Backblaze (BLZE) 2026年第3季財報電話會議：B2成長加速至47%

Backblaze 2026 財計第三季業績顯示，B2 雲端儲存業務成長加速、營運槓桿改善，且來自 AI 基礎設施提供商與 AI 開發者的需求上升。該公司亦調升全年的營收與調整後 EBITDA 獲利率展望。

主要重點

• 總營收年增 26% 至創紀錄的 4,700 萬美元，比管理層指引高標高出 200 萬美元。
• B2 雲端儲存營收成長 47%，高於第二季的 34%。若剔除 5 月 1 日的調漲價格影響，B2 的有機成長率為 35%。
• B2 年經常性營收 (ARR) 達 1.27 億美元，年增 53%，季增 1,400 萬美元。B2 淨營收留存率從第二季的 113% 改善至 115%。
• 調整後 EBITDA 獲利率年增 1,000 個基點至 33%。GAAP 淨虧損縮小至 200 萬美元，而 Non-GAAP 淨利則大幅成長逾兩倍至 670 萬美元。
• 管理層預計第四季營收為 4,980 萬至 5,020 萬美元，依中間值計算，B2 營收年增率約為 60%。
• Backblaze 將全年營收指引調升至 1.778 億至 1.782 億美元，並將調整後 EBITDA 獲利率指引調升至 30% 至 31%。

關鍵財務數據

業務與營運表現

B2 的成長加速普遍體現在直銷、自助服務客戶增加，以及低於預期的客戶流失率。Backblaze 新增的 B2 自助服務淨新客戶數約為去年同期的四倍。

該公司新增 19 家 ARR 超過 5 萬美元的客戶，使總數達到 254 家。來自該客戶群體的 ARR 年增 109% 至 5,000 萬美元。

Backblaze 在本季簽署了四份 ARR 超過 50 萬美元的協議。其中三份是與現有 AI 開發者的擴約，每家客戶在一年內皆將其承諾總額提升至原本的兩倍以上。一家訓練資料提供商則針對一項新 B2 工作負載簽署了一份價值超過 400 萬美元 ARR 的額外協議。

第四份大型協議是與一家全球網路技術公司簽署約 50 萬美元 ARR 的合約，採用該公司的白標儲存方案 Powered by Backblaze。

AI 開發者的動能也同步增強。進入銷售管道的新 AI 開發者數量比第二季翻倍，不過管理層強調管道中的機會並非已成交合約。來自 AI 搜尋引擎的造訪量年增 230%，且這些造訪者開立自助服務帳戶或聯繫銷售人員的比率是自然搜尋造訪者的兩倍以上。

CoreWeave 與另一家先前公布的 NeoCloud 客戶對第三季成長的貢獻有限，但已開始進入放量階段。Backblaze 提前完成了第二家 NeoCloud 客戶的開發工作，使其能夠比公司原先預期的 2027 年初提前數個月開始載入資料。

電腦備份業務營收年減 2% 至 1,600 萬美元，但季增。管理層將業務趨穩歸因於客戶獲取與留存舉措。

管理層指引

修正後的全年營收中間值比先前展望高出 500 萬美元。調整後 EBITDA 獲利率指引從 27%-29% 上調。

管理層表示，該展望未包含潛在價值超過 50 萬美元 ARR 的新協議，以及超過合約最低使用量的變動使用量。

根據營收指引的中間值，該公司預計第四季的 PP&E 資本支出為 6,500 萬至 8,300 萬美元，視交付時間而定。這筆支出的大部分預計將由可轉債所得資金資助，並反映在自由現金流中。

風險與重點領域

Backblaze 正在客戶使用量全面放量之前部署基礎設施。管理層預期由此產生的折舊將暫時拉低 GAAP 毛利率 3 至 6 個百分點，最大的影響將從 2027 年第一季與第二季開始。隨著客戶使用量增加，毛利率預計將有所回升。

該公司預計在未來四至五個季度內投入資本，以支援來自 CoreWeave、其他 NeoCloud 提供商及 AI 開發者的合約需求。管理層指出，每增加 1 美元的 ARR 約需要 2 美元的資本支出，預估回收期為兩年。

執行情況取決於硬體、資料中心和電力是否能及時到位。Backblaze 已取得硬體配額，而目前的訂單履約時間約為四至五個月，優於其保守的九個月規劃假設。

客戶放量的時程是另一個變數。CoreWeave 的五年合約總額為 3.35 億美元，產能承諾將按月增加直至 2027 年 7 月。管理層表示對時間表仍感滿意，且尚未發現任何可能影響承諾放量的跡象。

分析師問答焦點

• NeoCloud 策略：管理層表示並未放緩對新 NeoCloud 客戶的爭取。目前的首要任務是執行已簽署的協議，並成功交付數 Exabyte 級別的容量。
• CoreWeave 時程：CoreWeave 對第三季 47% 的 B2 成長貢獻極小。其貢獻自第四季起開始增加，並在 2027 年第一季與第二季加速，最終於 2027 年第三季達到初始容量部分的最低營運率。
• 需求多元化：管理層指出，需求涵蓋自助服務使用者、ARR 超過 5 萬美元的客戶、大型 AI 開發者與 NeoCloud 提供商。公司表示，這種廣度使其能夠對客戶與營收品質進行更多選擇。
• 資本經濟效益：CoreWeave 合約中約 70% 涉及由 Backblaze 資本支出資助的 Backblaze 資料中心容量。其餘 30% 則是在 CoreWeave 出資的資料中心使用 Backblaze 平台進行管理儲存，且具備更高的毛利率。
• AI 市場策略：Backblaze 將 AI 基礎設施提供商視為規模更大、週期更長的機會，而 AI 開發者則代表了具有即時儲存需求且更廣泛的客戶群。

法說會完整逐字稿

完整財報電話會議逐字稿

管理層陳述

Operator

Hello, everyone. Thank you for joining us, and welcome to the Backblaze Third Quarter 2026 Financial Earnings Call. [Operator Instructions] I will now hand the conference over to Mimi Kong, Director of Investor Relations and Corporate Development.

Mimi, please go ahead.

Mimi Kong

Thank you. Good afternoon, and welcome to Backblaze's Third Quarter 2026 Earnings Call. On the call with me today are Gleb Budman, Co-Founder, CEO and Chairperson of the Board; and Marc Suidan, Chief Financial Officer. Today, Backblaze will discuss the financial results that were distributed earlier.

Statements on this call include forward-looking statements about our future financial results, the impact of our sales and marketing initiatives, cost-saving initiatives, results from new features, the impact of price changes, supply volatility and pricing, our ability to compete effectively and manage our growth, and our strategy to acquire new customers, retain and expand our business with existing customers. These statements are subject to risks and uncertainties that could cause actual results to differ materially, including those described in our risk factors that are included in our most recent quarterly report on Form 10-Q and our other financial filings. You should not rely on our forward-looking statements as predictions of future events. All forward-looking statements that we make on this call are based on assumptions and beliefs as of today, and we undertake no obligation to update them, except as required by law. Our discussion today will include non-GAAP financial measures. These non-GAAP measures should be considered in addition to and not as a substitute for our GAAP results. Reconciliation of GAAP to non-GAAP results may be found in our earnings release, which was furnished with our Form 8-K filed today with the SEC.

You can also find a slide presentation related to our comments in the webcast, which will also be posted to our Investor Relations page after the call. Please also see our press release or presentation for definitions of additional metrics such as NRR, gross customer retention rate, and adjusted free cash flows.

We will be participating in the Craig-Hallum Alpha Select Conference on November 17 in New York, and the Needham Virtual Tech Week on November 18 and 19. We look forward to connecting with you all.

Thank you for joining us, and I would now like to turn the call over to Gleb.

Gleb Budman

Thank you, Mimi, and thank you, everyone, for joining us today. We had a strong third quarter as we continued to deliver on our strategy of being the high-performance capacity storage layer for AI. Revenue reached $47 million and B2 revenue growth accelerated again to 47% year-over-year. What stands out to me is also the breadth of that acceleration, ranging from our smallest self-serve customers to our largest sales-led customers.

We added roughly 4x the number of net new B2 self-serve customers versus the same period last year. Customers over $50,000 in ARR grew by 19 to 254, showing our upmarket momentum is repeatable. And we again closed 4 deals over $500,000 in ARR, a sign of the value we deliver to the largest data users. All of that reflects the value of our platform, along with the progress we're making on our AI strategy and go-to-market execution.

Now let me share what I'm seeing in the market. In conversations with customers, AI founders and infrastructure providers, I'm hearing two consistent themes. First, data growth keeps exceeding expectations. Our early AI customer storage needs consistently outpace even their own aggressive forecast. Playing models continue to require ever larger data sets and AI applications and agents are generating more data and giving companies new reasons to retain and reuse existing data sets.

And second, companies are increasingly focused on getting more efficiency and ROI from AI. They're leaning into open models using routers to select the optimal model for each task and working across multiple AI clouds to improve economics. Storage tiering is one of the important ways to improve that ROI also. If you have a router to select the right model, why wouldn't you apply that same discipline to selecting the right storage location.

Together, those trends create a growing opportunity for Backblaze. Customers need to collect, retain and use more data while getting a better ROI from their infrastructure spending. Our high-performance capacity tier helps them do both.

As we've discussed, we serve 2 groups: AI infrastructure companies to whom we provide a capacity storage tier for their customers, and AI builders who need capacity storage for themselves. Let me start with the progress we're making with our AI infrastructure customers.

While we're continuing to progress in conversations with multiple leading AI infrastructure companies, we believe the best way to build on this opportunity is to deliver for the customers who have already chosen us, bring their capacity online and give them a great experience as they scale. We've secured the necessary hardware allocations to fulfill our requirements and shipments are tracking ahead of schedule.

Our team also finished development ahead of plan for the large NeoCloud customer we announced at the start of the year, and that customer began loading data several months ahead of our early 2027 expectation. Additionally, CoreWeave continues to be a great partner and has begun to ramp on our platform. Thus, both agreements are moving into execution and each carries a multiyear minimum commitment, so revenue is underpinned by contract and the ramp follows our customers loading data into capacity we deploy.

Turning to AI builders, which use Backblaze directly. We're building a repeatable, scalable way to reach these customers, help them get started and support them as they grow. These builders work across areas like foundation models, physical AI and generative media. Their products differ, but many share the same need, a capacity tier that grows with them as they train and tune models and generate new inference data.

We're seeing strong momentum here. The number of new AI builders entering our pipeline doubled in Q3 compared with Q2. This is pipeline, not closed deals, but we're encouraged in driving it three ways.

First, helping builders discover Backblaze. Years of trusted original content, including nearly 13 years of Drive Stats have established Backblaze as an authority on storage. AI search tools draw on content like this to answer storage questions for people and agents, giving Backblaze more opportunities to appear in those answers. This visibility resulted in visits from AI search engines growing 230% year-over-year, roughly double the market rate. And those visitors create a self-serve account or contact sales at more than twice the rate of visitors from organic search, which traditionally had been the gold standard. We believe this is strengthening customer acquisition across our broader business through both self-serve and direct sales.

Second, deeper ecosystem relationships. Through Flamethrower, we reach builders via VCs, start-up communities and events like TechCrunch and coast-to-coast start-up lab. Hundreds of start-ups now participate in the program, letting us build relationships early and support their storage needs as they develop their businesses.

Third, making it easier for developers and their AI agents to build with Backblaze. We've published reference architectures for B2 and AI workflows and expanded the tools developers use to integrate our storage. For example, our new open-source B2 MCP server lets AI agents use B2 storage directly with governance through customer-defined permissions. This growing pipeline and engagement across the ecosystem reinforce our confidence in our product market fit for AI builders, and we're focused on turning that interest into lasting customer relationships and long term growth.

Now on to the 4 agreements over $500,000 ARR signed this quarter. We had 3 expansions with existing AI builders, showing how customers often start small and grow, and won brand-new agreement for a core workload. First, the training data provider we highlighted earlier this year that signed a nearly $1 million agreement in just 11 days and then expanded by more than $1 million just a couple of months later. This quarter, they signed a separate incremental agreement worth more than $4 million in ARR for a new workload on B2. Second, a generative AI media company expanded as its AI training workloads and data grew. Third, an AI voice company is connecting its GPU providers compute directly to Backblaze, allowing it to store more of its data with us.

Each of those 3 increased their total commitment to more than double its original level within a year. Across these examples, customers are bringing more workloads to Backblaze and making us a deeper part of their architecture.

The fourth agreement highlights continued demand for core workloads, a market we estimate at $26 billion for 2026. A global networking technology company signed a new agreement worth approximately $500,000 in ARR for Powered by Backblaze, our white label offering, replacing the varied storage setups its customers bring with one simpler experience and creating a new revenue stream. Our track record brought them to us. Our economics and ease of integration made the offering work.

Q3 was a great quarter. More customers chose Backblaze, existing customers expanded, and large AI infrastructure customers began to ramp. Together with accelerating B2 growth, that reinforces our role in AI and our ability to deliver on the opportunity ahead.

As we look forward, training, checkpoints, logs, and inferencing are all part of the AI workflow that's generating data. Foundation models, generative AI, multimodal, physical AI and many more are categories of AI organizations and use cases generating data and agents are excelling how all of that is being accessed and used. The net of that is companies seeing massive growth in their needs for storage while also aiming to drive clear ROI from their investments. That makes for an exciting time for Backblaze.

With that, I'll turn it over to Marc. Marc?

Marc Suidan

Thanks, Gleb, and good afternoon, everyone. Q3 was another quarter of accelerating growth and improving operating leverage. Total revenue grew 26% year-over-year to reach a new record quarterly high of $47 million, led by B2 Cloud Storage, where revenue growth accelerated to 47%, up from 34% in Q2, and 28% a year ago. Adjusted EBITDA margin expanded 1,000 basis points year-over-year to 33%. Both revenue and adjusted EBITDA exceeded the high end of our guidance, and we are again raising our full year guidance.

Let me walk through the quarter, starting with revenues. Total revenue was $47 million, up 26% year-over-year and $2 million above the high end of our guidance. The outperformance was broad-based, driven by strength in our B2 direct sales bookings, higher B2 self-serve customer growth and less-than-expected churn. B2 revenue grew 47% year-over-year, our strongest growth rate in 10 quarters.

The May 1 price increase contributed approximately 12 percentage points to B2's growth. Excluding it, B2 organic revenue grew 35% compared with 26% on the same basis in Q2. B2 ARR reached $127 million, up $14 million from Q2 and 53% year-over-year.

We also continue to move upmarket. ARR from our $50,000-plus customers grew 109% year-over-year to $50 million. As expected in their large committed contracts, CoreWeave and the other previously announced NeoCloud deal did not contribute significantly to Q3's growth acceleration. Both have now started to ramp, and we expect meaningful contributions from both of them to grow in the coming quarters. B2 net revenue retention was 115%, up from 113% in Q2. Computer Backup revenue was $16 million, down 2% year-over-year, but up sequentially as our growth and retention initiatives stabilize revenue.

Moving on to gross margin. Total company gross margin was 64%, up from 62% in the prior year with the pricing update continuing to help. Operating expenses were 67% of revenue, improving from 72% a year ago as revenue grew faster than our expense base.

Adjusted EBITDA margin reached 33%, exceeding guidance on stronger-than-expected revenue results. Our GAAP net loss narrowed to $2 million as we continue to make progress towards GAAP profitability. Non-GAAP net income was $6.7 million, more than 3x the $1.9 million reported a year ago. Cash flow from operations was $23 million year-to-date compared to $14 million a year ago.

Adjusted free cash flow was $2 million year-to-date compared to negative $10 million last year. Year-to-date PP&E additions, net of dispositions, totaled $33 million. We continue to expect full year PP&E capital expenditures of 55% to 65% of revenue. Based on the midpoint of our revenue guidance, this implies $65 million to $83 million in Q4, subject to delivery timing.

We expect to fund most of these expenditures with cash from our convertible debt proceeds, and those cash outlays will be reflected in free cash flow. Over the next 4 to 5 quarters, we expect to deploy capital to support contracted demand, including the CoreWeave, other Neo clouds, and AI builders commitments.

We are adding capacity ahead of customer ramps, which means depreciation will begin before the associated revenue, temporarily reducing GAAP gross margins by 3 to 6 percentage points. As customer usage ramps, we expect cash generation to strengthen and gross margin to recover.

Turning to the balance sheet. We ended Q3 with $230 million in cash, cash equivalents and marketable securities. Our 0 coupon convertible financing funds this build-out without the cash interest expense associated with financing the same equipment through capital leases.

We also purchased cap calls to mitigate potential equity dilution. Assuming the debt principal is repaid in cash, the cap call completely offsets dilution up to $33.76 per share. Our earnings presentation includes a table detailing net dilution at various stock prices above that threshold.

Turning to guidance. For Q4, we expect revenue of $49.8 million to $50.2 million. At the midpoint, we expect B2 revenue growth to accelerate to approximately 60% year-over-year in Q4. This strong exit rate gives us confidence in our 40% plus growth outlook for B2 in 2027.

For adjusted EBITDA margin in Q4, we expect a margin of 32% to 34%. For the full year, we are raising revenue guidance to $177.8 million to $178.2 million, a $5 million increase at the midpoint. For full year adjusted EBITDA, we are raising our 2026 fiscal year margin outlook to 30% to 31% from 27% to 29%. Consistent with our guidance philosophy, this outlook excludes potential new deals greater than $500,000 in ARR in variable usage above contracted minimums.

Q3 demonstrated continued progress against our objectives, accelerating B2 growth and translating that growth into operating leverage. We are investing against committed demand while maintaining operating discipline, positioning the business for the next phase of growth across a diversified set of customers.

With that, operator, please open it up for questions.

Operator

[Operator Instructions] Your first question comes from the line of Ittai Kidron with Oppenheimer & Co.

分析師問答

Ittai Kidron

Congrats on good numbers here. One question for you. I guess, Gleb, I'll start with you. I just want to make sure I understand your comments on the AI infrastructure customers. If I interpret that correctly, you said something to the effect of focus on current customers and make sure that you ship and ramp up them appropriately. Am I interpreting this to mean that you're -- I don't know, temporarily or permanently, you tell me -- are kind of slowing your chase of new Neo clouds and you're more focused on ramping existing ones? Is this a change in strategy? Am I interpreting that correctly?

Gleb Budman

Thanks for the question. It's not a change in strategy. We still are going after the Neo clouds. We have a team that's working on it, and we're in discussions with a number of them. It's more a statement that the ones that we have signed, both CoreWeave, but also the other large one in particular that we announced at the beginning of the year, these are significant scale-up opportunities for us.

And we want to make sure that we do a really good job. And so it's a lot of organizational focus on just executing relentlessly on that. And we think that by doing that, that's going to be the best path we have to really take advantage of the overall opportunity with the broader set of Neo clouds. So not a change in strategy.

We still believe that there's a big opportunity for us to service the broader NeoCloud landscape, just a recognition that there's a lot of work to be done to make sure that we hit the full multi-exabyte deliverability that we're doing for them. And then separate from that is, obviously, the AI builders opportunity for us is also very large. And so it's just executing on both in parallel.

Ittai Kidron

Okay. Very good. And then, Marc, I just want to make sure I understand the numbers and your guide for the next quarter. Again, just to clarify, it doesn't sound like CoreWeave contributed much, if any, to the third quarter. Your guidance to the fourth quarter, if you take into account the beat that you had in the quarter now, I would have to think that CoreWeave needs to and should contribute more in 4Q. So is it fair to say that there's -- just the step-up there hasn't been a substantial one relative to the beat and how you roll numbers? So am I interpreting that to mean that you're being extra conservative here still on CoreWeave itself? Or is CoreWeave more of a '27 ramp? Help me think about the timing here around this.

Marc Suidan

Ittai, CoreWeave ramps all the way through middle of next year. They're on-track for that. You are correct that in Q3, it was not that meaningful to the 47% growth. That was more broad-based across everything else. As it relates to the 60% B2 growth we guided for Q4, there's definitely a bit of CoreWeave there, and there's also the overperformance of Q3 as it relates to the 4 large deals greater than $0.5 million we announced in there. And then CoreWeave and the NeoCloud really start kicking-in in higher gear in Q1 and then Q2 of next year and then full ramp Q3 of next year.

Operator

Your next question comes from the line of Jason Ader with William Blair.

Jason Ader

I guess, Marc, just on the fourth quarter gross margin. So you didn't give specific guidance for that. You talked about kind of a 3- to 6-point impact from all the CapEx depreciation. But is it right to think that it won't be that big an impact in Q4 just because the CapEx is coming online in the quarter, and therefore, it's more of a next year or Q1 event?

Marc Suidan

Yes. That's correct, Jason. A lot of it starts arriving in Q4, right, because we signed our big deal in June. We started placing all our orders. If it comes at the tail end of Q4, there's a bit of an impact on depreciation in Q4, thus gross margin, but it really picks up more in Q1 and Q2 of next year. And then thereafter, it starts recovering for the 4 quarters after that.

Jason Ader

Okay. And just to be clear, you're still comfortable with the time line on the CoreWeave ramp, even though there's a lot of concern in the market right now?

Gleb Budman

Jason, this is Gleb. So the short answer is yes. We've been a great partner. We're working closely with them. We don't see any signs of any risk or any concern. It's a committed contract with committed minimum ramps. We're building toward that. It's obviously on us to make sure that we execute well on delivering everything that we need to deliver to them. And I think that there's clearly a lot of demand for storage out in the market. And there's nothing that we see that would jeopardize that at this point.

Marc Suidan

Yes. And Jason, this is Marc again. What I would add to that is we're very well capitalized. I mean, between the convertible debt offering we did, the cash balance we had and the capital lease lines, we have access to over $400 million to be able to fund whatever we need. We're not going to spend $400 million, but we have access to it. So we're well capitalized to continue down this route, both for CoreWeave as well as the broad-based demand we're getting and the additional signals we're seeing.

Jason Ader

Let me ask you this final question then. If for some reason, something went wrong with CoreWeave, how confident are you that you could find the demand for that additional capacity that you're bringing online?

Gleb Budman

I think, Jason, kind of just broadly, kind of as I was saying in the prepared remarks, I think we're seeing overperformance across all of our segments, everything from Flamethrower, which had a record month in new sign-ups. We had 100 -- over 100 companies sign up for Flamethrower this month alone, which is down at the very beginning at the start-up side of the market to the full self-serve customers to the over $50,000 ARR customers and then over $500,000 ones.

So I think we're seeing broad successful demand across the whole business. So if there's hiccups and bumps in demand from some parts, the overall demand that we're seeing is very strong. And it's actually allowing us to be more judicious with the customers that we're bringing on to the platform at this point, which has been nice. It's allowed us to be somewhat picky a little bit on quality of customers and revenue that we accept onto the platform.

And so obviously, CoreWeave was a really great partner. We're working closely with them. We expect to work well with them in delivering for them. They're a big deal for us, and we would -- but we do have a lot of demand coming from a lot of parts across the spectrum.

Operator

Your next question comes from the line of Mike Cikos with Needham.

Michael Cikos

Congratulations on a successful quarter. For Gleb or Marc, I just wanted to see, and I don't want to -- I'm trying to get a sense of the individual factors here when I'm thinking about the increased guide for 4Q, which is great to see, right? We're talking about the strength of the B2 self-serve customers. We're talking about these 4 large agreements. We're talking about CoreWeave, which I think you had said actually started loading their data several months ahead of what was expected.

And really, the question is, can you help us tease apart those various factors if we're trying to figure out what is the core 3Q outperformance that is benefiting 4Q here versus maybe some of these larger like a CoreWeave or that other NeoCloud scaling faster than anticipated?

Gleb Budman

Mike, this is Gleb. So first of all, just to clarify, it wasn't CoreWeave that started a few months before their plan. It was the other NeoCloud that we announced at the beginning of the year that where we had development work that we talked about that we needed to do in order for them to start being able to utilize the platform. We finished that development work early, which is a great testament to our team and the way that the team has been able to execute on an engineering and operations side. And then that customer has started loading data earlier by a few months. We had originally said we expected it to start at the beginning of 2027. So that's my take to it. I'll let Marc maybe touch on more of the individual components for the guide.

Marc Suidan

Yes. Mike, good to be talking to you. When you think about the -- we've announced 11 deals so far this year, of greater than $0.5 million ARR. That all fits into why this guide is getting to 60% year-over-year growth. And of course, there's CoreWeave and the other NeoCloud in there, but the acceleration that's been happening this year is really strong.

And a lot of these customers, as Gleb noted, they're -- I would say they're what we consider AI builders, companies with about $0.5 billion to $1 billion in market value. So pretty mid- to late-stage start-ups growing well. We already -- I mean, typically, we see ARR in them of $100 million to $200 million. So all that seems to be running well. So we feel pretty comfortable with what we're seeing, pretty well-diversified and kind of very healthy business models.

Michael Cikos

Okay. And I know you had commented as well on CapEx for Q4 and potential impact and headwinds on gross margin until the revenue ramps behind it. Are we at a point where you can help us think about CapEx next year, just given the market's interest or appetite for data points surrounding capital efficiency and returns on those capital requirements?

Marc Suidan

Yes. So we previously shared at the Investor Day that we got about a 2-year payback to recover our CapEx spend, right? So if you were to think about where ARR ended at the end of '25 and where it could potentially end at the end of '27, let's hypothetically say that's about $150 million. That would require $300 million of CapEx, right? And for the lifetime of that CapEx, that would generate $1 billion of revenue in aggregate, not per year. So that's the economics there. So you're seeing the CapEx ramp now really accelerated. Any CapEx acceleration from our side is predominantly tied to revenue growth because for every dollar of ARR we're adding, we need to fund $2 of CapEx. And so that's where you're seeing that pick up in Q4.

Like we said at the Investor Day also, we will probably pay for most of those in Q4 in cash. So you'll see that in the free cash flows. And then afterwards, we'll step back and see how much we want to re-blend in the capital lease model back into that. So we're continuously evaluating those options.

Operator

Your next question comes from the line of Erik Suppiger with B. Riley Securities.

Erik Suppiger

Congrats on a good quarter. Can you talk a little bit about how you would compare the market opportunity as you see it today between the AI builders and the AI infrastructure providers? It sounds like maybe with interest rates going higher, you're placing your bet more broadly on the AI builders. Is that a reasonable assumption?

Secondly, just on CoreWeave, can you describe the trajectory of how that business is going to ramp between now and third quarter of next year? Is it going to be -- is it going to ramp up right as you get into the middle of the year? Or how should we think of that trajectory?

Gleb Budman

Thanks for the question. So first of all, on the AI infra versus the AI builders. So one of the things we had talked about before is that AI infra companies, there's a couple of hundred of those companies out there. And they're at different stages of their maturity curve. And so what we found is that ones that are just at the earlier stages, all that they're focused on is building up their...

Lost my thought.

Marc Suidan

Somebody is typing hard, so.

Gleb Budman

Yes, the AI infra companies, like the ones that are earlier in their maturity curve, they're very focused on just being able to stand up data center, power, and GPUs. The ones that are further along in their maturity curve, they're actually thinking about the broader customer workflows and helping them solve those.

Those that are further along are the ones that are starting to lean into the realization that they need to provide storage as a part of their services. And as they do that, they realize that capacity tier of storage is critical to serving the actual required workflows of their customers.

So in part, it's -- some of these couple of hundred are further along. Many of them are still on the earlier stage. So we think that over the next couple of years, many of these AI infra companies are going to get to this place where they come to that step in their journey where they really are serving the broader workflows and they're going to need Backblaze as a core piece of that. In the meantime, we're working closely with a number of them to start further down that path.

On the AI builder side, part of the reason that we're talking about that side is there's so many of these companies out there. They have large storage needs and they have them today. So I was looking at some of our companies that are AI builders on our platform. We have a frontier lab that's focused on the radiology space. We have a frontier lab focused on biotech research. We have a frontier lab focused on quantitative modeling for financial services firms.

We have an AI data provider that's focused on providing a variety of data sets for enterprises. We have a surveillance platform company that's focused on leveraging AI to enable all their surveillance functionality. We have a company that's doing AI applications delivery for enterprises.

So these are all a variety of AI builders and there's 33,000 venture-funded start-ups that are in this kind of grouping. And that's just the AI start-ups, not including the midsized scale-ups and other enterprises that are doing building in the AI space.

So it's an area where we see a lot of opportunity to have scalability, repeatability, and the ability to service them today while we work on these individual larger opportunities on the AI infra.

Marc Suidan

Yes. And Eric, I can dive to the other part of your question, and then I'll start off with a data point we previously shared, right?

The NeoCloud capacity storage market is expected to be $14 billion by 2031. So whether it's CoreWeave or other players, I mean, that's them selling to these similar AI builders and AI start-ups. So that market, right, when they -- these companies raise money, unlike their predecessor start-ups pre-AI, right? They have a P&L line item, data storage, right? We got a report that says 15% of their spend is on data storage. And that's obviously varying types of data storage. So it's suddenly a much bigger part of what a start-up is today than what it was before because it's a key asset for them.

Now you asked about the CoreWeave ramp. So if you remember, the CoreWeave contract is $335 million over 5 years. About 70% of that is them operating in our data centers and our CapEx and about 30% of that is using our platform, but they fund the CapEx in their data centers. So that 30% is the managed storage part. It's obviously a much higher gross margin. That kicks in a bit later.

For the first part, the 70%, that's the ramp now that you're seeing. And there's a two-way minimum commitment, right? We got to deliver a certain capacity, ramps up every month all the way through July of 2027 next year, and that's their commitment to us as that ramps up as well. So it's a two-way commitment. So once we get to July 2027, that 70% of the contract reaches its minimum run rate for the remainder of the contract.

Operator

Your next question comes from the line of Eric Martinuzzi with Lake Street Capital Markets.

Eric Martinuzzi

Shifting over to the Computer Backup side of the business. Those numbers in Q3 outperformed certainly versus what I was modeling. I'm curious to know, I think you had talked last quarter about kind of a down mid-single digits range or maybe it was low to mid-single digits, but what's implied in the 2026 for CBU?

Marc Suidan

So for Q4, we're still implying that low single-digit decline year-over-year. That's what's in the guide. And what you saw in Q3 was the initiatives we've taken to retain or grow customer acquisitions have stabilized the revenues. So that's why sequentially, there was a slight increase. But as it relates to the guide, we're still guiding the same way we did in the previous quarter until we see that pattern solidify.

Operator

Your next question comes from the line of Jeff Van Rhee with Craig-Hallum Capital Group.

Unknown Analyst

This is Daniel on for Jeff. Congrats on the quarter, great quarter. In terms of maybe you can walk us through a little bit -- we've talked about the build-out that's going to be happening for CoreWeave starting in Q4 here. Maybe when we look to the rest of the business, you could just walk us through some of the key items you're looking at, hurdles, long lead items, sort of what are the things you need to get in place? How responsive is the build-out? If you see a surge in demand, how long does it take to build out to meet that? And I don't know any other thoughts in terms of building obviously, the rest of the B2 business has been accelerating.

Gleb Budman

Daniel, thanks. I appreciate the comments. So the short of it is it's a huge step-up, right? We're basically increasing the CapEx by something like a factor of 10 over the course of the next year-ish. So it's definitely a large increase across all the aspects of the business. The good news, I would say, and Marc can touch more on this, but the supply is something that we have allocated from the vendors. We work closely with them.

And we don't -- allocated does not mean that we have them in our possession, and we've paid for them. It means that we have the allocation and we pay for them once we actually get them. But it does allow us to have confidence in being able to deliver on our commitments for our partners.

Marc Suidan

Yes. And there's two aspects. One is the hardware. I mean, we used to plan -- to be conservative, we plan on 9 months to fulfill from order to fulfillment. It's coming in closer, I'd say, between 4 and 5 months. So we're doing well there. Things are on schedule as it relates to the hardware side. The data center, I mean, no secret, it is a top election issue. We actively monitor that to make sure we're expanding at the right rate because at the rate of growth we are growing -- we've had quite a few expansions, and we're actively making sure we have enough capacity there to sustain our growth.

Those are the two aspects we got to look into because as you noted, Daniel, right, you've got the CoreWeave commitment, you've got the other NeoCloud commitment. You've got these 11 deals greater than $0.5 million. You've got the self-serve sign-ups of more than quadruple. And there's been one consistent trend since B2 launched in 2016, which is every year, the data from the existing customer grows, and that's before even AI hit the scene. And now with AI, we're just seeing the velocity of growth of that data really accelerate.

So with that, those are the elements we look for. It's the data center and the hardware. The power comes with the data center. So they go hand-in-hand in the hardware. We are predominantly on high-capacity drives from the main vendors. There's also some -- a bit of compute and networking that fits into there as well.

Unknown Analyst

And then just in regards to Computer Backup and multiple exabytes that you have on the more traditional business, is there a potential over time for some of that capacity to potentially shift or be allocated toward B2 as you see more opportunities there?

Gleb Budman

Yes. It's an interesting question, Daniel. I mean, I think the short of it is that the Computer Backup business itself is kind of roughly a flattish business. So it becomes a smaller and smaller percentage of the overall infrastructure. We do look continuously at ways to optimize that business. Some of the optimizations are churn reduction and growth, which is optimizing on stabilizing the revenue like we did this quarter, which was an exciting moment. And some of it is about optimizing the data footprint.

And so we do have initiatives underway to look at how to do that further. To the extent that we do that, that space can be repurposed for some of the B2 use cases. It's not the, I would say, the core focus for the way we do it, but it is an opportunity for it.

Operator

We have reached the end of the Q&A session. I will now turn the call back to Gleb Budman for closing remarks.

Gleb Budman

Great. And thank you, everybody, for joining us. This was a great quarter. We showed broad-based momentum across our business, 53% year-on-year growth in B2 ARR. We see AI being a year's long tailwind for our business as we provide the capacity tier that AI desperately needs. Our mission is to make customers unstoppable by solving their toughest storage challenges and the pace of AI innovation is making those challenges dramatically more urgent and greater.

I want to thank our customers for trusting us to support your businesses and innovation, our partners and investors and our team for doing all the hard work to make all of this happen. Thank you for joining us, and we look forward to talking with you again next time. Operator, you can end the call.

Operator

This concludes today's call. Thank you for attending. You may now disconnect.