Ginkgo Bioworks Holdings, Inc. (DNA) 2026 Q1 法說會逐字稿

Good evening. I'm Daniel Marshall, Senior Manager of Communications and Ownership at Ginkgo. 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.

各位晚安。我是 Ginkgo 的傳播與股權資深經理 Daniel Marshall。與我一同出席的有我們的共同創辦人兼執行長 Jason Kelly，以及財務長 Steve Coen。感謝各位一如既往地加入我們。我們期待向各位更新我們的進展。

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 results, we're going to provide insight into how and why we see autonomous labs like Nebula, our autonomous lab, replacing the lab bench, which is where nearly all of biological science is done today. As usual, we'll end with the 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. #GinkgoResults or e-mail, investors@ginkgobioworks.com.

提醒各位，在今天的簡報中，我們將作出前瞻性陳述，其中涉及風險與不確定性。請參閱我們向 SEC 提交的文件以了解更多這些風險與不確定性，包括我們最新的 10-K。今天，除了向各位更新本季業績外，我們也將分享我們如何以及為何認為像 Nebula（我們的自動化實驗室）這樣的自主實驗室，將取代實驗台——而目前幾乎所有生物科學研究都是在實驗台上完成的。照例，我們將以問答環節作結，我會接受分析師、投資人及大眾的提問。您可以透過 X 以 #GinkgoResults 事先提交問題，或寄送電子郵件至 investors@ginkgobioworks.com。

All right. Over to you, Jason.

好的。接下來交給你，Jason。

Thanks, Daniel. We always start with this. Ginkgo's mission is to make biology easier to engineer. And I mentioned this at the last earnings call, but in 2026, our focus will be on investing to win the category of autonomous labs. And I'm really excited, even since we just spoke a few months ago, this category has really been growing in attention, new companies in Silicon Valley pursuing this, a lot of interest from the AI Frontier labs about the application of AI models in science via autonomous labs.

謝謝你，Daniel。我們一向從這裡開始。Ginkgo 的使命是讓生物工程更容易。我在上次財報電話會議也提到，到了 2026 年，我們的重點將是投資以贏得「自主實驗室」這個類別。我非常興奮，因為即使距離我們幾個月前才剛通話，這個類別已經獲得更多關注：矽谷有新公司投入其中，AI 前沿實驗室也對透過自主實驗室將 AI 模型應用於科學領域表現出高度興趣。

Government talking more about this. So I do think we're on to the right track with this focus for the company. The two big ways I'm going to be pursuing that goal in 2026, the first is to take our services in solutions, in data points and cloud lab and run them on top of our autonomous lab here in Boston that we call Nebula. That's a chance to prove out the capabilities of our system with real-world activities. And then the second big area of activity will be getting early adopters of autonomous labs out in the world to buy our systems like we've done already with Pacific Northwest National Labs that I talked about last time.

政府也更多在談論這件事。因此我確實認為，公司把重點放在這裡是走在正確的軌道上。2026 年我將以兩大方式來追求這個目標：第一，是把我們在解決方案、資料點與雲端實驗室方面的服務，運行在我們位於波士頓、稱為 Nebula 的自主實驗室之上。這是用真實世界的活動來驗證我們系統能力的機會。第二個主要工作領域，是讓全球的自主實驗室早期採用者購買我們的系統，就像我上次提到的，我們已經與 Pacific Northwest National Labs 做過的那樣。

So excited to pursue both of those, and you're going to hear more about it from me in the section. We also -- in the last quarter, we were able to close on a deal I talked about extensively last time, which is the spin-off of our biosecurity unit into a new company called Perimeter. I want to say congratulations to the team at biosecurity at Ginkgo and pulling that off, $60 million and a lot of great new investors coming into that focus really firmly in the area of defense tech and building sort of a biosecurity prime. Ginkgo is a shareholder in that company. We're super excited to see it succeed.

我很期待推進這兩項工作，稍後在我的段落中你們會聽到更多。我們也——在上一季——完成了我上次詳細談到的一筆交易，也就是將我們的生物安全（biosecurity）部門分拆成一家名為 Perimeter 的新公司。我想向 Ginkgo 生物安全團隊致上祝賀，成功完成這件事：籌得 6,000 萬美元，並引入許多優秀的新投資人，將重點牢牢聚焦在國防科技領域，打造某種「生物安全主承包商」。Ginkgo 是該公司的股東。我們非常期待看到它成功。

And I think this is really nice, as I talked about last time, opportunity, both for Ginkgo to keep our focus on the autonomous labs and for the team at Perimeter to grow under their own brand with a new set of defense tech-focused investors. Our focus over the last couple of years was very much on getting these numbers where they are today, bringing down our cash burn in the company. We guided towards this, and Steve will touch on that in his section. But again, happy to have a very strong cash position, $373 million with no bank debt as of Q1 2026. And so you'll hear a little bit more from Steve on this.

我認為這也非常好，正如我上次所說，這同時為 Ginkgo 提供了機會：我們可以持續專注於自主實驗室，而 Perimeter 團隊也能在自己的品牌之下成長，並獲得一批以國防科技為重點的新投資人。過去幾年我們的重點，很大程度上是把這些數字帶到今天的位置，降低公司的現金消耗。我們曾就此提供指引，Steve 也會在他的段落中談到。不過再次強調，截至 2026 年第一季，我們很高興擁有非常強勁的現金部位：3.73 億美元，且沒有銀行負債。稍後 Steve 會再多談一些。

But this sets us up very nicely. We're well capitalized to pursue this area of autonomous labs. We have these base service businesses to build on top of and the lead in developing the technology and you put all that together. And I think we're by far the best bet in this sector. All right.

這讓我們處於非常有利的位置。我們資本充足，能夠追求自主實驗室這個領域；我們也有這些基礎服務型業務可在其上建構，並且在技術開發上領先。把這些因素綜合起來，我認為我們在這個產業中絕對是最值得押注的標的。好。

I'm going to pass it on to Steve to dig into the financials.

接下來我把時間交給 Steve，請他深入說明財務表現。

Thanks, Jason. Before I walk through our financials, I want to take a moment to frame an important change in how we are presenting our results beginning in Q1 2026. As we announced in February, we entered into a definitive agreement to sell our biosecurity business, which was previously reported as a separate segment. Further, as Jason noted, we closed that transaction on April 3. The biosecurity transferred assets met the criteria under US accounting to be classified as held for sale and the financial results reported as discontinued operations as of March 31, 2026. This is the first quarter in which biosecurity is reflected as discontinued operations within our financial statements. And to close with the accounting rules, we have and will retrospectively recast all prior periods presented to conform to this presentation. That means the revenue, operating expenses and cash flows previously attributed to the biosecurity business are removed from each line item of our continuing operations and cash flows as the prior period information is presented, including for Q1 of last year. The former biosecurity results are now reported as a single net line loss from discontinued operations, below loss from continuing operations.

謝謝你，Jason。在我帶各位檢視財務數字之前，我想先花點時間說明一項重要變更：自 2026 年第一季起，我們呈現業績的方式有所調整。如同我們在 2 月宣布的，我們已簽訂具約束力的最終協議，出售我們的生物安全業務；該業務先前以獨立部門（segment）方式揭露。此外，如 Jason 所述，我們已於 4 月 3 日完成該交易。依照美國會計準則，生物安全所移轉的資產符合「待出售」（held for sale）的分類標準，且截至 2026 年 3 月 31 日，其財務結果須以「終止營業」（discontinued operations）方式列報。這是生物安全首次在我們的財務報表中以終止營業反映的季度。並且為符合會計規則，我們已且將回溯重編（retrospectively recast）所有所呈列的過往期間，以符合此一呈現方式。這表示，先前歸屬於生物安全業務的收入、營業費用與現金流量，將自持續營業（continuing operations）與現金流量的各項目中移除，並在呈列前期資訊時一併調整，包括去年第一季。原生物安全的結果現在以「終止營業淨損失」單一淨額列示，位於「持續營業損失」之下。

To be clear, all of the financial commentary I will provide today relates exclusively to continuing operations. We will not be discussing the biosecurity business further in our prepared remarks. On April 7, 2026, for your information, we filed a current report on Form 8-K that includes pro forma financial information for fiscal year's 2023, 2024 and 2025 on a continuing operations basis. Following the biosecurity divestiture, we now operate as a single segment. So with that, I'll now discuss our Q1 results.

為避免疑義，我今天提供的所有財務評論僅與持續營業相關。我們在預先準備的講稿中將不再進一步討論生物安全業務。供各位參考，我們已於 2026 年 4 月 7 日提交 Form 8-K 的即時報告，其中包含以持續營業基礎編製的 2023、2024 與 2025 會計年度擬制（pro forma）財務資訊。生物安全剝離後，我們目前以單一部門運作。接下來我將說明 2026 年第一季的業績。

Revenue was $19 million in the first quarter of 2026, down 49% compared to the first quarter of 2025. As previously disclosed, revenue in the first quarter of 2025 included $7.5 million in noncash revenue relating to the mutual termination of the BiomEdit agreement. Excluding this, revenue in the first quarter of 2026 was down 37% from the prior year period. 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 other nonrecurring items, we believe adjusted EBITDA is a more indicative measure of our profitability.

2026 年第一季營收為 1,900 萬美元，較 2025 年第一季下降 49%。如先前揭露，2025 年第一季的營收包含與 BiomEdit 協議相互終止相關的 750 萬美元非現金營收。排除此項後，2026 年第一季營收較去年同期下降 37%。需要注意的是，我們的淨損包含多項非現金及其他非經常性項目，詳情在我們的財務報表中有更完整的說明。由於這些非現金及其他非經常性項目，我們認為調整後 EBITDA 更能反映我們的獲利能力。

A full reconciliation between adjusted EBITDA and GAAP net loss from continuing operations can be found in the appendix. In the first quarter of 2026, R&D expense decreased 38% from $49 million in the first quarter of 2025 to $30 million in the first quarter of 2026. G&A expense decreased 35% from $20 million in the first quarter of 2025 to $13 million in the first quarter of 2026. These decreases were all driven by our restructuring efforts. Net loss from continuing operations was $76 million in the first quarter of 2026 compared to a loss of $83 million in the prior year period.

調整後 EBITDA 與依 GAAP 計算之持續營業淨損之完整調節表可見於附錄。2026 年第一季，研發費用自 2025 年第一季的 4,900 萬美元下降 38% 至 2026 年第一季的 3,000 萬美元。一般及行政（G&A）費用自 2025 年第一季的 2,000 萬美元下降 35% 至 2026 年第一季的 1,300 萬美元。這些下降皆由我們的重整（restructuring）措施所驅動。2026 年第一季持續營業淨損為 7,600 萬美元，去年同期為淨損 8,300 萬美元。

The reduction in loss year-over-year was due to our restructuring efforts. Moving further down the page, you'll note that adjusted EBITDA in the first quarter of 2026 was negative $42 million, which was down from negative $44 million in the first quarter of 2025. Since we are now only operating in a single segment, we only present a single measure of adjusted EBITDA, and it is important to note that adjusted EBITDA includes the carrying cost of excess lease space, which you can see was $16 million in the first quarter of 2026. Previously, this cost would not have been included in the former presentation of segment adjusted EBITDA. This cost represents the base rent and other charges relating to lease space which we are not occupying net of sublease income.

淨損年對年下降是由於我們的重整措施。再往下看，你會注意到 2026 年第一季的調整後 EBITDA 為負 4,200 萬美元，較 2025 年第一季的負 4,400 萬美元有所改善。由於我們目前僅以單一部門營運，我們只呈列單一的調整後 EBITDA 指標；並且需要注意的是，調整後 EBITDA 包含多餘租賃空間的持有成本（carrying cost），你可以看到 2026 年第一季為 1,600 萬美元。先前在舊的部門調整後 EBITDA 呈現方式中，這項成本不會被納入。此成本代表我們未實際使用之租賃空間的基本租金與其他費用，扣除轉租收入（sublease income）後的淨額。

This is a cash operating cost that is not related to driving revenue right now and can be potentially mitigated through subleasing. And finally, cash burn in the first quarter of 2026 was $48 million, down from $58 million in the first quarter of 2025, a 17% decrease. As previously reported, in October 2025, we amended and reset the annual commitments with Google Cloud for $14 million. Resetting the commitment reduced our future minimum commitments by more than $100 million compared with the original terms and extended the commitment term from three to six years. We paid this $14 million in Q1 of 2026, which is reflected in our cash burn for the quarter.

這是一項目前與推動營收無關的現金營運成本，且可望透過轉租來部分緩解。最後，2026 年第一季的現金消耗為 4,800 萬美元，較 2025 年第一季的 5,800 萬美元下降，降幅為 17%。如先前所報告，我們於 2025 年 10 月修訂並重設與 Google Cloud 的年度承諾金額為 1,400 萬美元。重設承諾使我們未來的最低承諾金額較原始條款減少超過 1 億美元，並將承諾期限由三年延長至六年。我們已於 2026 年第一季支付該 1,400 萬美元，並已反映在本季的現金消耗中。

Excluding the payment to Google Cloud, cash burn reflects a significant decrease in the first quarter of 2026 compared to the first quarter of 2025, which was a direct result of the restructuring. Now turning to guidance. As we discussed in February, 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. We have turned the page on our pure focus on restructuring actions to focus this year not only on cost efficiency, but on investing in what we see as our opportunities while continuing to provide our customers the advanced services that they have come to extend. For these reasons, we believe cash burn best reflects our continuing services and tools and further investments in autonomous labs.

若排除支付給 Google Cloud 的款項，2026 年第一季的現金消耗相較 2025 年第一季顯著下降，這是重組的直接結果。接著談指引。如同我們在 2 月所討論，2026 年的重點是在持續維持成本效率的同時，投資我們的 AI 機器人與軟體，將自主實驗室帶給我們的生物科學客戶，包括在波士頓建置我們的 Frontier Autonomous Lab。我們已翻過僅專注於重組行動的一頁，今年的重點不僅是成本效率，也是在持續提供客戶已習慣並持續採用的先進服務之同時，投資我們所看到的機會。基於這些原因，我們認為現金消耗最能反映我們持續提供的服務與工具，以及對自主實驗室的進一步投資。

In terms of 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 our continued improvements in cash burn efficiency and our business pursuits for 2026. And with that, I'll hand it back over to you, Jason.

就全年展望而言，我們重申 2026 年整體現金消耗指引，總計為 1.25 億至 1.50 億美元。此區間反映了成本效率、持續服務與工具，以及我們進一步投資之間的穩健平衡。總結來說，我們對於現金消耗效率的持續改善以及 2026 年的業務推進感到滿意。接下來我把時間交回給你，Jason。

Thanks, Steve. So I'm going to dive in on the strategic section. I'm excited to go into this today. Our mission is to make biology easier to engineer. And the way we're really aiming to solve that problem, we believe the bottleneck fundamentally is the laboratory work associated with bioengineering.

謝謝，Steve。接下來我將深入談策略部分。我很期待今天能談到這些。我們的使命是讓生物學更容易被工程化。而我們真正想解決這個問題的方式是：我們相信根本的瓶頸在於與生物工程相關的實驗室工作。

And so I'm going to dig deep today and talk about why autonomous labs will be replacing the lab bench. I want to highlight some of what we're doing with Nebula, our system because we have some news this month in terms of expanding that system. And then finally, the services that we put on top of Nebula, our cloud lab, data points and solutions. These are sort of like, we call it, like our Starlink, right? If you think about SpaceX, 70% of the launches last year were actually Starlink, their own internal product, in the coming year, the ability for us to scale up on autonomous lab and showcase that you can make money on services without having people in the middle of the lab doing those laboratory services, I think, is a real highlight and will help drive sales of our systems into the world.

因此我今天會深入說明為什麼自主實驗室將取代實驗台。我也想強調我們在 Nebula（我們的系統）上所做的一些事情，因為本月我們在擴展該系統方面有一些新消息。最後，還有我們疊加在 Nebula 之上的服務：我們的雲端實驗室、資料點與解決方案。這些有點像——我們稱之為——我們的 Starlink，對吧？如果你想到 SpaceX，去年 70% 的發射其實是 Starlink，是他們自己的內部產品。展望來年，我們擴大自主實驗室規模並展示你可以在不需要人在實驗室中間提供那些實驗室服務的情況下，透過服務賺錢的能力，我認為是一大亮點，也將有助於推動我們的系統在全球的銷售。

So I'm going to talk about all three, and let's dive in. Okay. So I gave this analogy. I'm going to do it again because I think this is for new folks listening on the call, it's worth understanding what we say when we say autonomous lab as distinct from traditional lab automation. So I'm going to give an analogy from the transportation industry.

所以我會談這三個部分，讓我們開始深入。好。我之前用過這個類比，我會再用一次，因為對於這通電話中新加入的聽眾，理解我們所說的「自主實驗室」與傳統實驗室自動化的差異很重要。因此我會用交通運輸產業的類比來說明。

On the y-axis here, we have the amount of automation for a certain type of transport. And on the x-axis, the request flexibility. In other words, the users asking the transportation system to do something different or not. And so for a low request flexibility and a high level of automation, that's your subway, right? It's the red line here in Boston, you sit down in the subway, and it takes you away.

在這裡的 y 軸，是某一種運輸方式的自動化程度；x 軸則是需求彈性，也就是使用者是否會要求運輸系統做不同事情的程度。因此，在低需求彈性且高自動化程度的情況下，那就是地鐵，對吧？就像波士頓的紅線，你坐上地鐵，它就把你載走。

You don't have to do anything, it is high level of automation, totally automated transport, but it is very inflexible. You have to want to go to one of the stops on the red line. Low amount of automation, high amount of flexibility. That's a car, right? You get your hands on the wheel, foot on the pedals and it'll take a ride, take a ride to your house or to the grocery store anywhere you want to go.

你不需要做任何事，它是高度自動化、完全自動化的運輸，但非常缺乏彈性。你必須想去紅線上的某一站。低自動化、高彈性的是汽車，對吧？你手握方向盤、腳踩踏板，它就能載你回家或去雜貨店——你想去哪裡都可以。

And that's roughly what the transportation system has looked like for the last 100 years. Let's go to the next slide. You've been to California in the last four or five years or now L.A. or Austin or soon in Boston and you sat in the back seat of a Waymo, and it is amazing, it is like sitting on a subway, you don't have to do anything, but it will take you right to your house. It has the flexibility of a car.

而這大致就是過去 100 年交通系統的樣貌。讓我們看下一張投影片。如果你在過去四、五年去過加州，或現在在洛杉磯或奧斯汀，或很快在波士頓，你坐在 Waymo 的後座，那真的很驚人，就像坐地鐵一樣，你什麼都不用做，但它會直接把你送到家門口。它具備汽車的彈性。

And so it's those two things together that sort of flexibility plus a high level of automation that mean we actually give it a new name, we don't call it an automated car, we call it an autonomous car. Because up until now, you've needed a human being with our brain in the loop in order to manage that amount of flexibility into the system. All right. So if you look at the next slide, you'll see the miles traveled by cars and trucks versus subways and trains in the United States, it's more than 99% in cars and trucks. And that's not because we don't know about subways.

因此，正是「彈性」加上「高自動化程度」這兩者結合，讓我們給它一個新名字；我們不稱它為自動化汽車，而稱之為自駕（autonomous）汽車。因為在此之前，你需要一個人類——用我們的大腦——在迴路中，才能把那麼高的彈性納入系統管理之中。好。如果你看下一張投影片，你會看到在美國，汽車與卡車的行駛里程相較於地鐵與火車，超過 99% 都在汽車與卡車上。這並不是因為我們不知道地鐵。

It's that we need that flexibility to do our day-to-day lives, that it's required for the transportation that humans need. All right. So now let's go into the lab bench and into the lab. So low amount of flexibility, high amount of automation. We have actually automation in the lab.

而是因為我們需要那種彈性來應付日常生活——那是人類所需的交通運輸所必須具備的。好。現在我們把視角轉到實驗台與實驗室。低彈性、高自動化——我們在實驗室裡其實已經有自動化。

It's called a work cell. That's a 3D schematic of a work cell we have here at Ginkgo. Companies like HighRes, Thermo and Biosero make these. And they're like a subway, right? They're great.

那叫做工作單元（work cell）。那是我們在 Ginkgo 這裡的一個工作單元 3D 示意圖。像 HighRes、Thermo 和 Biosero 這些公司都會製作。它們就像地鐵一樣，對吧？它們很棒。

They're fully automated. They'll do an experiment without a scientist in the loop, but it better be the experiment that you ordered yesterday. It's not going to do a new experiment for you today. Low amount of automation, high amount of flexibility. We have those two in the lab.

它們是全自動的，不需要科學家在迴路中就能完成一個實驗，但前提是那必須是你昨天就下單的那個實驗。它不會今天替你做一個全新的實驗。低自動化、高彈性——在實驗室裡我們也有這種。

It's called the lab bench. And you, as a scientist are basically the human glue, connecting all of these different devices in the lab together to do whatever protocol you want to do. And so again, here's the kicker. If you look at research budgets between laboratory work cells, which are used in things in pharma companies like high-throughput screening or combinatorial chemistry or things like that versus at the lab bench it's about 95% plus at the bench. And again, it's not like we don't know about work cells.

那就是實驗台（lab bench）。而你，作為科學家，基本上就是把實驗室裡各種不同設備黏合在一起的「人類黏著劑」，以便執行你想做的任何流程（protocol）。所以重點又來了：如果你看研究預算在實驗室工作單元（常用於製藥公司中的高通量篩選、組合化學等）與實驗台之間的分配，約有 95% 以上都花在實驗台上。同樣地，這並不是因為我們不知道工作單元。

It's that scientists need flexibility to explore all the different hypotheses they have for discovering a new drug or developing a new crop trait or whatever type of biotechnology they're doing. So this is what we're trying to build at Ginkgo. We're trying to get up to that top right corner and make a Waymo. We're trying to make an autonomous lab that has the flexibility of the bench so scientists can order whatever experiment they want, but the automation of a work cell. In other words, they don't have to be there to do each and every step and move the samples among the different equipment and program the equipment.

而是因為科學家需要彈性，去探索他們為了發現新藥、開發新的作物性狀，或不論他們在做哪一類生物科技時所提出的各種假設。這就是我們在 Ginkgo 想要打造的。我們想要走到右上角，做出一個 Waymo。我們想打造一個自主實驗室：具備實驗台的彈性，讓科學家可以下單任何他們想做的實驗；同時具備工作單元的自動化。換句話說，他們不必親自到場去完成每一步、在不同設備之間移動樣本，或為設備編程。

They can just hit go and have that protocol run end-to-end for them via the automation, but whatever they want to order that day. That's the target. And look, if you go to the next slide, the value prop here, I think, is very clear for getting rid of the lab bench, massive overhead cost savings. You've heard a lot about overhead costs at academic research labs and things like that. That's really paying for ultimately, millions of square feet of laboratory space, it's 50 million square feet of laboratory space just in the Boston area.

他們只要按下開始，就能透過自動化讓該流程端到端為他們執行，不論他們當天想要下單的是什麼。那就是目標。而且你看，如果你到下一張投影片，這裡的價值主張我認為非常清楚：把實驗台移除，帶來巨大的間接成本節省。你們已經聽過很多關於學術研究實驗室等的間接成本。那其實最終是在為數以百萬平方英尺的實驗室空間買單；光是在波士頓地區就有 5,000 萬平方英尺的實驗室空間。

So you can dramatically reduce that. You can increase the research productivity of your human scientists as we need more data from AI, and we'll talk about that in a minute. And then we can enable AI scientists to run these lab-in-the-loop experiments. We just announced a project with OpenAI a few months ago, where GPT5 ran our lab. That's that kind of lab in the loop.

所以你可以大幅降低那個成本。隨著我們需要更多來自 AI 的資料（我們等一下會談到），你也可以提升人類科學家的研究生產力。然後我們也能讓 AI 科學家去執行這些「實驗室在迴路中」（lab-in-the-loop）的實驗。我們幾個月前剛宣布與 OpenAI 的一個專案，由 GPT5 來運行我們的實驗室。那就是這種 lab in the loop。

Experiments that we're seeing increasingly also in the biopharma industry. But to put a point on it, like a typical large pharma, biopharma biotech spends $1 billion to $3 billion a year on research, not clinical trials, but spending on 1 million-plus square feet of lab benches. And if you look at their spending within that on automation, it's well below $100 million, usually much less. And frankly, I think those numbers should flip. I think really, the majority of the capital should be going towards automated laboratory work rather than manual benches.

我們也越來越常在生物製藥產業看到這類實驗。但把話說得更明確一點：一家典型的大型藥廠／生物製藥／生技公司，每年在研究上的支出是 10 億到 30 億美元（不是臨床試驗），而是花在超過 100 萬平方英尺的實驗台空間上。而如果你看他們在其中用於自動化的支出，遠低於 1 億美元，通常還更少。坦白說，我認為這些數字應該要對調：資本支出的大部分應該投入自動化實驗室作業，而不是人工實驗台。

And the reason for that is, I think, a relatively straightforward calculation. On the next slide, you can see comparison between a traditional manual lab and an autonomous lab. It's about a threefold space improvement. When you take all these -- this equipment that's often very spread out in a normal human-operated lab because people need to get around it and safety reasons and all these things. But in an autonomous lab, you can jam all that equipment right next to each other about as tight as you can make it for the arms to work and things like that.

原因在於，我認為這是一個相對直接的計算。在下一張投影片，你可以看到傳統人工實驗室與自主實驗室的比較。空間效率大約提升三倍。因為在一般由人操作的實驗室裡，這些設備往往分散得很開，因為人需要在其中走動、出於安全等各種原因。但在自主實驗室裡，你可以把所有設備盡可能緊密地擺在一起，只要機械手臂能運作、諸如此類即可。

So it's about a threefold space reduction. And then -- the manual labs really are just run 40 hours a week, right? I mean, it's when people come in the lab and humans are there, and they got to be there, and that's when you can get people to work. There's almost never multiple shifts in these types of sort of high-end research labs. It's really a 40-hour work week.

所以空間大約可以縮減三倍。然後——人工實驗室基本上只運作每週 40 小時，對吧？也就是人進到實驗室、人在那裡、而且必須人在那裡的時段，才能讓人工作。在這類偏高端的研究實驗室裡，幾乎從來沒有多班制。基本上就是每週 40 小時的工作週。

And our system here in Nebula is running 168 hours a week. So 24/7, and that's a fourfold improvement in sort of hours available for utilization of your laboratory. So I think a real clear, threefold, fourfold on two different axes. It is a clear value driver if you go to the next slide. I think there's little question the ROI is there.

而我們在 Nebula 的系統是每週運作 168 小時。也就是 24/7，這在可用工時／實驗室利用率上帶來四倍的提升。所以我認為在兩個不同的軸向上，都很清楚是三倍、四倍的改善。如果你到下一張投影片，這是一個非常明確的價值驅動因素。我認為幾乎沒有疑問，投資報酬率（ROI）是存在的。

I think the big question with autonomous labs is a technical one. It's how do you get that high level of automation and the high level of flexibility, that top right corner without a human being in the loop, right? It's the same question of the Waymo, how do you get it to navigate all these different environments and different roads without a human in the loop, if you can do it, it's an obvious win. If we can do this in the lab, it's an obvious win. All right.

我認為自主實驗室最大的問題是技術問題：如何在沒有人的迴路（human in the loop）下，同時達到高自動化與高彈性，也就是右上角那個象限？這跟 Waymo 的問題一樣：如何在沒有人的迴路下，讓它在各種不同環境與道路中導航；如果你能做到，顯然就是勝利。如果我們能在實驗室做到，顯然也是勝利。好。

So let me talk through a little bit the design constraints that we focused on at Ginkgo. On the next slide, you can see a work cell, one of those subways, the way it's designed is it's designed against a particular protocol. So if you're a biopharma company and you want to build a high-throughput screening work cell and you call a traditional automation vendor, they're going to -- first question they're going to ask you is, tell me about your protocol and tell me what throughput you need to run out. How many samples do you want to get through every week? Because they're building you a subway line.

所以我來談談我們在 Ginkgo 著重的一些設計約束。在下一張投影片，你可以看到一個工作單元（work cell），也就是其中一條「地鐵線」。它的設計是針對某一個特定流程（protocol）。所以如果你是一家生物製藥公司，想建一個高通量篩選（high-throughput screening）的工作單元，然後你打給傳統自動化供應商，他們——第一個問題會問你：跟我說你的流程是什麼，以及你需要跑到什麼通量。你每週想要處理多少樣本？因為他們是在幫你建一條地鐵線。

It's going to be built to do that protocol for you. But if you're a new facility head who is opening a lab in Central Square here in Cambridge, and you are building it for a scientific lead. You don't ask that scientific lead, hey, what's the protocol you're going to run in this lab that 30 of your scientists are going to use to do work? You say, what kind of science you're doing? And very specifically, what equipment do you want me to install in that lab so that your scientists can be productive over the next three to five years as they use the lab? And so it is oriented around the equipment rather than the protocol. And that's a sort of a subtle point, but it has a huge amount of consequences when it comes to how you design the hardware and software that responds to this challenge. And so if you go to the next slide, you can see our hardware solution here is what we call our rack carts, our reconfigurable automation carts.

它會被打造來替你執行那個流程。但如果你是一位新的設施主管，正在劍橋這裡的 Central Square 開一間實驗室，而你是為了一位科學負責人來建置它，你不會去問那位科學負責人：嘿，你要在這個實驗室跑什麼流程，讓你 30 位科學家用它來做事？你會問：你在做什麼類型的科學？更具體地說，你希望我在這個實驗室安裝哪些設備，讓你的科學家在未來三到五年使用這個實驗室時能保持高生產力？所以它是以設備為導向，而不是以流程為導向。這看似是個細微差異，但在你如何設計硬體與軟體來回應這個挑戰時，會帶來非常巨大的影響。因此如果你到下一張投影片，你可以看到我們的硬體解決方案，我們稱之為 rack carts，也就是可重新配置的自動化推車。

And we have -- it's basically a robot wrapped around each laboratory device. So we have control over that environment. There's a HEPA filter on the top, which is important for a lot of biological work. We have opportunities for contamination and things like that. We have a six-axis robotic arm and a piece of magnetic motion track, allows you to, if you go to the next slide, LEGO block these together into ultimately very large setups.

而我們有——基本上是每一台實驗室設備外面都包了一個機器人。這樣我們就能控制那個環境。頂部有 HEPA 濾網，這對很多生物工作很重要，因為會有污染等風險。我們有一個六軸機械手臂，以及一段磁性運動軌道，讓你——如果你到下一張投影片——可以像樂高積木一樣把它們拼在一起，最終形成非常大型的配置。

And we're at 50-plus right now in the lab and it's growing quickly. I'll show you some photos in a second. We have 103 racks will be coming online just in about a week, all in one big setup here in Boston. And so, if you go to the next slide, we can -- I want to show -- this is actually a video of the OpenAI protocol. So we had this project with OpenAI, where GPT5 controlled the lab, and we made a video of one of the samples just moving through the various racks for that protocol.

我們現在在實驗室裡已經有 50 多個，而且成長很快。我等一下會給你看一些照片。我們大約一週內就會再有 103 個 rack 上線，全部在波士頓這裡的一個大型整體配置中。所以，如果你到下一張投影片，我們可以——我想展示——這其實是 OpenAI 流程的影片。所以我們和 OpenAI 有這個專案，由 GPT5 控制實驗室，我們做了一段影片，展示其中一個樣本如何在該流程中穿梭於各個 rack 之間。

And so this just gives you a sense of like how does it work, right? So you have these tracks and we're able to move like our one sort of constraint on the system is that we pass things in what's called SBS format. So that little rectangle you saw there is like a 3 x 5-inch square -- rectangle and it can contain 96 or this is a 384-well plate or 1586 well plate. It can also carry consumables like tips and other things. And the arm picks up that piece of plastic wear or samples or tips or whatever it might be, and then puts it on to a particular device.

這能讓你感受一下它是怎麼運作的，對吧？你有這些軌道，我們能移動——我們系統的一個限制是：我們以所謂的 SBS 格式來傳遞物件。你剛看到的那個小長方形，大概是 3 x 5 英吋的長方形，它可以裝 96 孔，或這是一個 384 孔板或 1586 孔板。它也可以承載耗材，例如吸頭（tips）和其他東西。機械手臂會拿起那個塑膠器皿或樣本或吸頭或任何物件，然後把它放到某一台特定設備上。

So in this case, it's going -- just went through an acoustic liquid handler. Now it's going on to a bravo, liquid handler. You saw, again, that first thing that got put down there was actually the plastic tips that are now getting picked up by the liquid handler. And the other two plates or sort of sample and destination plate. So we're -- in this case, for the OpenAI project, we're picking up some synthetic DNA, and we're putting it into reaction mixes that were designed by the GPT5 model at the time, right?

所以在這個案例中，它——剛通過一台聲學式移液器（acoustic liquid handler）。現在要到一台 bravo 移液器（liquid handler）。你又看到，剛才放下去的第一個東西其實是塑膠吸頭，現在正被移液器拿起來。另兩個板則是樣本板與目的板。所以我們——在這個 OpenAI 專案中——會取用一些合成 DNA，然後把它加入到當時由 GPT5 模型設計的反應混合物（reaction mixes）中，對吧？

And so again, a key feature here is any device that accepts those SBS plates, we can integrate into the racks. It takes us usually like one month to 1.5 months if it's a new device. We've now got 80-plus devices on there. We're adding new devices all the time. If a customer asks us for one, if we want to add one to Nebula, we just bring it online.

所以再說一次，這裡的一個關鍵特性是：任何能接受這些 SBS 板的設備，我們都能整合進這些 rack。若是新設備，通常需要大約 1 個月到 1.5 個月。我們現在已經有 80 多種設備在上面。我們一直在新增設備。如果客戶要求某一種，或我們想把某一種加入 Nebula，我們就把它上線。

So now that plate is going to get shaken up and put on to ultimately thermocycler or the final analytical device to run the qPCR reactions and give a readout on the performance of each one of those samples back to, in this case, an AI model. That data at most of the runs on Nebulas going back to a human scientist, as opposed to an AI scientist. But we expect there will be a mix of both as science goes forward, we'll have both scientists and their agents ordering experiments on autonomous labs. Go to the next slide. Great thing about this system is it can expand.

所以現在那個孔板會被震盪混勻，然後放到最終的熱循環儀（thermocycler）或最終的分析設備上，去跑 qPCR 反應，並把每一個樣本的表現讀值回傳——在這個案例中，是回傳給一個 AI 模型。在 Nebula 的大多數運行中，這些資料會回到人類科學家手上，而不是 AI 科學家。但我們預期，隨著科學向前發展，兩者會混合並存：會同時有科學家與他們的代理（agents）在自動化實驗室上訂購實驗。請到下一張投影片。這個系統很棒的一點是它可以擴展。

We started Nebula, I think was about eight racks, doing NGS like next-generation sequencing prep for our samples here at Ginkgo and expanded ultimately now up to over 100 racks on the system. So let's dig in a little bit. I want to go to the next section. So that was sort of the theory, like how we design the hardware in order to solve some of the challenges of the autonomous lab and what autonomy means. And now I want to dig in a little bit on Nebula specifically, because I think what's really unique about Ginkgo is we're not just a hardware company.

我們一開始做 Nebula，我記得大概是八個機架（racks），在 Ginkgo 這裡為我們的樣本做 NGS（次世代定序）前處理，最後一路擴展到現在系統上超過 100 個機架。所以我們深入一點。我想進到下一個段落。剛才那算是理論：我們如何設計硬體來解決自動化實驗室的一些挑戰，以及「自主」代表什麼。現在我想更深入談談 Nebula 本身，因為我覺得 Ginkgo 真正獨特之處在於，我們不只是硬體公司。

We actually run BSL-2 labs here in Boston and do scientific partnerships with some of the largest biotech, ag biotech, industrial biotech companies in the world. And so we can actually show what it looks like to do real science on a system like this. And so if you go to the next slide, one of the things I'm quite proud of that we've been able to show in the last quarter is over 100 protocols with more than 30 of them being unique, submitted by scientists, and I'll mention this, but these are not being submitted by automation engineers or experts in robotics. Being submitted onto our system, Nebula here in Boston, which has 50-plus lab devices all integrated together where you can send point-to-point samples from any device to any other device as requested by those scientists. There is nothing else like Nebula in the world today, doing sort of open-ended science like this at this scale with this number of unique protocols and end users.

我們其實在波士頓這裡運營 BSL-2 實驗室，並且與全球一些最大的生技、農業生技、工業生技公司進行科學合作。因此我們可以真正展示，在這樣的系統上做真實科學研究會是什麼樣子。所以如果你到下一張投影片，我很自豪的一件事是：在上一季我們已經展示了超過 100 個 protocol，其中有 30 多個是獨特的（unique），而且是由科學家提交的——我特別提一下，這些不是由自動化工程師或機器人專家提交的。這些 protocol 被提交到我們在波士頓的 Nebula 系統上；該系統整合了 50+ 台實驗室設備，你可以依照科學家的需求，把樣本從任何一台設備點對點送到任何另一台設備。今天世界上沒有任何東西像 Nebula 這樣，以這種規模、這麼多獨特 protocol 與終端使用者，做這種開放式（open-ended）的科學。

And it's proof that autonomous labs are feasible. I mean there's work to do and we talked about that. Things break as we are scaling this up. That's for sure but it is evidence, in my view, that this is going to land, like we are going to beat the manually operated lab. And so if you go to the next slide, I want to walk through a few of the key things that you got to show if you're going to take out one of those laboratory floors at Takeda or Merck or Novartis or whatever or Bayer Crop Science or any of these companies that do a lot of laboratory work.

而這也證明自主實驗室是可行的。我的意思是，還有很多工作要做，我們也談過。當我們把規模拉大時，東西確實會壞——這是肯定的——但在我看來，這是證據顯示這件事會落地：我們會超越手動操作的實驗室。所以如果你到下一張投影片，我想帶大家走過幾個關鍵點：如果你要取代武田（Takeda）或默克（Merck）或諾華（Novartis）或拜耳作物科學（Bayer Crop Science）等公司其中一整層的實驗室樓層（或任何做大量實驗室工作的公司），你必須展示哪些能力。

So first, you want to connect 100-plus devices in a single automation setup, all right? So it can't just be five or 10. A scientist expects to have access to many different devices in order to do whatever protocol they might read about in a scientific paper this week. And then I think about 100 is the right number. So we've been able to -- this week, we'll find out, we're turning it on in about five days.

第一，你要能在單一自動化配置中連接 100+ 台設備，對吧？不能只是五台或十台。科學家會期待能使用很多不同設備，才能做出他們本週在科學論文裡讀到的任何 protocol。我認為大約 100 是合適的數字。所以我們已經能做到——這週我們會知道結果，我們大概五天後就會把它開啟。

105 all -- or 103 racks, all in one big setup. And the reason we can do that is because of that rack productized hardware, that cart I showed you. We just rolled in, they came in off a truck and we rolled another 50 in and those have all gone in the actual install over the last three or four weeks. So it's pretty fast to put that many new devices on an automated setup. We'll see if it works.

105 台全部——或 103 個機架——全部在同一個大型配置裡。我們之所以能做到，是因為那個機架式、產品化的硬體，也就是我給你們看的那台推車。我們只是把它們推進來；它們從卡車卸下後，我們又推進來另外 50 個，而這些在過去三到四週內都已完成實際安裝。所以要在一個自動化配置上新增這麼多新設備，速度非常快。我們看看它能不能運作。

Second, we have run 10 now like a 30-plus unique protocols, 100-plus different -- 100-plus total protocols. But that's kind of -- you got to get in that, I think, 50 to 100 to maybe 200 unique protocols, all running on the autonomous lab at the same time. And we do that with our catalyst. This is our software, our scheduler that we built is a very complicated scheduling problem. It's really easy to mess this up.

第二，我們現在已經跑了 10——呃，像是 30+ 個獨特 protocol、100+ 個不同——總計 100+ 個 protocol。但那只是——我覺得你必須達到大概 50 到 100、甚至 200 個獨特 protocol，同時在自主實驗室上運行。我們是用 Catalyst 來做到這點。這是我們的軟體、我們打造的排程器（scheduler）；這是一個非常複雜的排程問題，而且很容易把它搞砸。

Biology is very sensitive to timing. Things break all the time as we keep driving the scale up here. So we're getting to do that quick cycle of debugging and improving the system, but that scheduler is really the key piece of software driving that. And then finally, scientists, scientists, not automation engineers, and I think on a peak day on Nebula, we had 439 or so scientists submitting. So that's really exciting like to have that many different scientists submitting protocols on one automation system.

生物實驗對時間非常敏感。當我們持續把規模拉大時，東西一直會壞。所以我們得以進行快速的除錯與系統改進循環，但那個排程器真的是驅動一切的關鍵軟體。最後，使用者是科學家、科學家——不是自動化工程師。我想在 Nebula 的尖峰日，我們大概有 439 位左右的科學家在提交。能有這麼多不同科學家在同一套自動化系統上提交 protocol，真的很令人興奮。

Again, I don't know of any automation system in the world that's been able to do that before. And we're able to do that in part by leveraging AI coding tools with custom harnesses wrapped around them that basically understand how to transfer the scientist's intent in human language into code to operate the autonomous lab. And that is a big unlock. We're very thankful for what's going on with all the coding agents. That's a real help for improving the ease of use because at the end of the day, to make robots do something, you have to program them.

同樣地，我不知道世界上有任何自動化系統以前能做到這件事。我們之所以能做到，部分是因為我們利用 AI 程式碼工具，並在外面包了一層客製化的 harness，讓它基本上能理解如何把科學家用人類語言表達的意圖，轉換成操作自主實驗室的程式碼。這是一個很大的突破。我們非常感謝各種 coding agents 的進展。這對提升易用性幫助很大，因為說到底，要讓機器人做事，你就必須把它們程式化。

And to walk up to a lab bench and do your work by hand, you don't. So we have to solve this problem. We can't make it so that scientists have to become coders to do their job, and we've really just been giving a gift by these AI coding tools, again, like the Codex and the cloud codes and things like that, that can sit inside other tools that are specific to the automation to get this done. So those are the three big ones, and I'm pretty happy with the progress at all. So if you go to the next slide, as I mentioned several times now, we're going from 50 to 105 racks by the end of this month.

而你走到實驗台前用手做事時，你不需要寫程式。所以我們必須解決這個問題。我們不能讓科學家為了做本職工作就得變成程式設計師；而這些 AI 程式碼工具真的像是送給我們的一份禮物——再一次，像 Codex、cloud codes 之類的東西——它們可以嵌入到其他針對自動化的工具裡，來把這件事完成。所以這就是三個大的重點，我對整體進展非常滿意。那如果你到下一張投影片，如我已經提過好幾次，我們會在這個月底前把機架數從 50 增加到 105。

It's going to be awesome. It's a really cool system to see people should come out and visit it. If you go to the next slide, that scheduler is not trivial. So this is an example of our scheduler, I think, running 17 or 20 different protocols at the same time. Each color is a different protocol.

這會非常棒。這是一個很酷的系統，大家應該來現場看看、參觀一下。如果你到下一張投影片，那個排程器並不簡單。這是一個我們排程器的例子，我想它同時在跑 17 或 20 個不同的 protocol。每一種顏色代表一個不同的 protocol。

Each row is a different device on the system. The x-axis is time. And so you can imagine if you want to add a new protocol to that, and you're like, okay, I need to use the device on row three, the device on row seven and the device on row nine and I need the device on row three for the first three minutes, then I'm not willing to tolerate up to one hour gap then the second device for 15 minutes, then up to a 30-minute gap than the last device. It will check, can it fit you in. And if it can fit you in, or if it can fit you in by moving a couple of other things that doesn't disrupt them in a way that breaks the protocol, it will fit you in.

每一列（row）是一台系統上的不同設備。x 軸是時間。所以你可以想像，如果你想加一個新的 protocol，你會說：好，我需要用第 3 列的設備、第 7 列的設備和第 9 列的設備；我需要第 3 列的設備用前三分鐘，然後我最多只能容忍到一小時的空檔，接著第二台設備用 15 分鐘，然後最多只能有 30 分鐘空檔，再用最後一台設備。它會檢查：能不能把你排進去。如果能排進去，或者如果能透過移動其他幾件事、且不會以破壞 protocol 的方式干擾它們而把你排進去，它就會把你排進去。

That's awesome, right? That's very much not how the traditional lab automation, the subways work. They're running a batch, that subway line is showing up at a certain time. You can't just jump and insert yourself in the middle, but you can with our scheduling software here. On the next slide, the little green one, it's hard to see, but that column, third column over from the left over there has the names of all the different scientists submitting.

這很棒，對吧？這完全不是傳統實驗室自動化——那種「地鐵」式工作流程——的運作方式。它們是跑一個批次，那條地鐵線會在固定時間到站；你不能在中間跳進去插隊，但用我們這套排程軟體就可以。到下一張投影片，那個小小的綠色區塊不太好看，但左邊數過來第三欄那一欄，列出了所有不同提交者（科學家）的名字。

So I'd really love this. I love that we're seeing different people submitting different orders for protocols every day. It's really exciting. And again, I think it's unique. We're also seeing a lot of energy on the US government side. If you have the next slide, a lot of new policy action here, there's the genesis mission, which we're fortunate to be a part of from the White House to bring AI into the national labs. But there's a big motion right now where we're seeing an increasing amount of drug discovery work moving to China from Kendall Square, I was talking about earlier here in the Boston area. And that's because simply Chinese scientists are paid one-third as much and they're doing equal work to what's happening here in the US., like they're just as good. They're just as smart.

所以我真的很喜歡這個。我喜歡看到每天都有不同的人為各種協定提交不同的訂單。這真的很令人興奮。而且再說一次，我認為這很獨特。我們也看到美國政府端有很多動能。如果你看下一張投影片，這裡有很多新的政策行動；有一個 genesis mission，我們很幸運能參與其中，由白宮推動把 AI 帶進國家實驗室。但現在有一個很大的趨勢：我們看到越來越多的藥物發現工作，正從我先前提到的波士頓地區 Kendall Square 轉移到中國。原因很簡單：中國科學家的薪酬只有三分之一，但他們做的工作與美國這裡發生的一樣——他們同樣優秀、同樣聰明。

And so I think if we want to remain competitive, we got to think about doing our research in a fundamentally different way in the United States. I don't think we can just rest on our laurels of having the only smart scientists in the world in this area or at least versus China. And I think that era is over -- firmly over at this point. And so we've got to think about a new way to do it. I'm pretty heartened to see activity out of the National Science Foundation is funding $100 million for a network of cloud laboratories and autonomous labs.

因此我認為，如果我們想保持競爭力，就必須思考在美國以一種根本不同的方式來做研究。我不認為我們可以只憑著在這個領域擁有全世界唯一的聰明科學家、或至少相較於中國仍然如此，就高枕無憂。我認為那個時代已經結束——在此刻已經徹底結束。所以我們必須思考一種新的做法。看到美國國家科學基金會（NSF）的動作，我相當振奮：他們正投入 1 億美元，資助雲端實驗室與自主實驗室的網路。

There's a new bill introduced by Senator Young to sort of do more of this cloud labs and autonomous labs. So hopefully, we see more here, but I'm encouraged by what we see already. If you go to the next slide. We're obviously very fortunate. I had a chance in December to sort of ribbon cut the first our RAC robots going into Pacific Northwest National Labs and signed a new contract for $47 million, much larger autonomous lab set up nearly 100 RACs going in a new building in a couple of years at PNNL.

另外，Young 參議員也提出了一項新法案，想要推動更多這類雲端實驗室與自主實驗室。因此希望我們能看到更多進展，但我對目前已經看到的狀況感到鼓舞。如果你看下一張投影片。我們顯然非常幸運。我在 12 月有機會為第一批進入太平洋西北國家實驗室（PNNL）的 RAC 機器人剪綵，並簽署了一份 4,700 萬美元的新合約，規模更大的自主實驗室建置，將近 100 台 RAC，幾年後會在 PNNL 的一棟新大樓裡部署。

So this is really exciting, and I think sort of highlights the direction I believe our national labs will go, our scientific research in the country. If you go to the next slide, we were lucky to give ARPA-H, a tour of Nebula. We have a great project with them. And the work is accelerated by having these autonomous labs available to our scientist at Ginkgo. I think this is something that makes a lot of sense for a lot of labs at the National Institute of Health, for example, or NSF-funded labs or academic research universities.

這真的很令人興奮，我認為這也凸顯了我相信我們國家實驗室、以及全國科學研究將前往的方向。如果你看下一張投影片，我們很幸運帶 ARPA-H 參觀了 Nebula。我們與他們有一個很棒的專案。而有了這些自主實驗室可供 Ginkgo 的科學家使用，工作就能被加速。我認為這對例如美國國家衛生研究院（NIH）的許多實驗室、或 NSF 資助的實驗室、或學術研究型大學而言，都非常合理。

They would all be accelerated if our scientific talent could get many more of their hypothesis tested than are today due to the limitation of the manual lab. Next slide. listen, Nebula is showcasing what is possible, and that means that early adopters are getting excited about it. So we are also building autonomous labs for that left end of the chart here, the very earliest adopters, the people that are excited to try this out as a different way as an alternative to their lab benches. And so we'll keep leaning in there, building those systems as that demand comes in.

如果我們的科學人才能夠測試比今天更多的假設，他們都會被加速；而目前之所以做不到，是因為手動實驗室的限制。下一張投影片。聽著，Nebula 正在展示「可能做到什麼」，這也意味著早期採用者對此感到興奮。因此我們也在為圖表左端、也就是最早期的採用者建置自主實驗室——那些很興奮想把這當作不同方式、作為實驗台替代方案來嘗試的人。所以我們會持續投入，隨著需求進來就建置那些系統。

And we are seeing, if you go to the next slide, a lot of interest. So we've had 600-plus visitors in the first quarter. I'll show it at the end, we have a great -- like a little sign up. You can sign up. We do tours weekly, if any of you want to sign up or listening in, we're very happy to give you a tour.

而且我們也看到——如果你看下一張投影片——很多興趣。第一季我們接待了 600 多位訪客。我會在最後展示，我們有一個很棒的——像是一個小小的報名方式。你可以報名。我們每週都有導覽，如果你們任何人想報名或正在收聽，我們非常樂意帶你參觀。

So okay, so that's Nebula and that's the dive on that, right? Now I want to talk a little bit about our service businesses, Cloud Labs, data points and solutions, which I think of a little bit, like I said, like our Starlink, right? So last year, 70% of the launches at SpaceX were Starlink, if you go to the next slide. That's a huge advantage for SpaceX. That means they get to be creating an asset, a moneymaking asset in the form of Starlink while also getting to test over and over again, their launch platform.

好，所以那就是 Nebula，以及對它的深入介紹，對吧？現在我想稍微談談我們的服務型業務：Cloud Labs、data points 與 solutions；我把它們想成有點像——我之前說的——我們的 Starlink，對吧？所以去年 SpaceX 的發射中有 70% 是 Starlink，如果你看下一張投影片。這對 SpaceX 是巨大的優勢。這表示他們一方面能以 Starlink 的形式打造一個資產、能賺錢的資產，同時也能一再地測試他們的發射平台。

And their launch platform, ultimately, I think in their view, is the big product, right, that they can have that sort of transportation layer to space. But Today, they're 70% of the demand for that platform, right? I see a similar situation with the autonomous lab. We are able to have a big system here in Boston and basically prove out moving over our work from data points, Ginkgo Cloud Lab solutions, even our reagents business onto that platform. And if you go to the next slide, really excited.

而他們的發射平台，最終我認為在他們看來才是主要產品，對吧？也就是他們能擁有那種通往太空的運輸層。但今天，他們對那個平台的需求有 70% 來自他們自己，對吧？我在自主實驗室上看到類似的情況。我們能在波士頓這裡擁有一個大型系統，基本上驗證把我們的工作從 data points、Ginkgo Cloud Lab solutions，甚至我們的試劑業務，轉移到那個平台上。而如果你看下一張投影片，真的很令人興奮。

We got our cloud lab off the ground just in the last quarter, it's really been exciting. This is from The Times of London. Do you want to run an experiment for $39 -- there's a lot to do it for you. Go check out cloud.ginkgo.bio. You can go in the estimate tab at the top, type in whatever protocol you're interested in.

我們的雲端實驗室就在上一季才剛起步，真的非常令人興奮。這是《倫敦時報》（The Times of London）的報導。你想用 39 美元做一個實驗嗎——有很多人可以替你做。去看看 cloud.ginkgo.bio。你可以在上方的 estimate 分頁，輸入你感興趣的任何協定。

It will look up and see do we have the equipment needed to do your protocol? And if so, it will make an estimate of what the price would be to run that protocol in a cloud lab. And people are, I think, pretty surprised at how inexpensive it can be. And that is a reflection of where all the costs lie in doing lab work, which is in. Manual lab work done, 40 hours a week done at low equipment density, low equipment utilization in laboratories that cost a fortune to run.

它會去查詢並確認：我們是否具備執行你協定所需的設備？如果有，它就會估算在雲端實驗室執行該協定的價格。我想大家都很驚訝它可以這麼便宜。而這反映了做實驗室工作的成本到底在哪裡：在手動實驗室工作——每週 40 小時、設備密度低、設備利用率低——而且是在營運成本高得驚人的實驗室裡完成。

That then flows through, and it means all of the CRO services you order and so on are very expensive. We think we can solve that problem through automation and the cloud.ginkgo.bio or cloud lab service is really a great way to do that. If you go to the next slide, this is what OpenAI took advantage of when we did this project where GPT5 ran the lab, and we had an awesome result back in February, we showed that after six rounds of design, we had improved the cost of cell-free protein synthesis by 40% over scientific state-of-the-art, that opened a lot of eyes.

這些成本會一路傳導下去，導致你訂購的各種 CRO 服務等等都非常昂貴。我們認為可以透過自動化來解決這個問題，而 cloud.ginkgo.bio 或雲端實驗室服務，正是一個很棒的方式。如果你看下一張投影片，這也是 OpenAI 在我們做這個專案時所利用的：GPT5 來跑實驗室。我們在 2 月得到非常棒的結果：在六輪設計之後，我們把無細胞蛋白質合成的成本，相較於科學界最先進水準降低了 40%，這讓很多人眼睛一亮。

I think people weren't really -- we didn't know ahead of time whether the models would even be able to design experiments and interpret data at this level of sophistication. So really excited about that, really excited about future work we're going to be doing to keep proving this out with AI.

我想大家之前其實不太——我們事先也不知道模型是否真的能以這種精密程度來設計實驗並解讀數據。所以我們對此非常興奮，也對未來將持續用 AI 來驗證與推進這件事的工作感到非常期待。

It's a neat line of work. I would say it's distinct from the autonomous lab, I call this really an AI scientist. Using the autonomous lab, using a cloud lab to get its work done. But it is all a really important thing to watch if you're following kind of how AI is changing science. On the next slide.

這是一條很有意思的工作路線。我會說它與自主實驗室不同；我把這稱為真正的「AI 科學家」：使用自主實驗室、使用雲端實驗室來完成它的工作。但如果你在關注 AI 如何改變科學，這整件事都非常值得觀察。下一張投影片。

Also excited just in the last quarter, three new channels coming to our delivery business to our cloud lab and data point service. Amazon Biodiscovery got launched by AWS which is basically a platform to allow you to design antibodies. All three of these are sort of in the antibody space, Benchling similarly and then Tamarind Bio. These are -- Tamarind and Amazon are sort of ways for pharma companies to access these frontier bio models. So if you think of things like AlphaFold, which got the Nobel Prize for Demis at Google, those -- that was like one of the earliest protein design models.

另外就在上一季，我們也很興奮看到三個新通路進入我們的交付型業務——也就是我們的雲端實驗室與 data point 服務。AWS 推出了 Amazon Biodiscovery，基本上是一個讓你設計抗體的平台。這三個都算是在抗體領域：同樣還有 Benchling，以及 Tamarind Bio。這些——Tamarind 和 Amazon——某種程度上是讓製藥公司能存取這些前沿生物模型的方法。所以如果你想到像 AlphaFold，那個讓 Google 的 Demis 拿到諾貝爾獎的成果——那算是最早期的蛋白質設計模型之一。

There's many more now. They're computationally intensive. They're interesting and they help drug discovery scientists come up with a design for an antibody or a protein for their drug. But then you got to test it, right?

現在還有更多。它們需要大量運算。它們很有意思，並且能幫助藥物發現科學家為他們的藥物提出一個抗體或蛋白質的設計。但接著你還是得測試它，對吧？

Like we don't know if these things work in biology unless you go into the lab. And so the idea is, could you have these layers where you access the latest models and all the compute to power them. And then when you're ready to do your experiment, you hit a button and it kicks the designs to a cloud lab to do it for you and the data flows back very nicely, well packaged right to the model and you can run that loop as many times as you want. So that's sort of what's going on with Amazon and Tamarind and then Benchling is really the leader in electronic lab notebooks. And it's a similar idea.

就像我們不進實驗室，就不知道這些東西在生物學上是否真的有效。因此概念是，你能否建立這些層級，讓你可以存取最新的模型以及驅動它們所需的所有運算資源。然後當你準備好做實驗時，你按下一個按鈕，它就把設計送到雲端實驗室替你執行，而資料會非常順暢、包裝完善地回流到模型端，你可以想跑多少次就跑多少次這個迴圈。這大概就是 Amazon 和 Tamarind 的合作在做的事；而 Benchling 則是電子實驗室筆記本（ELN）的真正領導者。概念也很類似。

If you're in ELN as a scientist and you've designed this experiment, could you ultimately hit go and kick it off to a cloud and we partner there with our data point service again around antibodies. So super exciting to see these. I think this is like early indications of a way that could become a norm for how scientists do their work in the future and kind of order their laboratory experiments. I'll just say a couple of more quick things about data points. Really excited about the progress here, working with 10 of the top biopharma companies in the world just in the first year of running it.

如果你作為科學家在 ELN 裡設計了這個實驗，最終是否能按下開始，將它送到雲端啟動？我們在那裡也再次透過我們的 Data Points 服務合作，聚焦在抗體相關。看到這些真的非常令人振奮。我認為這像是一些早期跡象，顯示未來科學家如何進行工作、以及如何安排實驗室實驗，可能會逐漸成為常態。再快速補充幾點關於 Data Points 的事：我們對這裡的進展非常興奮，在啟動的第一年就與全球前 10 大生物製藥公司中的 10 家合作。

It's a good mix of pharma and government and even tech companies and tech bio companies. We've done a nice job on the next slide of really being a community leader here. We're running competitions. There's a virtual cell pharmacology initiative where we'll actually test compounds for free. People should definitely check that out if you're in the small molecule drug discovery space.

合作組合很不錯，涵蓋藥廠、政府，甚至科技公司與科技生物公司。我們在下一張投影片也很好地扮演了社群領導者的角色。我們在舉辦競賽；也有一個虛擬細胞藥理學（virtual cell pharmacology）倡議，我們會免費測試化合物。如果你在小分子藥物發現領域，真的應該去看看。

So really need opportunities and we hope to see summits and things like that. It's been good. I think AI as applied to the design of drugs is a big area. And with data points, we're sort of operating almost like a scale AI, like creating those just big data packages to train the models. All right.

所以確實需要這些機會，我們也希望看到高峰會之類的活動。這一直很不錯。我認為 AI 應用在藥物設計是一個很大的領域。而透過 Data Points，我們的運作方式有點像 Scale AI——建立那些大型資料套件來訓練模型。好。

Next slide. We have had a long-standing business in solutions, more than 250 of these research partnerships over the last 10 years. It's gotten us to work with the R&D groups of some of the largest companies in pharmaceuticals, industrial biotech and agricultural biotech. And uniquely at Ginkgo, it is a huge range of different kinds of research from microbes associated with the roots of corn and trying to engineer them to produce fertilizer to mRNA therapeutics or antibody development and pharmaceuticals to enzymes for industrial biotech, really wide range of different types of genetic engineering and biotech lab work that has happened at Ginkgo in sort of not totally automated way. In other words, not like no people in the lab, but like semi-automated.

下一張投影片。我們在解決方案（solutions）業務上有長期的基礎，過去 10 年累積了超過 250 個研究合作夥伴關係。這讓我們能與一些最大型公司的研發團隊合作，涵蓋製藥、工業生物科技與農業生物科技。且在 Ginkgo 很獨特的是，研究範圍非常廣：從與玉米根部相關的微生物、並嘗試工程化它們以產生肥料，到 mRNA 治療或抗體開發與製藥，再到工業生物科技用的酵素——在 Ginkgo 發生過的各種基因工程與生物科技實驗室工作類型非常多元，而且是以某種「不完全自動化」的方式進行。換句話說，不是實驗室裡完全沒有人，而是半自動化。

So human interacting with a liquid handling robot and a human interacting with various benchtop devices that can take a lot of samples at once. So we were sort of like not all the way to an autonomous lab, but we're doing a lot of variable work for years in semi-automated setups. And so if you go to the next slide, I'm most excited to move this kind of work on to Nebula. It is the hardest work to move, right? This is the stuff that really is that car, I mentioned earlier, the lab bench.

也就是人員與液體處理機器人互動，人員也會與各種桌上型設備互動，這些設備可以一次處理大量樣本。所以我們算是多年來在半自動化配置下做了很多變動性很高的工作，還沒有完全走到自主實驗室（autonomous lab）。因此如果你看下一張投影片，我最興奮的是把這類工作搬到 Nebula 上。這是最難搬遷的工作，對吧？這就是我先前提到的那台車——實驗室工作台（lab bench）。

It's totally variable. It's really different. It's not just doing the same experiment over and over again like you would in a traditional CRO. But if you remember my slide, it's where 95% of the spending is going at all of our customers. They spend a bit with us, but they mostly spend on huge internal research labs to do this kind of work.

它完全是可變的，差異非常大；不像傳統 CRO 那樣只是重複做同一個實驗一遍又一遍。但如果你還記得我的投影片，這正是我們所有客戶 95% 的支出所在。他們在我們這裡花一點，但大多數支出都投入在龐大的內部研究實驗室，用來做這類工作。

And so we want to replace the manual lab bench, migrating the work from our solutions business onto Nebula is a really critical demonstration. So I'm excited about the progress there. We're trying to share that publicly, and we bring people through. If you go to the next slide, one of the best things we do is we bring people through, show them a lab, let them talk to our scientists, see how scientists are submitting new protocols every day. And this has been really exciting to bring research leaders from -- I don't know, three heads of pharma or ag R&D come through to visit just this year, right, to see the system.

因此我們想要取代手動的實驗室工作台，把我們 solutions 業務中的工作遷移到 Nebula 上，是一個非常關鍵的示範。所以我對那裡的進展很興奮。我們也嘗試公開分享，並帶大家實地參觀。如果你看下一張投影片，我們做得最好的事情之一就是帶大家參觀，讓他們看實驗室、和我們的科學家交流，看看科學家如何每天提交新的實驗流程（protocol）。而且把研究領導者帶來參觀一直非常令人振奮——我不知道，今年就有大概三位藥廠或農業研發的負責人來訪，對吧？來看這個系統。

And so if you just want to visit, there's the link, you really should come by. But I think Nebula and our services on top of it is a truly unique asset to demonstrate what we think fundamentally is a better way to do biotech R&D. And we would love to get it in at every company out there and replace their benches. So if you go to the next slide, that is the world that I want to see. And so please, if you're interested, you can e-mail me at jason@ginkgobioworks.com.

所以如果你也想來參觀，連結在那裡，你真的應該過來看看。但我認為 Nebula 以及其上的服務，是一項真正獨特的資產，用來展示我們認為從根本上更好的生物科技研發方式。我們也很希望把它導入每一家企業，取代他們的工作台。所以如果你看下一張投影片，那就是我想看到的世界。因此拜託，如果你有興趣，可以寄信給我：jason@ginkgobioworks.com。

Happy to follow up and happy to take your questions now. Thank you.

很樂意後續跟進，也很樂意現在回答各位的問題。謝謝。

Great. Thanks, Jason. (Operator Instructions)

很好。謝謝你，Jason。（接線員指示）

We have one to start off submitted from Brendan at TD. We got it over e-mail. He has two questions. So the first one is, how should we think about the potential impact to revenues this year from the AWS and Benchling announcements? How have the launches gone thus far?

我們先從 TD 的 Brendan 透過電子郵件提交的一題開始。他有兩個問題。第一個是：我們應該如何看待 AWS 與 Benchling 公告對今年營收可能造成的影響？到目前為止，這些上線/推出的進展如何？

And what is baked into your assumptions for the rest of 2026 for these new platforms?

以及在你們對 2026 年剩餘期間的假設中，這些新平台納入了哪些內容？

Yes, I can take that one. So yes, we talked about AWS and Benchling. The other one in that same category as the Tamarind Bio partnership as well. I'm super excited about this. I mean this is the first time I've seen this sort of kind of like cloud layer talking directly to labs as a sales channel.

好的，我可以回答這題。是的，我們談到了 AWS 和 Benchling。同一類別裡還有 Tamarind Bio 的合作夥伴關係。我對此非常興奮。我的意思是，這是我第一次看到這種有點像雲端層（cloud layer）直接與實驗室對話、並作為銷售通路的模式。

So I'm excited to see where it goes. It is definitely new, right? So like seeing like a flood of inbound there. We are seeing some people are reaching out to us because of the channel, so that's exciting. I'm most excited that it's starting around antibodies, right?

所以我很期待它會走到哪裡。這確實是新的，對吧？所以像是看到大量的主動詢問湧入。我們確實看到有些人因為這個通路而主動聯繫我們，這很令人振奮。我最興奮的是它從抗體開始，對吧？

Because that's just kind of naturally there's a number of these AI models associated with antibodies and so on and because there's a few different providers that will do these antibody services for you. But what I'm most excited about is with our cloud lab, we're not limited to testing an antibody binding, right? If you look already on the, I don't know, eight or , 10 protocols we posted, we're posting a new one every week. It's a pretty wide variety of stuff. We're doing mass spec metabolomics, all kinds of things.

因為很自然地，確實有不少與抗體相關的 AI 模型等等，而且也有幾家不同的供應商可以替你做這些抗體服務。但我最興奮的是，透過我們的雲端實驗室，我們不僅限於測試抗體結合（binding），對吧？如果你看我們已經發布的——我不知道——8 個或 10 個實驗流程（protocol），我們每週都會新增一個。內容相當多元。我們在做質譜代謝體學（mass spec metabolomics）以及各式各樣的事情。

And so you can come and ask for a protocol and Cloud lab, we'll add it. I'd love that to turn into a channel straight from an electronic lab notebook or whatever, where a scientist is like, this is the protocol I want, price it. You get a price back from cloud.ginkgo.bio and then you go run your experiment. I think that's a much -- that feels a lot closer to AWS and sort of like what we saw is successful with cloud compute. Than where these are today, which is really much more just in a more narrow lane around antibodies, which I think is an exciting place to start.

所以你可以來提出一個流程需求，雲端實驗室（Cloud lab）就會把它加進去。我很希望它能變成一個直接從電子實驗室筆記本或其他工具連過來的通路：科學家說「這就是我想要的流程，幫我報價」，你從 cloud.ginkgo.bio 收到報價，然後就去跑你的實驗。我認為那樣會更——那感覺更接近 AWS，以及我們看到雲端運算成功的那種模式。相較之下，現在這些還比較像是在抗體這個較窄的跑道上，我覺得作為起點很令人興奮。

But I am super excited to fan that out. I think that -- then it could become really quite an interesting channel and something that scientists just don't have access to today. At the end of the day, you can't get custom stuff done. So I think that's what I am most excited about there.

但我非常期待把它擴展出去。我認為——那樣它就可能成為一個相當有趣的通路，也會是科學家今天根本無法取得的東西。歸根結底，你沒辦法把客製化的事情隨時外包完成。所以我想這就是我對此最興奮的地方。

Cool. All right. Next question from Brendan. What are you hearing on data points and the collective AI-driven offerings with Ginkgo are especially attractive for customers as biotech and pharma companies continue to roll out their own AI capabilities. In other words, what kind of demand dynamics are you seeing here?

酷。好。下一個問題來自 Brendan。關於 Data Points 以及與 Ginkgo 的整體 AI 驅動產品，你們聽到哪些回饋？隨著生技與製藥公司持續推出自家的 AI 能力，這些產品對客戶特別有吸引力。換句話說，你們在這裡看到什麼樣的需求動態？

And are there any potential revenue funnel unlocks we should watch for over the coming quarters from this part of the business?

另外，從這塊業務在接下來幾個季度，有沒有任何我們應該留意的、可能打開營收漏斗的觸發點？

Yes. So I've been super -- I mean we launched data points, almost 1.5 years ago now, and to have 10, the top pharma companies as customers now is really exciting. I think the revenue unlock is just repeat business from those customers. And so we are starting to see that and what we saw, what sort of like pilot projects, data gen projects and then now you've got again because you are seeing people trying to build in-house models. Now remember, like these are not reasoning models.

有的。所以我一直非常——我是說，我們推出 Data Points 到現在差不多 1.5 年了，而現在能有 10 家頂尖製藥公司成為客戶，真的很令人振奮。我認為營收解鎖點就是來自這些客戶的重複採購。因此我們開始看到這種情況：一開始是一些試點專案、資料生成（data gen）專案，然後現在你又看到——因為大家都在嘗試建立內部模型。請記得，這些不是推理模型。

These are not like in-house versions of Claude or Codex or OPUS or whatever or GPT5. They are models trained on biological data. So they're much more specialized. And so I do think it makes sense actually in the field that you're going to see a lot of people having their own data sets, their own models that are sort of tuned up versions maybe of various protein models. That's not going to be uncommon at all, much more common than I think you'll see in the reasoning model and coding space because these things are very different and people have different data sets.

這些不是像 Claude、Codex、OPUS 或者 GPT5 之類的內部版本。它們是用生物資料訓練的模型，所以更專門化。因此我確實認為，在這個領域你會看到很多人擁有自己的資料集、自己的模型——可能是各種蛋白質模型的調校版本——這一點其實很合理，而且一點也不罕見；我認為這會比在推理模型與寫程式（coding）領域更常見，因為這些東西非常不同，而且每家都有不同的資料集。

And so I'm, sort of, hopeful as people are building these models, we'll keep seeing the sort of repeat demand as they're like, okay, I found one. I like what I'm seeing in terms of return on data and performance of my internal model, give me more data. And so that's the revenue unlock. And the more that we see, then I think we become sort of like a default provider that's certainly what happened with scale and other places in the early days of image models and then language models when people saw, oh, I'm seeing performance increase with more data, they turn around and bought more data. That's what we're going to be watching as these protein models and other and it does not disrupt other types of models come out too in the future.

所以我有點希望，當大家在建這些模型時，我們會持續看到重複需求：他們會說，好，我找到一個了；我喜歡我看到的資料投報（return on data）以及我內部模型的表現，給我更多資料。這就是營收解鎖點。而我們看到得越多，我認為我們就越會成為某種預設供應商——這在早期影像模型、後來語言模型的時候，Scale 以及其他公司就發生過：當人們看到，喔，更多資料會帶來效能提升，他們就回頭買更多資料。當這些蛋白質模型以及其他模型——而且未來也會出現不會干擾其他類型模型的模型——我們會觀察的就是這件事。

I think that's the lane for data points.

我認為這就是 Data Points 的賽道。

Cool. Sort of on the theme of AI, we have someone who is on X who asked us a question. I think this is sort of based on our project with OpenAI. How much efficiency improvements after using GPT 5.5. Any idea for space left for improvement, will this be a transitional factor?

酷。延續 AI 這個主題，我們有一位在 X 上的人問了一個問題。我想這大概是基於我們與 OpenAI 的專案。使用 GPT 5.5 之後，效率提升了多少？對於還有多少改善空間有概念嗎？這會是一個過渡性的因素嗎？

Yes. So we had this project and just to remind that we announced back in February with OpenAI, our first project with them, where we had GPT, it's actually not 5.5, it's 5. We started much earlier, and that was when -- that was the model that was out and we kind of kept the same model through the whole thing for like more scientific paper purposes.

是的。所以我們有這個專案——提醒一下，我們在 2 月宣布了與 OpenAI 的第一個專案——當時我們用的是 GPT，其實不是 5.5，而是 5。我們更早就開始了，當時——那就是當時可用的模型，而且我們在整個過程中基本上都維持同一個模型，主要是出於科學論文用途之類的考量。

And so we were able to show over a series of six rounds of running the model with 100, 384-well plates designed by GPT5 per round, a 40% improvement over state-of-the-art in the scientific goal we were trying to achieve. I think there's real interesting questions, a, how much further could you push that, like sort of what is actually diminishing returns look like in some of these scientific areas?

因此我們能夠展示：在連續六輪的模型運行中，每一輪由 GPT5 設計 100 個 384 孔板（384-well plates），在我們試圖達成的科學目標上，相較於最先進方法提升了 40%。我覺得這裡有一些非常有趣的問題：a，還能把它推到多遠？也就是在某些科學領域裡，所謂邊際報酬遞減（diminishing returns）實際上會長什麼樣子？

Can the model have sort of breakthrough ideas that create really new ways of doing this? TBD. And then as the models have gotten better, yes. And would 5.5 be better than what we got with 5, right? I think that's all going to be exciting stuff to test.

模型能不能提出某種突破性的想法，創造出全新的做法？尚待觀察（TBD）。然後，隨著模型變得更好，是的——那 5.5 會不會比我們用 5 得到的結果更好，對吧？我覺得這些都會是很令人興奮、值得測試的事情。

So we're excited to do more with OpenAI and we're planning to. And so I think this is an open terrain in terms of how good the reasoning models can be at basically experimental design and experimental analysis. That's -- those are the two things it's really doing. It's like here's an experiment. I want to run, give me back the data cloud lab, autonomous lab, give me back what are the results of my experiments I just designed and then I'm going to analyze them and design more experiments.

所以我們很期待與 OpenAI 做更多合作，而且我們也計畫這麼做。因此我認為，在推理模型究竟能在「實驗設計」與「實驗分析」上做到多好這件事上，這是一片開放的疆域。那——它真正做的就是這兩件事。就像是：這是我想跑的實驗，把資料回傳給我——雲端實驗室、自主實驗室——把我剛設計的實驗結果回傳給我，然後我會分析它們，再設計更多實驗。

We'll say, I think it's real exciting to watch what's going to be capable of there. It's a new way to do science. It really is. Like -- and I won't belabor this too much, but I think it roughly can turn individual -- like the access to a model like that plus an autonomous lab can let individual scientists operate closer to how a principal investigator of an academic lab or a head of a drug discovery group who has lab of eight people or a lab of 30 people and it's sort of assigning hypotheses to different people and kind of pursuing that over time.

我會說，我覺得觀察那裡將會具備什麼能力真的很令人興奮。這是一種做科學的新方式，確實是。像——我不想在這裡講太久，但我認為大致上，它可以讓個別——也就是取得那樣的模型再加上一個自主實驗室，能讓單一科學家的運作方式更接近學術實驗室的主持人（principal investigator），或是一個藥物發現團隊的主管：他們有 8 人或 30 人的實驗室，會把假說分派給不同的人，並隨時間推進。

An individual could push that out for probably close to the same cost as they are currently costing to be themselves at a lab bench in terms of their just -- their utility costs and everything else and utilization, low utilization of equipment, they could push out five agents on top of an autonomous lab to go pursue a bunch of experiments.

一個人可能可以把這件事擴展出去，而且成本大概接近他們目前在實驗台前「作為自己」的成本——就他們的水電等公用成本、以及其他一切與設備低利用率相關的成本而言——他們可以在自主實驗室之上再部署 5 個代理（agents），去追一大堆實驗。

That is real exciting if that works. I think it really fundamentally changes the rate we can do science. That's why you see the Genesis mission in the US investing in this sort of stuff because their goal is to 2 times the output of US science.

如果那真的可行，會非常令人興奮。我認為它確實會從根本上改變我們做科學的速度。這也是為什麼你會看到美國的 Genesis mission 投資這類事情，因為他們的目標是把美國科學的產出提高到 2 倍。

That's a way that will do it. And our science-based industries, of which pharma is the biggest will be completely changed by this. If you can, 2 times or 3 times the rate, no question about it.

這就是能做到的方法。而我們以科學為基礎的產業——其中製藥是最大的——將會因此被徹底改變。如果你能把速度提高到 2 倍或 3 倍，毫無疑問。

All right. Our next question is really a bundle of questions from DK, who's writing from South Korea. And these questions are all about how the move on to Nebula, our autonomous lab has sort of changed the science that we're doing. So the questions are, how does the use of Ginkgo's automated lab affect overall costs? Are there meaningful differences in speed, for example, turnaround time for experiments?

好。下一個問題其實是一組問題，來自在南韓寫信來的 DK。這些問題都在問：轉向 Nebula——我們的自主實驗室——之後，如何改變了我們正在做的科學。所以問題是：使用 Ginkgo 的自動化實驗室會如何影響整體成本？在速度上是否有顯著差異，例如實驗的周轉時間（turnaround time）？

And have you observed improvements in success rates, reproducibility or scalability since moving to the autonomous lab?

另外，自從轉到自主實驗室之後，你們是否觀察到成功率、可重現性（reproducibility）或可擴展性（scalability）有所提升？

Yes. So on cost, I tried to touch on this a little bit in the talk. But I think like the clear ROI not just for us but for any one of our customers looking at an autonomous lab, is about a threefold reduction in space utilization compared to a manual lab and a fourfold increase in that time. In other words, like the amount of time the lab is being used to do lab work, right, from that 40 hours to 168, 24/7 week. That's really that -- those improvements is where it's going to yield the cost reduction.

是的。先講成本，我在談話中有稍微提到。我認為很明確的投資報酬（ROI）——不只是對我們，對任何考慮自主實驗室的客戶也是——在於：相較於手動實驗室，空間使用效率大約可降低到三分之一（也就是空間利用率改善約 3 倍），而時間利用率可提升約 4 倍。換句話說，實驗室被用來做實驗工作的時間，從每週 40 小時提升到 168 小時，也就是 24/7。真正帶來成本下降的，就是這些改善。

But that is a huge amount because those are really the two -- like sort of people time and space time are thetwo2 big things we spend money on in research. On the speed front, yes, it's interesting. An individual protocol doesn't really get shorter like than necessarily you would do it at the bench. You can imagine ways to do that in the future rebuild protocols differently. But the first thing scientists are going to do is just take work they're doing at the bench and move it on to the autonomous lab.

而這個幅度非常大，因為這兩個——也就是人力時間與空間時間——是我們在研究上花錢的兩大項。就速度而言，是的，這很有意思。單一流程（protocol）本身不一定會變得更短——不一定比你在實驗台上做更快。你可以想像未來有方法透過重新設計流程來做到。但科學家第一件會做的事，就是把他們在實驗台上做的工作搬到自主實驗室上。

And in that world, it does not need to get faster in terms of like end-to-end time for the protocol. However, in practice can get faster because you can start a protocol at 4:00 PM. in the afternoon where you never would have planned to spend the next seven hours in the lab, kick it off and have the thing run overnight. So in that world, you took an experiment that you would have started tomorrow at 10:00 a.m. and started at 4:00 PM and have the results by tomorrow at 8:00 AM or 10 AM.

在那種情境下，就流程端到端時間而言，它不需要變快。不過在實務上它可以變快，因為你可以在下午 4:00 開始一個流程——你原本不會打算接下來 7 小時都待在實驗室——把它啟動，讓它夜間運行。所以在那個世界裡，你把原本明天上午 10:00 才會開始的實驗，改成下午 4:00 就開始，然後明天早上 8:00 或 10:00 就能拿到結果。

And so that can shave a whole day off. So I think you will see actually a massive speed up because scientists will start taking advantage of the 4x more time that they have available every week. So if they plan it right, in theory, you can see a fourfold improvement in a lot of the times, depending on how serialized your experiments need to be. So I think that's really exciting in our side, I think I really like that.

因此那樣可以省下整整一天。所以我認為你們實際上會看到大幅加速，因為科學家將開始善用每週多出來的 4 倍時間。所以如果他們規劃得當，理論上，你可以在很多情況下看到四倍的改善，取決於你的實驗需要多大程度的序列化（串行）進行。所以我覺得這在我們這邊真的很令人振奮，我想我真的很喜歡這一點。

And then on the -- just sort of like improvement in like I say, I would call this like the quality of the experiments. I think reproducibility is inherently advantaged on automation. And that's mainly to do with like the audit trail. Like you kind of if an instrument errors, if a liquid handler makes mistake, these are all tracked. So you kind of know like those experiments that you don't catch at the manual lab bench, you catch if there's such a mistake on the autonomous lab.

接著在——就像我說的那種改進，我會把這稱為實驗的「品質」。我認為自動化天生就有利於可重現性。這主要與稽核軌跡（audit trail）有關。像是如果儀器出錯、如果液體處理機（liquid handler）犯錯，這些都會被追蹤記錄。所以你會知道——那些在手動實驗台上不一定抓得到的實驗問題，如果在自主實驗室發生這種錯誤，你就能抓到。

So if you saw really, wow, that's a surprising result. You might go back, look at your experiments and say like, oh, I see what I did there, I like design this experiment in a way that was like a little silly, and that's actually what's giving me this result as opposed to assuming you did the experiment you wanted to do and that was the origin of this like amazing result you got. I think yes, that's a common thing that can happen. For no nefarious reasons from scientists at the bench. And so I think that -- you will see a big improvement in reproducibility.

所以如果你看到真的，哇，這是個令人驚訝的結果。你可能會回頭看你的實驗，然後說，喔，我知道我那裡做了什麼了，我把這個實驗設計得有點蠢，而那其實才是造成這個結果的原因；而不是以為你做的就是你想做的那個實驗，然後那就是你得到這個驚人結果的來源。我想是的，這是很常見會發生的事。並不是因為實驗台上的科學家有任何不良動機。所以我認為——你會看到可重現性有很大的提升。

And then the other thing that got brought up there was throughput, the throughput increase is going to be the same. I think people are surprised when they go to cloud.ginkgo.bio, which I encourage people to do and type in a protocol and see how much it costs. Because I'm basically pricing that protocol based on reagent use and equipment time and a markup on that. And it is not the insane costs that you have when you have a whole team doing this work at the bench, it's just not. Like -- so if scientists really understood, just how low cost each sample could be in an experiment, and they did -- in order to do many more, they just hit a button rather than have to slave in the lab for three days is doing 1,000 experiments.

然後另一個剛剛提到的是吞吐量（throughput），吞吐量的提升也會是一樣的。我覺得大家會很驚訝，當他們去 cloud.ginkgo.bio（我鼓勵大家去）輸入一個 protocol（實驗流程）看看要多少錢。因為我基本上是依照試劑用量與設備時間來為那個流程定價，並在其上加一個加價（markup）。而它並不是你在實驗台上需要一整個團隊做這些工作時那種瘋狂成本，真的不是。像是——所以如果科學家真的理解，每個樣本在一個實驗中的成本可以有多低，而他們為了做更多實驗，只要按個按鈕，而不是得在實驗室苦幹三天來做 1,000 個實驗。

They're going to just order those 1,000 experiments. And so I think you will see an explosion in the amount of data. And this is 100% what happened in every other field that's ever been automated, right? It's like the beginnings of the automation of computation, right? Like when we went from slide rules to automated computation and explosion in the amount of compute you use, and a massive increase in the return on investment from what people who understood how to design a computation could do.

他們就會直接下單那 1,000 個實驗。所以我認為你會看到資料量爆炸式成長。而這在每一個曾被自動化的領域都 100% 發生過，對吧？就像計算自動化的開端，對吧？像是我們從計算尺（slide rules）走向自動化計算，計算使用量就爆炸成長，而那些懂得如何設計計算的人，能從中獲得的投資報酬率（ROI）也大幅提升。

And that's what I want to do for the scientists for drug discovery leads when they have access to an autonomous lab compared to the ROI and the throughput that they can get out of manual labs. It's just going to be no comparison. So yes, I think all three, you're going to see big gains on. And the cool thing is we're going to keep showing this on Nebula. So we just had -- Head of R&D here today, and we went through with his team and showed all the gains, and it's -- yes, it's really exciting right now.

而這就是我想為科學家在藥物發現（drug discovery）線索（leads）上做到的事：當他們能使用自主實驗室時，相較於手動實驗室所能得到的 ROI 與吞吐量，將完全沒有可比性。所以是的，我認為這三項你都會看到很大的增益。而很酷的是，我們會持續在 Nebula 上展示這些成果。所以我們剛剛——今天研發主管（Head of R&D）在這裡，我們和他的團隊一起走了一遍並展示所有的增益，而這——是的，現在真的非常令人振奮。

So I think we will end on the note, kind of, related to that, which is you guys mentioned in the call, you've mentioned other places you're trying to get to 100 RACs. When do you actually expect to get there?

所以我想我們就以一個與此相關的重點作結：你們在電話會議中提到，也在其他地方提到，你們正試著達到 100 台 RAC。你們實際上預期什麼時候能到達？

Yes. So it's been pretty fun. We have to put behind-the-scenes videos up, but we have been installing RACs for the last three weeks here at Ginkgo. I just showed up on trucks RACs built by our team in Emeryville. And we just added the additional 50.

是的。所以這一直相當有趣。我們得把一些幕後影片放上去，但過去三週我們一直在 Ginkgo 這裡安裝 RAC。我剛剛看到卡車把我們位於 Emeryville 的團隊打造的 RAC 運來。而我們剛剛又新增了額外的 50 台。

They are all fully connected now in lab. I took a tour of it, it's insane. And so -- and we can run them now, like the original system is running and now the new 50 are running, and there is a connection between the two, and that connection is going to get turned on, I think on the 14th -- next week. So it is imminent. So I'm really excited to see it all come together.

它們現在在實驗室裡都已經完全連接好了。我去參觀了一圈，簡直瘋狂。所以——而且我們現在就能跑它們：原本的系統在運行，新的 50 台也在運行，而且兩者之間有一個連接，而那個連接我想會在 14 號——下週——啟用。所以已經迫在眉睫了。所以我真的很期待看到它全部整合起來。

But we already have it up now running as two separate loops. So to put in 50 new pieces of equipment in three weeks. Again, these are just things that no one's ever done in laboratory automation. So I do think we are doing a very unique thing here at Ginkgo. That's the bet.

但我們現在其實已經以兩個獨立迴路（loops）的方式在運行了。所以在三週內新增 50 件新設備。再說一次，這些都是在實驗室自動化領域從來沒有人做過的事。所以我確實認為我們在 Ginkgo 正在做一件非常獨特的事。這就是那個賭注。

That's certainly what I'm leaning in on the company. It's what we're investing our capital into. It's where our new customers are coming from and so if you like that idea, I think that is a really exciting time to get involved with the company in any way. But yes, we're going to be at 100 next week. 103 or 105.

這也確實是我在公司裡最著力推進的方向。這是我們投入資本的地方。這也是我們新客戶的來源所在。所以如果你喜歡這個想法，我認為現在是以任何方式參與這家公司的一個非常令人興奮的時點。不過是的，我們下週就會到 100 台。103 或 105。

I got to count them, yes.

我得數一下，是的。

All right. And if you want to follow us on that journey, you can go to X or LinkedIn, Instagram and keep watching. We'll have a lot of content coming about the unveiling of the new full system. And as always, if you have questions, you can reach out to us at investors@ginkgobioworks.com. Thanks so much, everyone, until next time.

好。若你想跟著我們一起走這段旅程，你可以到 X 或 LinkedIn、Instagram 持續關注。我們會有很多內容，介紹全新完整系統的揭幕。一如往常，如果你有問題，可以透過 investors@ginkgobioworks.com 聯絡我們。非常感謝各位，我們下次見。

Thanks, everybody.

謝謝大家。