Innventure (INV) 2026财年第二季度业绩电话会议：Accelsius营收目标暂停

核心要点

• Innventure报告2026财年第二季度综合营收为100万美元，而2025财年第二季度为50万美元，2026财年第一季度为140万美元。其中，Accelsius贡献了90万美元，占总营收的96%。
• 本季度净亏损从第一季度的2780万美元扩大至3490万美元。调整后EBITDA亏损为2260万美元，而上一季度亏损为1840万美元。
• 在电力供应、GPU获取以及场地分配等限制因素延迟了较小规模早期采用者的部署后，Innventure暂停了Accelsius的营收目标。
• 在客户失去已确定的开发场地后，Accelsius将DarkNX项目从其2026财年预测中移除。管理层表示，这与该场地的电力承载上限有关。
• Accelsius不再预计能在2026财年实现现金流收支平衡。Innventure也撤回了到2028财年实现综合正现金流的目标。
• 管理层将通过四个推广采用里程碑来评估Accelsius：芯片厂商的参考设计参与、与OEM和ODM的联合开发、与超大规模云厂商签订的工作说明书，以及额外的第三方基准测试数据和部署。

核心财务数据

业务与运营表现

Accelsius

Accelsius仍是Innventure的主要营收贡献者，但管理层表示，AI基础设施市场的结构性限制打乱了其近期商业化进程。GPU分配、可用电力以及对服务器设计的影响力仍集中在大规模超大规模云厂商手中，这限制了原本可以作为早期采用者的较小规模公司。

因此，该公司已将其几乎所有的合作和市场推广活动重新聚焦于四大客户群体：芯片厂商、服务器OEM、服务器ODM以及超大规模云厂商。管理层表示，Accelsius正在与多家大型超大规模云厂商进行积极的概念验证和评估流程，不过这些机构的采购和部署周期较长。

Accelsius还报告称，其NeuCool两相芯片直接液冷技术获得了第三方验证。在配备8颗英伟达B200 GPU和约4万个运行点的同款戴尔PowerEdge XE9680L服务器上，NeuCool使GPU的温度比出厂预装的单相系统低9°C至14°C，同时在芯片端的冷却液流量仅约三分之一。

管理层表示，在设施供水温度为50°C时，单相系统超过了B200的84°C降频临界点，而NeuCool则保持了9°C的余量。根据Jacobs的参考设计，该公司表示，两相液冷可以在相同电力消耗上限内使GPU数量平均增加5%。

管理层引用了相关预测，即液冷市场规模将在2030年突破300亿美元，其中两相液冷市场规模将达到90亿美元。这些数据系行业预测，而非公司本身的营收展望。

AeroFlexx

AeroFlexx的商业项目管线达到近3500万美元，较上一季度增长9%。该业务扩大了在拉丁美洲和欧洲的合作伙伴关系。

在与Packaging Himalayas达成联合制造合作伙伴关系后，AeroFlexx在合作方位于意大利的工厂安装了灌装设备。目前该设备已投入运行，产品认证工作正在进行中。

Refinity

Refinity针对一座1万吨示范工厂工程设计的交付工作按计划将于2026年底前完成。该公司还在开展长时间工艺运行测试，管理层此前将该测试的目标时间定为2026年夏季。

管理层业绩指引

Innventure此前预计，Accelsius将在2026年底接近现金流收支平衡，年化营收运行率约为1亿美元。管理层目前预计收支平衡将在2026年之后实现，并已暂停Accelsius的营收目标，直至部署限制得到缓解或公司达成基础性的推广采用里程碑。

该公司还撤回了Innventure在2028年实现综合正现金流的目标。管理层计划在Accelsius的推广采用和营收创造能见度提高时，重新评估这一时间节点。

对于2027年，管理层表示现在对具体预订额或营收能见度给出预测还为时过早。超大规模云厂商的采用周期可能需要两到三年，具体取决于客户及评估阶段。

Innventure预计AeroFlexx和Refinity将在运营子公司层面筹集更多资金。然而，管理层承认，Accelsius现金流实现时间的推迟可能需要在母公司层面引入额外资金，潜在可能包括在2026年下半年开展进一步融资。

风险与关注要点

• 在两相液冷技术得到更广泛的应用之前，预计Accelsius的预订额和营收将继续保持不均衡且难以预测的状态。
• 较小的潜在客户在获取电力、GPU和合适部署场地方面面临限制。
• DarkNX必须确定替代场地并满足其他条件，其订单才能重新纳入Accelsius的预测中。管理层预计订单将会转移，但时间仍不确定。
• 大型超大规模云厂商的工程、采购和基础设施规划流程漫长，这为商业部署的时间节点带来了不确定性。
• 由于Accelsius不再预计能在2026年实现现金流收支平衡，Innventure可能需要在母公司层面筹集额外资金。
• 未来的融资可能会稀释股东股权，尽管管理层表示打算相机筹集资金，并保持Innventure在Accelsius的权益敞口。

分析师问答环节亮点

管理层拒绝透露在此前提及的Accelsius 5000万美元预订额中，有多少与DarkNX挂钩。公司表示该订单预计将转移至新场地，但由于获取替代地点和所需配额的时间表尚不确定，Accelsius已取消该笔预订。

Accelsius表示与超大规模云厂商的洽谈非常积极，且处于不同阶段。在初始工作说明书之后，典型流程可能包括技术验证、概念验证部署、融入客户的数据中心参考设计，并最终进行批量部署。

管理层强调，与芯片厂商、服务器OEM或ODM以及超大规模云厂商的对接必须并行推进。终端客户需求可以加速OEM和ODM的支持，但Accelsius在出现大规模需求信号之前就已经在建立这些合作关系。

关于流动性，管理层表示Innventure的资产负债表上有现金，并拥有多种融资渠道，包括备用股权购买协议。公司未就潜在追加融资的具体结构或时间表提供细节。

业绩电话会议完整文字记录

---

完整财报电话会议逐字稿

管理层陈述

Operator

Good afternoon, and welcome to Innventure's Second Quarter 2026 Earnings Conference Call. [Operator Instructions] As a reminder, this conference call is being recorded. If you have any objections, please disconnect at this time.

I would now like to turn the call over to Kyle Nagarkar, Investor Relations. Please go ahead.

Kyle Nagarkar

Thank you, Mariana, and good afternoon, everyone. Welcome to Innventure's Second Quarter 2026 Earnings Call. With me today are Bill Haskelll, Chief Executive Officer; Dave Yablunosky, Chief Financial Officer; Dr. Bill Grieco, our incoming Chief Executive Officer; and John Hewitt, Chief Executive Officer of Accelsius.

Earlier today, we issued a press release announcing our financial results, which is available on our Investor Relations website, along with the supplemental slide presentation. As referenced on Slide 6, we will be discussing non-GAAP financial measures during this call. The most directly comparable GAAP financial measures and a reconciliation of the differences between the GAAP and non-GAAP financial measures are available on our earnings release and supplemental slide presentation on our website.

In addition, certain statements being made today are forward-looking statements that are based on management's current assumptions, beliefs and expectations concerning future events impacting the company. These forward-looking statements involve a number of uncertainties and risks, including, but not limited to, those described in our earnings release Form 10-Q for the period ended June 30, 2026, and other filings with the SEC. The actual results of operations and financial condition of the company could differ materially from those expressed or implied in our forward-looking statements.

With that, I'll turn the call over to Bill Haskell.

Gregory Haskell

Thank you, Kyle. Good afternoon, everyone, and thanks for joining us. We're going to run today's call a little differently by focusing the majority of our time on Accelsius. You'll hear from four executives today. Dave will take you through the financials, then I'll say a brief word about the leadership transition we announced in June, followed by Bill Grieco to share what to expect under his new leadership. And finally, John Hewitt, who took over as CEO of Accelsius in July, will walk you through where the business is headed and where the industry is headed with it.

Let me give you the headline for Accelsius before we get into it. First, we believe the market is now debating when two-phase will be adopted, not if. Second, allocation of GPUs and memory, difficulties accessing power and two-phase enabled servers are impacting smaller early adopters. That has consequences for near-term revenue expectations, which Dave will address directly in his remarks. But here's the more important point. We believe those same forces have made the long-term picture for two-phase cooling better, not worse. John will walk you through exactly what changed and why.

Before Dave addresses the numbers, let me give a quick update on AeroFlexx and Refinity. At AeroFlexx, the commercial pipeline continues to build and is now close to $35 million, up 9% since last quarter. The company's global reach continues to expand with new partnerships in Latin America and Europe. In addition, following the May 11 announcement of the co-manufacturing partnership with Packaging Himalayas, AeroFlexx filling equipment has been installed and is operational at the Italian facility with product qualification underway. At Refinity, engineering design on the 10-kiloton demonstration plant is on track for delivery of a plan by the end of this year. You'll hear Bill Grieco come back to that in a few minutes.

Now let me pass it to Dave to take us through the financials.

David Yablunosky

Thanks, Bill. Good afternoon, everyone. Consolidated revenue for the second quarter was $1 million. That compares to $0.5 million in the second quarter of last year and $1.4 million in the first quarter of this year. Of the $1 million, Accelsius contributed $0.9 million or 96% of the total.

Net loss for the quarter was $34.9 million compared to $27.8 million in the first quarter. Adjusted EBITDA was a loss of $22.6 million versus $18.4 million in Q1. General and administrative expenses were $14.5 million, down 22% compared to the second quarter of 2025.

We ended the quarter with $46.5 million of cash and restricted cash. That compares to $60.4 million at the end of Q1, which also includes $5 million of restricted cash. Year-to-date, we used $59.5 million of cash in operating activities and generated $41.6 million from financing activities.

During the quarter, we took several steps to strengthen our balance sheet and manage our capital efficiently. We raised approximately $13 million through draws on our standby equity purchase agreement at an average price of $6.21. We also continue to reduce debt, including the full repayment of our convertible debentures earlier this year.

Shifting now to our outlook. We previously expected Accelsius to exit this year near cash flow breakeven at an annualized revenue run rate of roughly $100 million. We now expect the timing for Accelsius to breakeven to extend beyond this year. The primary driver for the change is market dynamics, not a change in our conviction around the technology or the market opportunity.

Smaller early adopters are facing constraints around power availability, GPU access and site allocations. Those resources continue to be concentrated among the largest hyperscalers.

I'd like to now directly address our DarkNX purchase order, given they too are not immune to these dynamics. The purchase order referenced a development site that DarkNX had previously identified. DarkNX recently informed Accelsius that this site is no longer available and that it's working towards developing alternate sites. Accelsius has removed the DarkNX project from its 2026 forecast, pending identification of an alternate deployment location in satisfaction of other conditions.

This single customer setback does not change the fact that the Accelsius remains at the front edge of market adoption. But until that adoption is established, order bookings and revenues are going to be lumpy and hard to predict. Due to these structural constraints, which limit early adopter deployments, we are suspending our revenue targets. We intend to reinstate forward-looking targets once those constraints ease or we achieve the foundational KPIs that drive broad industry adoption.

Here's what we are committing to instead. We will report on the key milestones best representative of our progress towards market adoption, which John will detail in his section. At the parent level, due to the revised timeline for Accelsius to achieve positive cash flow, we are no longer targeting consolidated positive cash flow for Innventure in 2028. We will revisit that expected timing when we have greater visibility into the pace of Accelsius adoption and revenue generation.

Taking a step back, let me revisit our capital strategy in the context of this revised outlook. First, it reinforces the need to be disciplined about where capital is raised and how we protect shareholder ownership. Second, we expect to be opportunistic in how we fund the business from here as revenue delays naturally precipitate a need for capital.

Our intent is to finance AeroFlexx and Refinity increasingly at the operating company level, which limits the amount of capital we need to raise at Innventure and helps minimize dilution for Innventure shareholders. At the same time, given the extended timeline for Accelsius to reach positive cash flow, we recognize there will be a need for additional capital at Innventure. When we raise capital, we intend to do it thoughtfully, opportunistically and with a goal of preserving Innventure's pro rata exposure to a Accelsius.

With that, I'll pass it back to Bill Haskell.

Gregory Haskell

Thanks, Dave. Before I introduce our next two speakers, a brief personal note. As we announced on June 30, I'll retire as CEO of Innventure on October 1 after almost six years leading the company and more than four decades in the industry. This was a planned succession, and I'll be working closely with Bill Grieco through the transition.

I'm confident in the handoff because Bill is not an outside hire learning about the company. He helped build it and had previously served as Innventure's Chief Technology Officer. For the past one and half years, he has been the founding CEO of Refinity, and he took that business from a blank sheet of paper to the doorstep of commercial demonstration. Before Innventure, he built and led innovation and new business creation at various large organizations and held a PhD in chemical engineering from MIT. Finally, he has served on the Boards of both Refinity and Accelsius. So he knows every one of our businesses from the inside and is the right leader for Innventure's next phase. Bill, over to you.

William Grieco

Thank you, Bill, for the introduction and for your six years of leadership that got Innventure to this point. I'm honored and I'm excited to take the baton.

Let me start with what will not change. Innventure's company building philosophy is the same under me as it was under Bill. We build operating companies around breakthrough technologies in partnership with multinational corporations, and we run those companies to win in their respective markets. We provide the initial funding and ongoing back-office support to allow the operating companies to focus on efficient operations and to meet their commercialization objectives. That's our value proposition, and it's why many of you have invested in us.

In the past, we tried to communicate in a way that's more typical of established public companies, especially in regard to providing revenue targets. For businesses like ours, innovative technology businesses addressing markets that are still forming, predicting revenue quarter-by-quarter is difficult, if not impossible. As a result, revenue targets are not the most useful yardstick at this stage. Revenue guidance will become more appropriate in the future as these companies mature. But in the growth phase, the better measure of progress is whether we're setting the right milestones, whether we're achieving them and whether we're building towards the inflection points that can create significant long-term value.

This framework speaks to how we will evolve our operating discipline. I'm an engineer by training and an operator by career. I believe management teams earn credibility one milestone at a time. We set a milestone, we meet it or exceed it, then we do it again. That's how we've run Refinity. We operate with a lean organization focused on achieving our objectives on time and on budget. For example, we shared that we would be scaling up our process for extended duration runs by summer of this year, and our team is doing that now. We told you that the engineering design for our 10-kiloton commercial demonstration plant would be complete in the fall, and it's on track for delivery by the end of the year. I expect Innventure and its operating companies to be run the same way.

Now let me speak to our capital allocation policy, which remains unchanged. The capital allocation framework we announced in April still stands. Capital above a parent reserve is intended to be distributed to shareholders. Innventure will remain committed to maximizing shareholder value. We do that through smart deployment of capital with every dollar aimed where it serves our shareholders best.

One more thing about how I intend to communicate with the market. I believe in showing results, not promising them. And I believe we owe the market clarity about what we're seeing. That's what today's call is, giving you a better look into what our companies, particularly Accelsius are seeing and telling you what we're doing about it.

Now let me tell you why I'm so enthusiastic about Accelsius. I sit on its Board, and I've been involved since we started the company, and I've never been more excited about this opportunity. It's rare in a career to watch a technology this differentiated, meet a market this large at the moment the market needs it. Accelsius' $65 million Series B round led by Johnson Controls with Legrand participating tells you what sophisticated industrial players think of the technology.

Like me, John Hewitt has been involved with Accelsius from the beginning as a founding Board member. So he stepped into the CEO role with a deep understanding of the company, the technology and the market opportunity. Since taking over early last month, he's worked with the Accelsius team to review and refine the commercialization strategy, which he recently reviewed with the Board. That is exactly the kind of work he's well suited to lead.

John previously ran the Americas for Vertiv, a multibillion-dollar business at the center of the AI data center build-out. He was most recently CEO of Robertshaw, a global design, engineering and manufacturing company with more than 6,000 employees. And earlier in his career, he held senior roles at TE Connectivity, Motorola and Baker Hughes. Josh Claman built Accelsius into what it is today, and he remains fully engaged as Executive Chairman. John has joined him to help scale it.

I'm more bullish than ever on Accelsius and on two-phase direct-to-chip cooling. And John will now walk you through the business, the strategy and the magnitude of the opportunity ahead. John, welcome. The floor is yours.

John Hewitt

Thanks, Bill. Good afternoon, everyone. Bill just gave you my resume, so I won't read it back to you. Let me tell you why I took this job, and then I'll spend my time where it belongs on the business and the industry.

At Vertiv, I had about the best vantage point in the industry to watch what AI compute is doing to the thermal limits of the data center. I saw every cooling technology in the market, what worked, what scaled and what hit walls. I joined the Accelsius Board four years ago because I concluded that two-phase direct-to-chip cooling would ultimately be the answer for the most demanding AI and high-performance workloads. I took this job because I believe that ultimately is arriving faster than most people expected. You maybe get one or two opportunities like this in a career, and I am excited to be here.

One more thing before I move on. As Executive Chairman, Josh Claman remains actively engaged in this company. We have been great partners for the last four years, and I am excited to continue working with him in a different capacity.

Four years ago, Accelsius was founded on a conviction that physics would drive the industry to liquid cooling and that two-phase would earn a two-phase portion of that market. At that time, we didn't think it would become so widely understood that two-phase will be required. The market didn't just make room for us. It's coming toward us, and that's made us aim higher.

We are no longer planning like a scrappy start-up buying for single-digit market share over the next 10 years. We believe we can hold a much more significant share, and this calls about how we plan to do that by building our product thoughtfully alongside key ecosystem players, holding the attention of the companies that define the AI sector and focusing where the expected return is greatest.

So what instills that conviction in us? We believe important things outside our control are breaking our way. Physics favors too phase. Every AI generation runs hotter, and the industry has discovered what servicing single phase actually costs. There's another force accelerating all this. Data center developments are experiencing significant pushback from communities being asked to host these facilities over water and power usage concerns. Two-phase changes the energy profile of a data center. In greenfield designs, it can lower cooling-driven energy use by 1/3. At a moment when $700 billion in planned 2026 data center CapEx is colliding with $130 billion in blocked and delayed projects, and New York has just enacted the first statewide moratorium. That isn't a nice to have. It's how the industry earns the right to keep building.

We believe the question is no longer if, only when. We can't control when, but we do control how we execute and where we focus. Over the last few years, we've been pursuing two goals simultaneously. The first was building the foundation for a great company, one that could gain a meaningful share of the liquid cooling spend and one that can deliver a highly differentiated product reliably and at scale.

The liquid cooling market is forecasted to exceed $30 billion in 2030, of which $9 billion is expected to be two-phase. For context, today, there are no mass scale two-phase direct-to-chip deployments in the United States. I'm proud of the work the team has done to lay the foundation, but the next year is critical, and we'll talk about that.

The second was delivering at-scale revenue from a hard tech company inside of five years and not just any hard tech. This is an advanced technology one few companies have ever solved designed to protect GPUs, the asset whose demand far exceeds supply and easily among the most valuable line items on any AI company's balance sheet.

For context, one B300 GPU runs over $50,000 and eight-way server built on them runs $400,000 to $500,000 and a loaded rack of those servers can run between $3.5 million and $4.5 million. When we benchmarked ourselves against successful hard tech companies in the cooling space, very few had any commercial revenue in year four. Against the relevant comparisons, we're tracking ahead of the pace. The problem was never the pace. It was the yardstick we measured against.

We expected Accelsius to travel the normal tech adoption curve with smaller early adopter companies as our main revenue source for a few years. Then we learned something about the AI market. Adopting this technology requires GPU allocations, access to power and the scale to influence server designs, and those are precisely the things smaller companies can't get. This is exactly what we saw happen with the DarkNX deal Dave spoke about earlier. Our analysis and the feedback we are receiving indicates that the market structure is sidelining many of the customers who would normally take the first risk. As a result, in this market, there are very few early adopters.

Here's why that's good news. The relationships we are now focusing on are the companies that dominate this market, companies worth hundreds of billions. And they haven't just noticed us. They're showing deep and promising interest, active proof of concepts with several key hyperscalers and impressing results that are driving next steps. And make note of this because I'll come back to it. For these companies, benchmarked data-backed proof of superior performance is what drives adoption. We just delivered a major proof point.

Now these companies move slower than early adopters would have, but I want to be clear about why. We believe it's a product of how a good business makes major decisions. They have shareholders, countless customers and established procurement and build cycles. They evaluate in a mature way, deliberate studies between engineering teams, starting with single-loop cold plate level tests, then proof of concepts, then operating impact analysis and then a dedicated haul deployment. Ultimately, they're built into their IT procurement plan and into their data center road map. Some iterations run over multiple quarters, and we are in various stages of progress with many of them.

So the trade we ended up with is this. Instead of seeking quick revenue from small companies that aren't likely to scale, we're focusing instead on the technology leaders, and we are deep in the evaluation cycles with some of the largest companies in this industry. Our earliest customers are also our largest possible customers. We believe that this isn't a phase, but the GPU allocation and power scarcity, among other factors, define AI infrastructure, and they aren't easing.

When I stepped into this role last month, we did a detailed review of the Accelsius commercialization strategy and made major updates. Substantially all of our partnership and market adoption work now focuses on four customer segments: chip manufacturers, server OEMs, server ODMs and hyperscalers. Our goal is to have chip manufacturers reference our solution, OEMs and ODMs design for it and end customers incorporate those requirements into their IT and infrastructure designs. And when those players move, the market moves.

Remember, almost all the data center footprint deployed or in process today uses either air or single-phase liquid cooling. Until two-phase adoption crosses the line, bookings and revenue are going to be lumpy and hard to predict. So as Dave said, we will not guide until we see that adoption. We believe bookings and revenue are lagging indicators in this market. The milestones we will report are the ones we view as the leading indicators.

As I mentioned earlier, the next year is critical. We are actively engaged in advancing progress around major milestones and the four we are focused on now are as follows: one, chip maker engagement leading to reference designs. Inclusion in a silicon vendor's partner ecosystem would be the strongest validation this market offers. It would put us in front of every customer designing around that silicon.

Two, server OEM and ODM relationships expanding into co-development initiatives. This would be the first step toward factory integration and server warranty coverage, key enablers to market adoption. Number three, moving beyond proof of concept to an executed statement of work with a leading hyperscaler, one that scopes the power usage effectiveness and operational impacts of two-phase in their data centers. That's the difference between being evaluated and being planned for. And four, continuing to deliver benchmark data and deployment with leading thermal labs, giving the industry's strongest thermal minds the proof needed to adopt.

Hitting these four milestones is how we'll measure progress and how you'll know we're creating meaningful company value. Given the decision-making timeline within large organizations that I spoke about earlier, we don't anticipate having material updates every quarter. That said, each one of these milestones already has its own work stream underway, most with significant progress, and we look forward to updating you on further progress when warranted.

And to that end, let me update you on one huge milestone just achieved. As I said earlier, for mature customers, data backed proof is everything, and we just completed a major study. In July, we published the most important technical validation in the company's history, and I want to walk you through it because the numbers deserve more than a headline.

An independent third-party systems integrator took a commercially available Dell PowerEdge XE9680L and eight-way NVIDIA B200 server drawing roughly 10 kilowatts and benchmarked it with its factory-installed single-phase cooling. Then they retrofitted the same server with our new cool cold plates and ran it again. Same server, same GPUs, same simulated workloads, roughly 40,000 operating points. The only thing that changed was the cooling.

The results, new cool ran the GPUs 9 to 14 degrees centigrade cooler at the system level, using roughly 1/3 of the coolant flow at the chip. At 50 degrees C facility water, the single-phase system pushed the B200 past its 84 degrees C throttle point, the temperature where the GPU slows itself down to survive. Ours held 9 degrees C of headroom below it, same server, same chips, different outcome.

Now here's what we believe those degrees are worth. NVIDIA has pointed the entire industry toward warmer facility water as a key lever for AI factory efficiency and their current single-phase designs top out around 45 degrees C. Our headroom means the performance single phase delivers at 45C, we deliver it up to 54 degrees C and beyond. At those temperatures, chillers convert from a necessity into a contingency in most of the world for most of the year. That's the energy story that I opened with. This is how the industry earns the right to keep building.

But for an operator, the energy savings isn't really about the utility bill. Every data center lives inside a fixed power envelope. Whatever the grid gives you, that's your budget, and every watt spent on cooling is a watt not spent on compute. Cut the cooling load and two things happen. You make the most of the power you were allocated and you redirect those savings into the only thing that generates revenue, which is more GPUs doing more work.

Based on the Jacobs reference design, two-phase enables on average, 5% more GPUs inside the same power envelope. At the scale of a gigawatt campus, 5% more revenue-generating compute from the same grid connection is an enormous number. This is why I say the benchmark validates the strategy, not just the product.

Remember what I told you to hold on to, the behemoths are evaluating us and the evaluations are going well. This test is what going well looks like. The companies that can adopt this technology have gigawatts to consider and PhD teams who will take a claim like ours apart line by line. This test was built for that audience, widely available hardware run by a third party at the warm water conditions their own road wraps require. We didn't hand them a marketing claim. We handed them a data set. The full white paper, warm water ready is on our site, and I'd encourage you to read it the way our customers are reading it.

So back to that question of when. We can't answer it definitively, but I can give you two data points. First, the chips. IDTechEx after interviewing chip makers, cold plate suppliers and integrators across the value chain identified 1,500 to 2,000 watts per package as the point where single phase begins to struggle. The B300 shipping today is already at 1,400. Every generation on NVIDIA's public road map goes higher.

Second, the racks. Beyond heat removal at the chip, single phase stays competitive only by pushing more and more water. As industry analysts have pointed out at extreme rack densities, the pipe sizing and physical volume required becomes constraints of their own. So whether the limit arrives through the chip's heat or the racks density, the limits exist and every generation moves us closer to them. That's the moment we're preparing for, and the four milestones I laid out are how we will measure progress.

Accelsius is positioned to scale when that time comes, and our deployments to date tell the story of an evolving company. Our earliest shipments were demo systems, an in-rack CDU with a load sled or two, built to show nucleation, the boiling physics at the heart of two-phase and to build awareness. Then we matured to shipping thermal simulation racks, pack with load sleds that simulate real AI workloads, letting users test our cooling and prove the physics for themselves.

Today's deployments are different. They're built around specific servers, specific chips, specific hyperscale computing solutions. Our customers are no longer testing whether two-phase works. They're testing how well it works with their equipment.

I'm proud to be leading Accelsius through this moment. We believe that we have proven the technology. Now it is time for us to prove our reliability, then to scale it with maturity and commercial discipline. That's exactly what we're doing. I'm glad to be here, and I look forward to your questions. Bill, back to you.

Gregory Haskell

Thank you, John. Let me sum up briefly. We were candid with you today about what we're learning about Accelsius, and we told you exactly how we report progress from here. The leadership of this company at Innventure and at Accelsius is stronger than it has ever been. We are more bullish on the Accelsius opportunity than we have ever been. Operator, let's open the line for questions.

Operator

[Operator Instructions] Our first question comes from Aashi Shah with Sidoti & Co.

分析师问答

Aashi Shah

And previously, you've mentioned about $50 million of Accelsius bookings. How much of that is associated with DarkNX? And with the original DarkNX site no longer moving forward, are those bookings still intact, or do they automatically transfer to the new site, or would you need a new agreement?

Gregory Haskell

John, do you want to field that question for us?

John Hewitt

Sure. Thank you, Bill. We don't normally disclose specific dollar amounts with respect to individual customers. But what I can tell you -- and thank you for the question, by the way. What I can tell you is we expect that order to transfer to a new site. The reason that we debooked the order, as Dave mentioned earlier, is it's going to take them time to find a new site to get the appropriate allocations, et cetera. And because of that, timing is uncertain.

Aashi Shah

Right. But if you can just give us a little more context on what happened with the original site. Was it related to power availability, financing, GPU availability, or was it something else?

John Hewitt

They lost the site and our understanding of it had to do with the power envelope.

Aashi Shah

Okay. And you've identified an executed hyperscaler SoW as a key milestone. How advanced are those discussions today? And once you secure one, what does the timeline typically look like from there to commercial -- deployment and revenue?

John Hewitt

It's a great question. So we have a number of hyperscale conversations that are in various stages. And I hope to be able to report to you soon that we've crossed that particular milestone. But I'll leave it this way. There are very active conversations happening even as we speak.

With respect to the timeline, each one of the hyperscalers has their own timeline But in general, they will go from early statement of work where the technology is validated. They will then do a proof-of-concept deployment, sometimes as big as a row in an existing data center. And then from there, you're designed into the data center reference design. They're effectively their blueprint and then volume production volume deployment happens. That cycle, depending on hyperscaler can run between two and three years, and we are in various stages with many of them as we work down that path.

Aashi Shah

Right. And so -- again, we've been looking at 2027 as the meaningful commercialization year for Accelsius, but now it's too soon for that, and too early for any visibility into 2027. Is that right?

John Hewitt

I think it is too soon for visibility into 2027 from a specific revenue and bookings perspective. The -- what we are heavily focused on now is the -- those four milestones that we walked through just a few minutes ago.

Operator

Our next question comes from Nehal Chokshi with Northland.

Nehal Chokshi

Okay. So Slide 13 has the status of the third-party validation. When did this third-party system integrator start to work on this testing?

Gregory Haskell

Is this -- which chart is 13? I don't have it in front of me, Nihal.

Nehal Chokshi

It's the one that, it's the one that talks about the third-party integrator validation results of NeuCool being able to operate the system, 14 degrees lower.

Gregory Haskell

And so the question is...

Nehal Chokshi

When did this third-party integrator start the work to do this validation?

Gregory Haskell

John, do you want to field that?

John Hewitt

Yes. I don't know the specific date, but I think it was about 60 days or so ago. They completed it at the end of July, and we announced it right after that.

Nehal Chokshi

Okay. And -- in this slide, you referenced a Jacobs reference design. What is that? And is it fair to assume that Jacobs is actually the third-party integrator?

John Hewitt

No, Jacobs is a -- and there's some information on our website with respect to that. But about a year or a year and a half ago if memory serves, we did some work with them evaluating the benefits at the overall data center level, thinking about what happens to the cooling infrastructure, what happens to the broader building power envelope and then how could that get deployed. That's the Jacobs engineering study that we're referring to. That was a -- that's completely a different proof point.

Nehal Chokshi

Okay. Can you give us a sense as to who are the type of customers that this third-party integrator has?

John Hewitt

This is -- I want to be careful about the word integrator, but the -- this particular third party would serve many of the Neocloud and enterprise customers.

Nehal Chokshi

Great. Okay. Hyperscaler statement of work and chip maker engagement as key milestones. Arguably, to a certain extent, both of these are the same because hyperscalers are becoming chip makers as well. Is that not true?

John Hewitt

It's a great question. As we look at and we analyze the market, each hyperscaler has their own strategy. And depending upon which one we're talking about, some of them have used up to our estimates, 50% or so custom chips and the other half is a mix of off-the-shelf chips from one of the major suppliers. There are other hyperscalers that the mix is much, much lower than that.

So I think there are probably, I'd say, three to four major chip makers that we have to engage with. And each one of those engagements lead you to a little bit different part of the market. And then each hyperscaler has their own chip strategy, as you pointed out just a second ago. They also have their own server ODM or OEM strategy.

So we've got to knit together, and that's why these milestones are so critical, particularly in the chip maker engagement, the server ODM and OEM relationships and the hyperscaler because all three of those pieces have to almost work in concert together.

Nehal Chokshi

Got it. All right. My last question is that at least from my perspective, I would say for at least a year, I have been more in the camp of when, not if. And so I'm curious from your perspective, when did you guys go from if to when, and what was the catalyst?

John Hewitt

That's a really great question. I think we've been pretty convicted internally that it's not if, but it's when. And I think for us, that moment was relatively early on when in the first couple of years of operation when we proved that the technology worked, and we started getting really good proof points around the performance relative to single-phase cooling and air cooling. Then for us, then, the reason that we talk about now the conviction of if versus when is it's not just us that's convicted, it's the rest of the market that is in the rest of the ecosystem.

Nehal Chokshi

And so what do you think of the rest of.

Gregory Haskell

If I can.

Nehal Chokshi

Yes, please Bill.

Gregory Haskell

I was just going to say, if you kind -- if you look at the industry information out there, there's a lot of conviction now that virtually all of the major players are migrating or believe they'll have to migrate to two-phase at some point. But the other big indicator, too, is if you look at the projection for the size of the two-phase market in 2030 of $9 billion, that's materially higher than it was even a year ago.

And so one of the key messages here is that while it's true that some of the smaller players that would typically be early adopters can't access the market for various reasons, the bigger players that we're engaged with are moving in sooner. And so the inflection point of value, we think actually is coming at us more readily than we had initially anticipated.

Operator

Our next question comes from Chip Moore with ROTH Capital Partners.

Alfred Moore

I want to follow up there, I guess, on cracking a hyperscaler. It sounds like you've got discussions that have been ongoing for a while, with more than one, certainly. Can you just expand on maybe those milestones, getting integrated with a server OEM? Is this something that's running concurrently? Can the hyperscaler, as the need arises, make that happen faster? Are they the real pivot point? I think in the past, we've talked about potential for orders in, what, the seven to nine figure range. So I assume that's some of these type of entities. Just any more color. Thanks.

Gregory Haskell

Sure.

John Hewitt

It's a great question. Sorry, thank you, Bill. I assume you were going to throw that one to me, apologies. Thank you for the question. The -- there are server OEMs and server ODMs. And depending upon whether you're a Neocloud or you're a hyperscaler or your strategy is, you're going to deploy one of those two solutions.

But you put your finger on the pulse a second ago that the hyperscalers or a Neocloud and end customer demand will motivate a OEM or ODM very significantly. So we're doing two things. We are working with the server OEM and ODM for building relationships with them. We're deep in all of those conversations. And they, depending upon their own internal strategy, will either test and do some deployment and some engineering work in advance of a hyperscaler, to your example, asking for support, but they move much, much quicker when there's a demand signal as well.

Alfred Moore

Right. Okay. That switch could flip faster, but it's uncertain.

John Hewitt

100%. So I have to -- sorry for stepping in front of you, but I have -- so I have to work both of those in parallel for exactly that reason.

Alfred Moore

Yes. Great. And then maybe just my follow-up more so around cash runway, cash burn, right? It sounds like clearly, there'll be a need at some point for capital, but talk about potential to raise money at the company level and puts and takes.

Gregory Haskell

Dave, do you want to handle that? All right.

David Yablunosky

Sure. Chip, thanks for the question. In our consolidated statements there, we have $41.5 million of cash. So we have cash on the balance sheet. It's really not an issue. And we have access to multiple different avenues to raise cash. We want to do it opportunistically.

I did say in my remarks that with the announcements today, there could be a need for cash to do additional cash raises in the second half of the year. I don't want to really get into any details what those might look like. But just we have cash on the balance sheet. We have access to the standby equity purchase agreement. We're good on -- we'll be fine on cash.

Gregory Haskell

And the goal really, obviously, is to minimize any dilution we have and exposure in particular to Accelsius for our shareholders, which I know are eager to participate in that.

Alfred Moore

And AeroFlexx and Refinity, any -- probably more so AeroFlexx with some of the commercial momentum, but any update on ability there maybe to do something strategic or otherwise?

David Yablunosky

Well, we did say at the operating. Go ahead, Bill. No, go ahead.

Gregory Haskell

I was just going to say there -- first of all, both Refinity and AeroFlexx have initiatives underway where they're raising their own capital to be self-funding. So the amount of cash that we need to participate in both of those is very, very, very small for the rest of the year, and they should be self-funding thereafter. So that's good news.

I think with respect to AeroFlexx, they are turning the corner. There's a growing pipeline and the players that they're dealing with are getting bigger. It's these CPG companies, as we all know, move very, very slowly, and it's very difficult to move them at a pace that we like. Nevertheless, they have turned the corner or are turning the corner, and we have had quite a few announcements out in the marketplace. So I think -- we'll see how the rest of this calendar year goes and kind of where they end the year. But I think there are any number of avenues we can take with respect to AeroFlexx.

Operator

Our next question comes from Nehal Chokshi with Northland.

Nehal Chokshi

A follow-up question from me, and really John, for John, actually. When do you expect the Fenman architecture, the cooling architecture, to be announced?

John Hewitt

That's a really good question. And the direct answer is I don't know. There have been some -- yes, I don't know. We're monitoring that one very, very carefully.

Nehal Chokshi

Okay. And what's your understanding? And when does the design get locked down? Because I think the design gets locked down well ahead of the announcement times.

John Hewitt

Correct. I think that's true. Sorry I could not be more helpful on that one. If you find out, let me know, would you?

Operator

This concludes today's call. You may now disconnect.