibody and see it to buying something. And that's great and that people have been training really large from kind of frontier bio models on protein design for binding things to use in drug discovery.

But that's not all you need to know to make an antibody drug. You also have to track how developable it is. In other words, does it make for a good drug? Is it soluble, right? Like is it going to all just like lump up in your blood. Is it immunogenic? Is it going to freak out your immune system. There's a whole list of assays that you run on antibodies that aren't just about at binding. You can have an antibody that binds great, that ends up being a bad drug. And so as we're starting to win on the binding side with these models, we believe the other key assay that everyone is going to want will be developability. And this is a much harder thing to do, [indiscernible], but it's much more complicated assays, more steps. It really benefits more from what we've been doing here, at Ginkgo on the integrated robotics side.

So we have this up and running. It's really great. Marshall is the best in the industry. I'm sure -- and we have launched AbbVie, Argenx, Lundbeck and Takeda as our sort of founding members of our antibody developability Consortium. They're each submitting antibody designs. We build, test and give back and make this big developability data set. What's really cool is that then gets with our partner, a Ferris put into a federated learning environment where nobody gets to see each other's designs but everybody gets to train models on the full data set. I think this is a super exciting way to again, these are more expensive assays than binding. Binding is pretty cheap, being able to do developability is harder. And so this is a way to make that cost lower on a per company basis, even though everyone is generating a lot of data. So we're really excited about that. If you're in this consortium of your biotech or biopharma listening, you should definitely reach out. If you add yourself as a new member, you'll get access to historical data as well. So very good opportunity, and I really like this model.

Also, we announced in the last quarter a partnership with Lilly on lab, where our kind of suite of Ginkgo data point services is available to members, these are a bunch of start-ups and others that are kind of affiliated with Lilly to and labs, just making that super easy for people to order our services.

Okay. Last thing I'll talk about. The question I get asked the most by investors is are the Frontier AI labs buying biodata. And the answer is yes, yes, we're working with a few of them. The -- what I would say is what you're seeing is a realization that the smartest models are going to have value in a bunch of these advanced engineering and science areas like things like chip design, things like drug discovery, things like new materials and energy. And so -- but in order for them to get good at that, like they got good at math, you need to tell them if they're doing a good job with their ideas, right?

So the reason things work so well in chess and math is the model can come up with something and then go get validated, reinforcement learning and see, hey, is my math proof right? Is this code I wrote, right? Did I play a good game of chess. And the beautiful thing is you can answer all those questions inside of a data center, right? So those are like computable answers. That's not true in a lot of these other areas. And so in these other areas, what you're going to want is basically the feedback of the laboratories so that you can make your model smarter. And so as long as I think the Frontier AI labs want to keep pushing the boundary of their capabilities in these physical areas, I do think they're going to want to train on physical data.

In our case, biological and chemistry data. So super excited about this. I do think it's a very cool area, and we're happy to be participating in this along with others. All right. I'm going to take questions about all of that. As always, we should grow the world we want to see. We appreciate your interest and your investment. Those of you who are investors in Ginkgo, and my e-mail is up there, if you ever want to email me and ask questions offline. All right. Thanks a lot.

Daniel Marshall

Thanks, Jason. As usual, I'll start with the question from the public and remind the analysts on the line that if you'd like to ask a question, please raise your hands on Zoom, and I'll call on you and open your line. Thanks all.

Daniel Marshall

First, just want to thank everyone for sticking with us through those technical difficulties earlier. We're extra excited to be there with you tonight. I want to start with a question from Brendan who centered in from TD. He actually sent 2 questions. We'll start with the first.

How should we think about the evolution of the cloud lab offering and how Ginkgo plans to monetize this moving forward? Is the idea to ultimately become a kind of AI model training data generation service? And if so, what kinds of data is Ginkgo able to supply the customers might not be able to get elsewhere?

Jason Kelly

Yes. I can take that. So yes, the -- I talked about this a little bit at Dan there around the antibody developability consortium. So I think -- the sweet spot for Ginkgo on this. And by the way, I'd say this is for Cloud Lab data points are basically our services that we run on top of the platform. And so we think for these particularly complicated assays, that's a really strong fit where we can generate data that you're going to have a hard time getting from other CROs.

So things like binding, there's now a lot of players doing that. I think we're our hat is in the ring there, that's fine. I think we'll be able to play there. But things like developability is really not. It's much more sparse because those are harder to do. The other area is in target discovery. So -- we offer our functional genomics offering where you can take your favorite cell line, and we can make either genetic pregrabations. So like CRISPR edits to the jeans. You want to say knockout every gene in that genome 1 by one, we can put those in wells and then give you a functional genomics, so like an RNA readout of what's going on inside that cell. Or we can add small molecules. So we can do a small molecule perturbation by adding thousands of those to the cell line.

That sort of stuff is much more complicated. And so those are the types of things, I think, as PAUSE we get past this like initial phase where if you kind of go after the easy stuff and it's good for storytelling and you move more into the phase of, boy, I would love to have the models that ultimately give me all the data I need to develop a drug or to find a new target against a disease that's been resistant to having a good target, those types of big activities that happen across the industry, I think AI is going to play there, too. And that I think will be quite favorable for us. I think we'll be unique in that respect.

Daniel Marshall

All right. Second question for Brendan. And I'll just remind all the analysts on the call. If you have a question, just raise your hand, and I'll call on you after this. So the next question is, what is the expected time line to reporting revenues from the sale of the RAC and catalyst systems from both the recently confirmed partnerships and any potential future deals?

Jason Kelly

Yes. Steve, do you want to talk about that a little bit?

Steven Coen

Yes. I just want to remind anyone, we're not giving guidance on revenue, and we're not in to guidance on 2027 yet. I think at the right time, we will bring the investors along as we feel it's important to know. We'll be resetting expectations for 2027 in February and our guidance. And we will look at although not that information is meaningful to that at that point in time. So -- and then we'll begin to our.

Jason Kelly

Yes. The other thing I'd add is as we build these systems, so we'll recognize revenue when the system is installed, right? So that's when we would get the sort of like CapEx revenue for the system. And then we have on an ongoing basis in these sales, we have a sort of services and SaaS, software licensing that's recurring. And so you'll see those over the life as we get a larger fleet deployed, that will sort of ramp up and stay up, hopefully, as people keep these systems for a long time, we hope.

Steven Coen

Yes, that's exactly right, Jason. As we mentioned in the last meeting, one before, it is multifaceted, delivering equipment, lodge deployed. That takes a good bit of time depending on size of the lab or is it CleanSpace lab. And then, of course, we have all the front-end services and then the back-end services and on the back-end services are more SaaS like. So that's why it takes time to build on a lab. That's why we'll be looking at it, and we'll provide the guidance when it's...

Jason Kelly

Yes. Maybe the other thing I would add, we are taking a different approach than people have traditionally taken to integrated automation. So the normal mental model, again, for like a higher as or Thermo or Biosero, the folks that build these subway work cells is that each one is unique to a customer. So as a customer saying, I have this protocol, I want to run at high throughput and the vendor is going to basically do some nonrecurring engineering generate a 3D model, design the system, build it all for you over a series of months, and then install it.

And so there's sort of a 6- to 12-month period where there's like kind of custom engineering happening? That's not how the RACs work. So the good thing about our system is we're producing rack enclosures every week. We can accumulate an inventory of RAC enclosures so that over time, cut orders come in, and we just need to integrate equipment and ship them. And so I do think what will happen as we get more mature in this area is the sort of time it takes to get a system installed will speed up relative to what you've seen in historical integrated automation just because our PAUSE offering is more productized. If It is a standard thing that we're manufacturing on the regular rather than a system we build one off against the customers back.

Daniel Marshall

Thanks, Jason. All right. Next question. Can you give a little more color on the AI Frontier labs ordering biodata, -- do you think that is durable? Does that help your business with traditional biopharma companies in any way?

Jason Kelly

Yes, it's super interesting. I mean I think there's a little bit of a mature thesis on open and entropic getting into the pharmaceutical industry, and there's a lot of talk about that. You're certainly hearing about it a lot from Dario, for example, obviously, entropic invested in that wet lab. So that's sort of the -- I would say, the big overlay. What PAUSE -- what I'm most excited about, I think there's like 2 lanes. There is generate bio data to train models that understand biology.

So that's your -- the protein models that you've seen from people that virtual cell initiatives, the types of data we were just talking about with the developability consortium. This is all biological data training biological models. We're like alpha fold as an example from Google was an example of a protein biological model. The other thing we have, obviously, is the reasoning models. And the reasoning models I'm quite excited about because I believe they will be able to operate robotic labs, and they essentially need to learn to write protocol code for lack of a better word.

So you can think of them as a coding agent, programming robots and laboratory protocols to do science. And that's also a type of training data that we expect labs are going to want to purchase, and we are excited to see models get better at doing that. And the reason I'm excited about that as I watch our system at Ginkgo, Nebula and I look at protocols coming on to that system. And it's not perfect, but it looks a little bit like a workday, right? So scientists are coming in, in the morning, they're submitting new protocols based on their project. You see our number of protocols running simultaneously going up. Mid-day, it's quite high and the opportunity is high and then by like 9:00 at night, it's starting to drop and buy 3 in the morning, there's a tenth as many protocols running as we're running at noon.

That makes basically no sets, right? You have an autonomous lab. It's running 24/7. There's no reason you wouldn't want to utilize that asset. And so what I want to get to is increasingly stuff like we've been doing with Open AI in the announced project, a genetic generation of protocol so that really just to keep the whole thing busy. And I think our customers when they buy our autonomous labs over time, they're going to have that same need. They're going to want a genetic science happening on the system because it feels like a waste not to be using it all hours once you have a system not going to actually do variable work all hours. And that's just a change, right? Today, the manual labs run 40 hours a week. Not 168 hours like the autonomous lab. So filling that is, I think, going to require some new tools. And so that's the -- those are the 2 areas.

That's why I say the AI companies are interested in autonomous labs for data generation, but the autonomous labs are also interested in frontier AI reasoning for the purposes of protocol coding. We need each other.

Daniel Marshall

All right. Next question. What size of autonomous labs are customers starting with? And how are you all thinking about product moving forward?

Jason Kelly

Yes. So sort of I would say we're at like kind of -- if you're building a small system, it's harder to be smaller than like -- and something more like 10 to 15, I think, is like a healthy like initial system. And then if something is closer to what you'd normally buy a work cell like if you were going to go buy a system to do some hydro put screening or would be a couple hydro screening systems, that's more like 30, just to give you a ballpark. And so one of the things I'm excited about on the product side is sort of multi-purpose work cells, right? So a work cell that could do 10 or 50 different protocols so that instead of having to buy 3 work cells to do 3 different things, you could just combine all of that into one autonomous lab that does what you would have previously used 3 work cells to do.

That gives us actually an advantage even in the work cell space before you go all the way to the lab bench Waymo. And so I'm pretty excited to just demonstrate to more pharma partners -- what's possible there because we're certainly seeing gains of that sort when we run our own services internally at Ginkgo.

Daniel Marshall

All right. Next, we have how is Ginkgo able to price ADME services and data points below Wuxi? And how durable is that cost advantage as competitors respond?

Jason Kelly

Robust. I think there's just a lack of awareness of how much of the cost when you're doing manual lab work is in the technician doing the work and the lab space, you need to do it. And that's true in China, too. I mean it's not the sites are less expensive, but they're not 10x less extensive. They're more like half. And so there is actually a lot of room. As we get closer to the cost of the services being majority reagents, then I think there is less of a reason to offshore.

Also, altering is expensive, right? You're kind of dealing with us, like -- so shipping and the timing and all this stuff, right? Like the -- so there's a -- it's a pretty interesting moment right now. You're also seeing a lot of companies wanting to split where there used to be a moat that Wushi had, which was hey, you're making the molecules there, you should test them there because that's more convenient and faster like they're doing the chemical synthesis, which is still fairly resistant to automation. Why not just take it down the hall and do the assay. And it turns out the reason not to is if the same organization knows the molecule structure, which is like secret IP of the pharma company, plus which of the 10,000 molecules you tried performed really well as a drug. Well, that's all you need to walk out the door and start a competing biotech company in China based on that molecule.

And so for -- and because you're seeing this now more and more of these companies that are started, it's like, wow, it looks a lot like the molecule that was shipped to us a week ago. That is driving companies to split where they do it. So they're doing molecule synthesis one place, shipping plates molecules, not saying what the molecules are to a different place to do the assays, just so that, that information is not held by one party. That's good news for us, right? So great to your chemical synthesis wherever you want and ship it to us for the assays, I'm happy to do it.

Daniel Marshall

Thanks, Jason. All right. There's 1 more. It's somewhat similar to 1 of the ones that Brendan asked earlier, but I'm just going to ask it again in case there's sort of anything else to add. This one is we saw a lot of inbound interest this quarter in our autonomous lab offerings. And so the question is like how should investors think about converting that interest into revenue and kind of over what time line should they be thinking about?

Jason Kelly

Yes. I think it's pretty close to what we covered before. Again, I think we want -- we'll be as public as we can, not every customer wants us to announce it, but we'd like to announce that we sell systems because it helps improve our brand. So you'll see us announcing as we get systems sold. And then I think you're going to see a period of time, again, it depends on the customer. They're often expecting a system installed like 6 months later or something.

Right, because they're 6 to 12 often because the normal cycle they're on is like, okay, we're planning well in advance of this because that's what it normally takes to get an automation system in.

And so we've got a room and it's going to get cleared out or we're building a building and it's going to open, and I'm on the team that's putting in the automation. So there is that kind of lag today. Again, I don't know what will happen as we start to if I can be like sure, but I've got these enclosures. And if you have the equipment, I can integrate it in about a month, and they could just install, I don't know if that would like to change the dynamic or people are sort of like, no, that's okay, like I already started asking for this with an RFP 6 months in advance anyway.

So please bring it in 6 months.

So we'll see. But I would say that's pretty typical 6 to 12 months.

Daniel Marshall

All right. I think that's all the questions that we have for today. I just want to thank everyone again for joining us and remind you that you can always reach out to us at investors at gincobioworks.com. Jason also shared his e-mail earlier, Jason Gigobioworks.com, in case you want to contact him directly. And yes, thank all so much, and you'll hear from us next quarter.

Jason Kelly

Yes. Thanks, everybody.