Coreweave conference call: This year's capital expenditure guidance is lowered by 40%, mainly due to delivery delays and the continued strength of older generation GPU prices

On Monday after the US stock market closed, CoreWeave announced a mixed third-quarter report, with strong revenue and squeezed order performance. The third-quarter revenue reached $1.4 billion, a year-on-year increase of 134%, and the revenue backlog soared to $55.6 billion, nearly double that of the second quarter.

However, due to delivery delays, the short-term capital expenditure guidance plummeted, with expected capital expenditures in 2025 projected to be between $12 billion and $14 billion, a sharp decline of 40% from the previous expectation of $20 billion to $23 billion. Nevertheless, long-term capital expenditures remain robust, with capital expenditures in 2026 expected to "far exceed twice that of 2025."

In the subsequent conference call, CoreWeave's management detailed the reasons for the capital expenditure downgrade and the issue of overcapacity. Here are the key points from the conference call:

• Contracted power capacity increased to 2.9 gigawatts, but the delivery delay of a "power shell" from a third-party data center developer led the company to significantly lower its full-year capital expenditure guidance for 2025 from previous highs to $12 billion - $14 billion. This represents a decline of about 40% from prior market expectations. Affected customers have agreed to adjust the delivery schedule while retaining the total contract value.
• Given the strong growth in backlog orders and "unmet" demand, capital expenditures in 2026 are expected to "far exceed twice that of 2025." This means that CapEx in 2026 could exceed $24 billion, or even higher.
• The delays primarily affect the power shell, rather than the GPU or power itself. This is a temporary, systemic supply chain issue, not a demand issue. The company is in a "supply-constrained" environment, with customer demand for AI computing power far exceeding current capacity.
• The customer base is diversified, with nine of the top ten customers having signed multiple contracts. Currently, no single customer's revenue accounts for more than about 35% of total revenue, down from about 50% in the previous quarter and, more importantly, down from about 85% at the beginning of the year. Additionally, as of the end of the third quarter, over 60% of revenue came from investment-grade customers.
• The market demand for AI cloud technology remains strong across various GPU generations. For example, in the third quarter, the first contract for over 10,000 H100s is about to expire. Customers proactively renewed infrastructure contracts two quarters in advance, with prices only 5% lower than the original contract prices.
• Financing costs are significantly decreasing (for example, the interest rate for DDTL 3.0 has dropped by 900 basis points compared to before), thanks to its top-notch execution and the stable cash flow visibility brought by "no questions asked" contracts. This lays the foundation for its future continued low-cost financing.
• The company emphasizes that it provides "full-stack services from hardware to software," including its proprietary orchestration software, newly launched AI object storage (claimed to reduce customer costs by over 75%), and enhanced software toolchains through acquisitions (such as OpenPipe, Marimo).

Here is the full transcript of the conference call: Operator:

Good afternoon, everyone. Welcome to the CoreWeave Q3 2025 earnings call. Joining me today to discuss our performance are CEO Mike Intrator and CFO Nitin Agarwal. Before we begin, I would like to take this opportunity to remind everyone that our remarks today will contain forward-looking statements. Actual results may differ significantly from those anticipated in these forward-looking statements. Factors that could cause these results to differ materially are outlined in today’s earnings press release and in our upcoming 10-Q quarterly report to be filed with the U.S. Securities and Exchange Commission. Any forward-looking statements made during this call are based on assumptions as of today, and we undertake no obligation to update these statements due to new information or future events.

During this call, we will provide GAAP and certain non-GAAP financial metrics. A reconciliation of GAAP to non-GAAP metrics is included in today’s earnings press release. The earnings press release and the accompanying investor presentation are available on our website at investors.coreweave.com. A replay of this call will also be available on our investor relations website.

Now, I will turn the call over to Mike.

Michael:

Good afternoon, everyone, and thank you for joining us. CoreWeave has once again achieved outstanding quarterly performance, demonstrating strong momentum as our business continues to accelerate with the rapid adoption of artificial intelligence across industries worldwide. We are still facing a highly constrained supply situation, with demand for CoreWeave's leading AI cloud platform far exceeding our current capacity.

The strong demand from our customers clearly indicates that leading global enterprises trust CoreWeave and leverage our platform to support their most critical AI workloads. In the third quarter, we exceeded expectations with revenue reaching $1.4 billion, a year-over-year increase of 134%. In just the third quarter, we added over $25 billion in revenue backlog, bringing our total revenue backlog at the end of Q3 to over $55 billion, nearly double that of Q2 and close to four times that of the year-to-date. Additionally, CoreWeave's remaining performance obligations have reached $50 billion, growing faster than any cloud platform in history.

These results reflect the deep confidence our customers have in CoreWeave, as they rely on us as their essential cloud service provider in the AI space. Even as the industry continues to face capacity constraints, we are actively expanding. Our active power capacity increased by 120 megawatts quarter-over-quarter, reaching approximately 590 megawatts, while contracted power capacity grew by over 600 megawatts to 2.9 gigawatts. This lays a solid foundation for our future growth, with over 1 gigawatt of contracted capacity available for sale to customers, most of which is expected to come online in the next 12 to 24 months.

In the third quarter, we signed large-scale computing contracts with numerous major clients, including Meta and OpenAI. Each contract marks a significant expansion of our existing partnerships and diversifies our customer base, reducing reliance on any single client In addition, we have deepened our partnership with a leading hyperscale data center operator and have signed the sixth contract with this client to date. In fact, nine out of our top ten clients have signed multiple agreements with us, with the only exception being a new client we onboarded in the third quarter. CoreWeave is able to amplify the power of pioneers in the field of artificial intelligence innovation, accelerating their breakthroughs.

These are some of the world's most advanced AI institutions, and once they experience the performance, flexibility, and reliability of CoreWeave Cloud, they will continue to expand their collaboration with us. This is the highest recognition we can achieve. Our exceptional growth indicates that the application of AI is rapidly surpassing frontier AI laboratories and hyperscale data centers, achieving breakthroughs. The broader global demand and our recent significant successes are driving the diversification of our revenue sources.

For example, the number of clients with revenue exceeding $100 million in the past 12 months has doubled year-over-year. AI-native enterprises and companies across various industries are adopting CoreWeave to transform operations and unlock new innovation, productivity, and growth potential. Poolside, a leader in foundational model development, has chosen CoreWeave to support its mission of building general artificial intelligence and deploying enterprise-level intelligent agents; meanwhile, Periodic Labs is leveraging CoreWeave to push the boundaries of scientific discovery and computational research. At the application layer, we have added AI-native clients like Jasper, who have chosen CoreWeave as their cloud partner to transform the digital marketing landscape.

We are also seeing remarkable growth momentum among our enterprise client base. CrowdStrike has chosen CoreWeave to advance the development of cybersecurity AI agents, while Rakuten is utilizing our platform to transform its visual language models, achieving greater transparency, reproducibility, and speed in AI workloads. Additionally, we have expanded our collaborations with numerous enterprise clients, including a leading software design platform and a major telecommunications operator in the United States. Our business has now extended into the public sector, a market with unique performance and security requirements. We recently launched CoreWeave Federal, aimed at providing cloud services to U.S. government agencies and the defense industrial base.

Currently, NASA is utilizing our services at its Jet Propulsion Laboratory to advance scientific exploration. We are honored to contribute to strengthening the AI infrastructure in the United States, enabling various agencies to accelerate innovation, tackle critical missions, and safeguard national interests. These recent successes also highlight that we are ready to serve enterprises. As our client base continues to expand across various verticals and regions, we are very pleased to welcome John Jones as our first Chief Revenue Officer. John previously served as the head of global startups and venture capital at AWS John's addition will greatly enhance our team's strength, and he will play a key role in expanding global revenue scale and driving the next phase of growth. Next, when discussing the growing scale of our data centers, I would like to briefly mention our previously proposed acquisition plan for Core Scientific, which was terminated in October. While this transaction was strategically beneficial for both companies, the valuation demanded by its shareholders was not suitable for CoreWeave, especially considering that the outcome of this transaction would not adversely affect our growth targets in the coming years. We will continue to work closely with Core Scientific to jointly operate the approximately 590 megawatts of data center capacity we have leased.

We are expanding our capacity in a rigorous manner to ensure we meet the growing global demand for CoreWeave cloud services. As I mentioned earlier, this quarter we have increased our contracted power capacity to 2.9 gigawatts and enhanced the resilience and flexibility of our overall product by diversifying in terms of scale, geography, and developers. As of the end of the third quarter, no single data center provider accounted for more than 20% of our contracted power portfolio. Last quarter, we added eight new data centers in the United States, further strengthening our coverage in the domestic market. Additionally, we are also expanding in Europe, including the construction of a large data center in Scotland, a project developed in collaboration with the UK government.

As we announced during the summer, we have initiated self-built projects to further expand our business footprint and enhance operational control. Despite the strong demand for our platform, data center developers across the industry are facing unprecedented supply chain pressures. For our part, we have been affected by temporary delays due to the lagging progress of third-party data center developers. This will impact our expectations for the fourth quarter, which Nitin will elaborate on later.

Nevertheless, the customers affected by the current delays have agreed to adjust delivery schedules and extend contract expiration dates. Therefore, we have maintained the total value of the original contracts, and customers are able to retain their capacity throughout the entire term of the initial agreement, reflecting their confidence in our ability to provide the highest performance solutions in the market. We are incredibly proud of our technological achievements, and customers continue to provide feedback that CoreWeave is the best platform for running AI workloads. In the third quarter, we continued to deliver a significant number of GB200 initial scale deployments while once again leading the market with the GB300, further showcasing our outstanding operational record.

CoreWeave's leading position in the industry is unmatched. We are the only cloud service provider to submit GB300 level MLperf inference results, setting a benchmark for actual AI performance. Just last week, Semianalysis once again recognized our outstanding performance, awarding CoreWeave the highest honor—Platinum Cluster Max level for the second time, surpassing over 200 vendors, including hyperscale cloud service providers and emerging new cloud service providers No other cloud service provider has ever received this honor.

CoreWeave has successfully achieved this twice, further confirming CoreWeave's leading position in the AI cloud space. The market demand for AI cloud technology remains strong across various GPU generations. For example, in the third quarter, our first contract exceeding 10,000 H100s is about to expire. Customers proactively renewed their infrastructure contracts two quarters in advance, with prices only 5% lower than the original contract price.

This strongly indicates customer satisfaction and the long-term practicality and differentiated value of GPUs running on the CoreWeave platform. CoreWeave is the world's first hyperscale AI cloud platform, featuring computing, storage, networking, and software specifically designed for AI workloads. Our continuously expanding cloud product portfolio is built on an increasingly refined suite of software and services, helping customers build, train, and deploy new products faster. In addition to task control, we also have proprietary orchestration solutions that are crucial for autonomously operating our cutting-edge AI cloud.

We recently launched CoreWeave AI Object Storage, a fully managed storage service that eliminates any friction in moving data across different regions, cloud platforms, and storage tiers, without charging any outbound traffic or transaction fees. CoreWeave's AI Object Storage offers the highest throughput for AI workloads while saving customers over 75% in costs. We have seen significant market interest in this service and have added several first customers, including leading AI labs like Mistral. Our entire storage platform has also seen rapid adoption by customers, with annual recurring revenue (ARR) surpassing $100 million in the third quarter.

With our unique global network backbone designed specifically for AI, CoreWeave has become a central hub for customers and their critical AI workloads, ensuring consistent excellence in performance and a seamless user experience, whether using CoreWeave cloud or other service providers. We have also further expanded our observability and security suite to ensure that CoreWeave can optimally handle all customers' critical workloads, regardless of their use cases or regions. Over the past few years, we have been dedicated to supporting pioneers in AI development and improvement. Today, we are expanding our role to facilitate the practical application of AI. From the tools developers need to build AI to the solutions required for real-world AI adoption, we have consistently viewed acquisitions as a key means to accelerate this process, including the recent announcements of acquiring OpenPipe, Marimo, and Monolith.

With OpenPipe, we quickly integrated its solutions into our broader fine-tuning product suite and launched the first publicly available serverless reinforcement learning tool. Through Marimo, we are expanding CoreWeave's influence in the open-source community, starting with entry-level exploration and prototyping OpenPipe and Marimo perfectly align with the functionalities of Weights and Biases, and we are rapidly growing a developer community that relies on the CoreWeave platform. Through Monolith, we are extending these capabilities into the real world to unlock the monetization potential of AI, initially focusing on industrial use cases with established enterprise customer bases and mature workloads, including leading automotive manufacturers like Nissan and Stellantis.

By successfully launching new products and services quickly, we are expanding our target market and growing alongside our customers. We are fundamentally enhancing CoreWeave's capabilities to establish a foothold in new markets and expand our business, all to further support the rapid development of AI and help AI developers and innovators bring products to market faster and more reliably, thereby improving return on investment. Our collaborative model is maturing, exemplified by our partnership with CrowdStrike, which will unleash and accelerate partner-driven growth. Our newly launched storage products and collaboration with Vast Data are further evidence of accelerating product portfolio and partner market expansion, enabling us to compete in new markets where previous products were limited or absent. This has facilitated customer-driven platform adoption and product-driven growth, creating favorable conditions for our business development.

Finally, I want to emphasize what truly sets CoreWeave apart. We are an indispensable cloud platform in the AI space, with unparalleled technology and operational strength, as well as a rapidly growing customer base. We provide the industry's most powerful infrastructure, the fastest time to market, and cutting-edge features. Leading AI innovators around the world choose CoreWeave because we help them act faster, scale smarter, and achieve results that cannot be realized anywhere else.

Our momentum has never been stronger, and future opportunities continue to expand. With outstanding products, an exceptional team, and unparalleled execution capabilities, CoreWeave is poised to enter the next phase of growth as a full-stack AI service provider and hyperscale solutions provider. The future is driven by CoreWeave, and we are just getting started. Next, please hear from Nitin.

Nitin Agrawal:

Thank you, Mike, and good afternoon, everyone. Our impressive third-quarter performance reaffirms the strong market demand for CoreWeave and the exceptional achievements we have made in building a critical cloud platform for AI. As Mike mentioned, we continue to operate in a highly supply-constrained environment, which is expected to persist for a considerable time. We remain committed to providing the best-performing solutions in the market and are continuously increasing our investments across the entire technology stack, which has driven the growth and diversification of our customer base, achieving significant results from emerging enterprises and AI-native companies to business expansions of existing customers. Now, let’s take a look at the third-quarter performance Revenue in the third quarter reached $1.4 billion, a year-on-year increase of 134%, mainly due to strong customer demand and efficient execution. The backlog at the end of the quarter was $55.6 billion, nearly doubling in just the third quarter. Demand for the Blackwell platform and our entire GPU product portfolio remains strong. In the third quarter, we signed multiple orders for older GPUs, added new customers, and renewed existing contracts.

The broad demand for CoreWeave cloud services has allowed us to significantly reduce customer concentration. Currently, no single customer's revenue accounts for more than approximately 35% of our total revenue, down from about 50% in the previous quarter, and more importantly, down from about 85% at the beginning of the year. Additionally, as of the end of the third quarter, over 60% of our revenue came from investment-grade customers. This reflects our success in achieving platform and customer diversification goals.

Operating expenses in the third quarter were $1.3 billion, which included $144 million in stock-based compensation. We continue to increase our investment in data center and server infrastructure to address the growing backlog of orders, which has led to increased cost of revenue and technology and infrastructure spending in the third quarter. Additionally, the increase in sales and marketing expenses was due to investments in marketing and expanding our public listing organization to capture the rapid growth of AI opportunities among enterprises and AI-native companies. The increase in general and administrative expenses was driven by professional services and personnel growth.

Adjusted operating income for the third quarter was $217 million, compared to $125 million in the third quarter of 2024. The adjusted operating margin for the third quarter was 16%. The adjusted operating income exceeded expectations, mainly due to revenue growth, cost reductions from reasonable delivery schedules of third-party partner data centers, and improvements in fleet operational efficiency. The net loss for the third quarter was $110 million, compared to a net loss of $360 million in the third quarter of 2024.

Interest expense for the third quarter was $311 million, compared to $104 million in the third quarter of 2024, primarily due to increased debt to support infrastructure scale expansion, but improvements in debt interest rates partially offset this increase as we further reduce capital costs. The adjusted net loss for the third quarter was $41 million, while the third quarter of 2024 was essentially break-even; adjusted EBITDA for the third quarter was $838 million, compared to $379 million in the third quarter of 2024, more than doubling year-on-year. The adjusted EBITDA margin was 61%. In terms of capital expenditures, total capital expenditures in the third quarter were $1.9 billion, below expectations due to the delivery delays mentioned by Mike regarding third-party data center suppliers.

The significant increase in construction projects to $6.9 billion, a quarter-on-quarter increase of $2.8 billion, is a direct result of this situation. It should be noted that construction projects refer to infrastructure that has not yet been put into use and is not counted in capital expenditures until it is operational. Now, let's take a look at our balance sheet and strong liquidity position As of September 30, we have $3 billion in cash, cash equivalents, restricted cash, and marketable securities.

Rapid growth and scaled operations require strategic financing methods. CoreWeave has become a leading AI cloud platform and is at the forefront of innovative infrastructure financing needed to provide cutting-edge workloads for enterprises and AI labs. We have made significant progress in strengthening our capital structure and reducing our cost of capital. In the third quarter, we revised the DDTL 2.0 financing mechanism, increasing its remaining drawable amount by over $400 million, creating a new $3 billion tranche at an interest rate of SOFR + 425 basis points, which is significantly lower than the initial cost of this financing arrangement.

As we discussed earlier, we completed the DDTL 3.0 financing in the third quarter, priced at SOFR +400, which is a reduction of 900 basis points compared to the interest rate on the non-investment grade portion of our previous financing arrangements. Looking ahead, as funding providers increasingly recognize our top-notch execution capabilities and the stable cash flow and visibility supporting our "no questions asked" customer contracts, we expect to continue financing at lower spreads. Additionally, we issued $1.75 billion in senior notes in July at a cost that is 25 basis points lower than our initial issuance in May, further expanding our business in the high-yield bond market. Year to date, CoreWeave has successfully completed $14 billion in debt and equity transactions to support our rapidly growing backlog and efficient expansion, achieving long-term growth.

Aside from payments related to financing with OEM manufacturers and self-liquidating debt generated through committed contract payments, we have no other maturing debt before 2028. Regarding taxes, in the third quarter, we recognized a non-cash tax benefit primarily due to the impact of a substantial tax bill. Although this tax bill had a one-time impact on the third quarter due to tax payments made year-to-date, we expect changes in legislation to result in cash tax savings for CoreWeave in future periods. Regarding performance outlook, as mentioned earlier, the PowerShell delivery delays from data center providers will impact our fourth-quarter performance.

These delays are temporary, and as Mike pointed out, the affected customers have agreed to adjust their delivery schedules to ensure their capacity can be maintained within the original agreement timeframe and achieve the total value of the original agreement. In this context, we expect revenue in 2025 to be between $5.05 billion and $5.15 billion. Additionally, we expect adjusted operating income in 2025 to be between $690 million and $720 million, and we anticipate that active generation capacity will exceed 850 megawatts by the end of the year. In the fourth quarter, we will launch several of the largest projects in the company's history.

Due to the timing mismatch between when data center costs first occur and when we begin to recognize revenue, this will have a short-term impact on adjusted operating profit margins. We expect interest expenses in 2025 to be between $1.21 billion and $1.25 billion, primarily due to increased debt to support demand-driven capital expenditure growth Some will be offset by the continuously declining cost of capital. Regarding capital expenditures, we now expect capital expenditures for 2025 to be between $12 billion and $14 billion. This expenditure has been reduced from previous expectations, mainly reflected in the corresponding increase in construction and progress, as the infrastructure awaiting deployment is under construction after the delivery of powered room capacity.

Therefore, the vast majority of the remaining capital expenditures we previously expected to be recognized in the fourth quarter will now be accounted for in the first quarter. Additionally, given the significant growth in our backlog of orders and the continued strong demand for our cloud services, we expect capital expenditures in 2026 to far exceed double that of 2025. These investments in the infrastructure platform will enhance our competitive advantage and support our continued rapid growth. In summary, we achieved record performance in the third quarter and are more confident than ever in the long-term development prospects of the company's business.

This year, we have made tremendous progress. Our revenue backlog is growing at an accelerated pace, currently exceeding $55 billion; at the same time, we have diversified our customer base, strengthened and expanded our platform through strategic partnerships and acquisitions, secured new sources of funding, significantly reduced capital costs, and expanded capacity and organizational scale at an unprecedented rate. These advancements enable us to seize current opportunities and lay a solid foundation for growth in the coming years. With the widespread application of artificial intelligence across various industries and applications, as well as our prudent business decisions to expand our product portfolio and enhance market share, our target market is continuously expanding.

CoreWeave is growing at an astonishing pace, with rapid and efficient scale expansion, solidifying our leading position as a key cloud platform for artificial intelligence. We thank all investors and analysts for their support and participation. We look forward to reporting our progress to you in the coming quarters. Next is the Q&A session.

Q&A Session

Q1 Mark Murphy: Every time we discuss the artificial intelligence sector, we hear about a surge in order volume, and CoreWeave is clearly no exception. However, the bottleneck issues regarding power and manpower are becoming increasingly severe. Can you talk about the situation with third-party suppliers? Specifically, is it a shortage of power or manpower, or are there other issues related to GPUs, memory, or storage? Additionally, have you communicated with other third-party suppliers to understand their progress schedules and whether they believe they can complete or deliver on time by early next year?

Michael Intrator: Let me analyze this issue from several different angles. Okay. First, you are right. This is indeed very frustrating for our customers.

What is very frustrating for us is that the supply chain supporting global infrastructure construction faces numerous systemic challenges. The realization of artificial intelligence is also dependent on these challenges. Nevertheless, we have taken multiple measures to enhance our ability to respond to future challenges. We have invested significant time in diversifying data center suppliers. In addition, we have established a considerable portion of the company dedicated to assisting in the operation of infrastructure We have launched our own self-built projects, including in Kenilworth and Lancaster, Pennsylvania. Therefore, you can see that we are actively expanding our business, doing everything possible to minimize the losses or delays that may occur during the infrastructure delivery process, as these delays have already put tremendous pressure on the supply chain. Now, when you have diversified infrastructure delivery channels, the relative impact of each delay diminishes. You can call upon different data centers at any time based on the delivery progress.

We believe this is an important infrastructure module, just with some delays in delivery time. However, the end users (that is, the customers using this infrastructure) have extended the contract terms so that we can deliver the full contract value. Despite the delays, this indeed reflects the value that customers derive from our infrastructure. Therefore, when you communicate with CoreWeave and other companies in the industry, you will repeatedly hear similar statements.

This is the real challenge at the power supply shell level. This is not a challenge of electricity itself, right? Currently, electricity is abundant, and we believe there will be sufficient electricity in the coming years as well. But the real challenge lies in the power supply shell.

Q: So, Michael, is this completely unrelated to Core Scientific, or is it completely unrelated to what you have experienced?

Michael Intrator: So I won't specifically talk about any of our data center providers.

We are working closely with all data center suppliers, doing everything we can to ensure they ultimately deliver the infrastructure we need. We have made remarkable progress in infrastructure delivery. As you can see, our capacity continues to grow and has reached approximately 590 megawatts. Since the last conference call, our capacity has increased by another 120 megawatts.

As we continue to scale our delivery, we have seen significant success. But I think it doesn't matter which specific data center provider it is. This is a systemic issue that the entire industry must face in the foreseeable future. The key is—or rather, for me, the key is that the delays in infrastructure construction will not affect our backlog orders, nor will they affect our ability to extract the full value from the contracts that are about to be delivered.

Q2, Keith Weiss: Thank you all for answering this question, and congratulations on achieving remarkable quarterly results in completing backlog orders.

You are right. We have never seen any cloud service provider able to scale so quickly. Mike, I want to ask you a question that we are often asked, which actually relates to the risk of overcapacity, but I think it is more specific; people are concerned that the capacity contracted by AI labs may be excessive.

However, I want to ask your question: how should we view your infrastructure and the actual substitutability of the infrastructure you are building? When you build for specific customers, are those data centers applicable to any customer? Can they be used for inference and training? Or are you really customizing on demand, building for specific customers, which would limit your flexibility, and if a certain customer performs poorly, your freedom would be reduced? Michael Intrator: Yes, Keith, that's a great question. In fact, we spend a lot of time thinking about this issue as we advance relationships with all our clients. In short, the infrastructure is interchangeable.

It can be transferred from one client to another. The infrastructure is built to the most stringent specifications, so it can be used for both training and inference. We have indeed thought a lot to ensure that we maintain as much selectivity and flexibility as possible in our infrastructure construction. I want to emphasize that this flexibility and interchangeability is largely due to the excellent software suite we provide, which allows the infrastructure to be utilized so effectively, right? Just like when SemiAnalysis conducts its annual alternative review, CoreWeave is nominated time and again as the best solution for such infrastructure in the world, and there’s a reason for that.

This includes hyperscale cloud vendors, emerging cloud companies, and all other participants trying to deliver this infrastructure. We just do it exceptionally well, and we believe that by delivering infrastructure with such a powerful software suite, we are protecting a lot of value.

Q3 Kash Rangan: The backlog of orders has grown significantly. I want to make two quick points. The first point is, Mike, I think you mentioned before that you would diversify contractors in the data center space. Perhaps you could give us a concrete update on how far we are from reaching a point where any disruptions unrelated to your business should not affect your revenue outlook. How far are we from that point?

Secondly, when you observe industry developments, I mean, no one anticipated—maybe some did—the $250 billion contract between OpenAI and Microsoft, nor the $300 billion contract between OpenAI and Oracle. Suddenly, of course, CoreWeave has a unique value proposition that allows it to build GPU clusters very quickly and efficiently, almost at the speed of thought. However, in a scenario where we are discussing hundreds of millions of dollars in contracts awarded to hyperscale cloud giants, what keeps you unique three or four years from now when supply and demand balance out? When we look back at CoreWeave, what will be the shining value proposition that keeps you at the table?

Michael Intrator: Yes, thank you. I will break this question into two parts to answer, okay? The first question you asked is about diversification and when it starts and stops, leading to deviations in our quarterly performance data. I want to emphasize that as the size of individual data center projects shrinks relative to the overall scale of our data center portfolio, the impact of delays of a few weeks on overall performance will become smaller, right? For example, when you deliver 590 megawatts of power, if there is a 200 megawatt or 300 megawatt increment, that increment represents a significant portion when delivered in the next quarter, right? As we scale up and begin to build out the full 2.9 gigawatts of power we have, a delay of one or two weeks for a 100 megawatt data center will not have a substantial impact. As Nitin mentioned, we expect that the vast majority of this 2.9 gigawatts of power will be put into use within the next 12 to 24 months. This way, you can clearly see that as the scale expands, the curve will become smoother, and the relative impact of each data center will become smaller. This is the first part of scalability.

The second part is the question you have been asking since our founding: Why can CoreWeave deliver GPUs faster? Why are we able to deliver GPUs for NVIDIA to run its machine learning performance tests? Why are we able to develop software that defines the field? Each quarter, you will see us continuously expanding our leading edge, thanks to our ability to customize cloud services based on actual use cases. As you have seen in previous analyses, we are unique in this regard.

We are actively expanding our product line. We are building or acquiring additional computing capacity to further reduce the commoditization of the computing services we offer. Therefore, a company specifically established to provide such computing services will have a greater advantage in future developments.

Q4: Tyler Radke: Regarding some of the delays you mentioned this quarter, can you further explain how these delays will impact 2026? I know, Nitin, you have provided some high-level analysis on capital expenditures. But what I mean is, given the information you currently have, especially regarding the 24-month portion of RPO (Recovery Point Objective), how should we view the impact of this change on revenue? Do you think these delays will be fully resolved in the first quarter? Will we see a growth rate next year that is higher than this year? Any relevant information would be helpful to us.

Michael Intrator: I'll start and then hand it over to Nitin. I think it's important to clarify that the capacity increases we are currently seeing are related to the infrastructure of a single supplier. At the same time, we are also working with other suppliers to conduct parallel PAF (Procurement, Authorization, and Financing) for other contracts. Therefore, delivery will be somewhat affected in the short term.

Next, you will see that we have the ability to accelerate progress this year to make up for previous delays. So, most of the delay issues you see should be resolved in the first quarter of next year. Nitin?

Nitin Agrawal: Yes, Tyler, you are correct. Most of the capital expenditure delays we experienced in the fourth quarter will be completed in the first quarter. As you might expect, we will gradually increase capacity in the first quarter. As Mike mentioned earlier, since we can adjust delivery dates related to customers, ensuring that customers retain all capacity and their contract value, the total revenue related to that customer will not be affected. We will share more details about the construction and related revenue plans for 2026 in the next earnings report, but as we emphasized this quarter, given the strong customer demand (reflected in the growth of our revenue backlog and ongoing customer demand), **we expect capital expenditures in 2026 to be more than double that of 2025 **

Q5 Michael Turrin: I want to connect some comments because the growth in bookings is very significant, and there are many questions surrounding the order of bookings. It sounds like you are saying that the supply chain impacts you are seeing are more specific to certain customers. I would like to understand more clearly whether this will affect your pace of signing new customers, or if it is more related to the contracts after RAMP and a specific customer environment? Additionally, a small question, does the NVIDIA deal specifically reflect in the backlog order metrics? Given that NVIDIA's contract structure is slightly different, I would love to hear you explain in detail the impact of this deal. Thank you very much.

Michael Intrator: Of course. This does not affect our ability to attract more customers. I think it is important to understand that we are advancing the construction of infrastructure in parallel. There is one data center that had issues affecting us, but we have 32 data centers in our portfolio.

They are all making progress to varying degrees. Therefore, each project is conducted independently. As Nitin mentioned earlier, we have 2.9 gigawatts of contracted power that will come online in the next 12 to 24 months. We will work hard to find customers who can utilize this power, which will have a significant impact on our future revenue.

This data center will complete its upgrades first, and then we will continue to advance based on that. Would you like to talk about the NVIDIA deal?

Nitin Agrawal: Yes, Michael, regarding the NVIDIA deal, we are very excited about it. This contract allows us to interrupt the capacity previously reserved for NVIDIA and resell it to other customers.

Therefore, the nature of this contract allows us to serve a multitude of smaller customers profitably, such as those high-growth AI labs that tend to prefer short-term, low upfront payment arrangements, while eliminating any risk to ourselves regarding capacity utilization. Given the flexibility allowed in the contract to interrupt and resell capacity, we are very excited about it. Accounting standards require us to exclude the capacity expected to be resold to other customers from RPO. It is important to clarify that if this capacity is not resold, it will continue to belong to NVIDIA and will be recognized as revenue. Therefore, you will see this NVIDIA contract in our revenue reserves, but it largely does not appear in our RPO.

Michael Intrator: So, to add, as Nitin said, we are very excited about this because this contract will enable us to provide infrastructure for emerging companies, startups, and those companies that struggle to obtain the computing infrastructure needed to build their businesses.

Thus, this interoperability is an extremely powerful tool for enhancing the resilience of the system and creating opportunities for new companies to join the broader product portfolio of CoreWeave. We are very excited about this. We believe it is an excellent architecture. This is an agreement reached with NVIDIA, and they will bear the full economic cost as we will sell computing resources to them I want to make it clear that this indeed represents an extremely rigorous method of financing in order to reach market areas that we have not been able to touch so far, or rather, market areas that no one has been able to touch so far.

Q6 Brent Thill: I just want to confirm, your capital expenditures have been reduced by 40% this year. Let me confirm, this is just feedback from one client, right? You are not assuming that other clients are generally experiencing delays, correct?

That's right. This is related to working with a single data center provider partner and the delays associated with that. As we mentioned in our prepared remarks, most of the issues, the vast majority of the issues, will be resolved in the first quarter. By the fourth quarter, you will see significant impacts on related construction and progress.

Q7 Raimo Lenschow: When considering capital expenditures for next year, could you also talk about funding sources? Because we see many other companies adopting leasing more. You just mentioned capital expenditures, which is what you need to do. How do you view the future development path, and how can various financing methods provide the company with greater flexibility? Thank you.

Michael Intrator: So you see, we are driving innovation in both technology and financing, right? My view is that we will comprehensively examine all possible financing and expansion methods, and then choose the most cost-effective way to scale and better serve our customers. If leasing is the best option, we will choose leasing. But we have already considered many different financing structures.

Over the past three years, we have established various financing structures to secure ample funding. Looking ahead, we will continue to explore and fully utilize these structures. We will not easily sign contracts with clients before determining the funding sources. We will evaluate each transaction to ensure that all aspects, including data centers, power, GPUs, and financing, are in place to successfully provide computing services to our clients.

Q8 Amit Daryanani: I would like you to talk about how your shift from relying on third-party data center providers to building more data centers yourself will impact your capital expenditures and the time to market for power products. We are very interested in understanding what you think the best approach is, and what the capital expenditure demands will be when you rely more on self-built data centers rather than third-party data center providers.

Michael Intrator: I want to clarify. We are not saying that we will build data centers ourselves and not use third-party data center providers. What we want to express is that building our own data centers is a way to reduce the delivery risk of the entire project portfolio. Therefore, we will continue to work with partners who provide data center capacity, allowing us to use the facilities they build for us. All of this will continue to be effective. We need this capacity to continue operating and conducting business at our current speed and scale.

We view self-built projects as an integral part of the entire system. It brings us closer to the physical infrastructure and deeper into the global supply chain, allowing us to gain first-hand information We believe that in order to efficiently reduce risks in a complex supply chain environment, it is essential to possess both capabilities simultaneously.

Q9 Brad Zelnik: Currently, 2.9 gigawatts of power have been committed for delivery, with over 1 gigawatt of power yet to be contracted with customers. Meanwhile, we see other large transactions being announced throughout the industry. Given the strong market demand, how do you view the contracting progress of the remaining capacity, and how would you explain this issue to us?

Michael Intrator: The completion of other transactions strongly validates the supply-demand environment we have been describing for years, right? No single company has the capability to provide infrastructure globally to meet the growing demands of the world's largest tech companies, the largest AI labs, government agencies, enterprises, and others. Therefore, the completion of other transactions reflects the immense demand pressure that we (as well as hyperscale data centers, AI labs, and data centers) are all facing. This further reinforces and validates the theme we have been emphasizing. We believe that ultimately, the product we deliver—a full-stack solution covering everything from hardware to software—is the most valuable infrastructure that can be brought to market.

We still believe this will significantly enhance the demand for our infrastructure. As for the remaining capacity, we are seriously considering how to continue diversifying our cloud services. We are continuously monitoring various applications that will have a significant impact on how the world operates in the future. We will allocate this infrastructure to relevant parties as soon as possible to ensure they can successfully launch products and conduct business.

Nitin Agrawal: There are a few points to note, as we mentioned in our prepared remarks. Currently, no single customer's orders account for more than 35% of our total revenue, a significant decrease from 85% at the beginning of the year. Additionally, 60% of our orders come from investment-grade customers. Therefore, when managing our existing sellable capacity, we will be very cautious in considering these factors.

Q10 Brad Sills: I would like to ask a question regarding the concept of PowerShell "power shell" as a bottleneck. Does CoreWeave have any intellectual property that you use for the construction of data centers? From this delay, do you have any lessons learned that can be applied to other data center projects you are responsible for? I just want to understand what you can do to address the bottleneck issues encountered with this contract itself. Thank you.

Michael Intrator: Yes, I would say, Brad, that our lessons learned cannot be said to come solely from this delay. We have been operating in a globally constrained supply market for three years. We understand how difficult this is. With each wave of new demand, the market becomes increasingly tight.

So, when you ask what measures we will take in the future to position ourselves, what I really want to encourage you to think about is that we have established a complete organization within CoreWeave that can help us enhance capacity in self-building. This allows you to delve into the supply chain. You understand where the power is and how it is allocated You also understand the conditions required to build a power shelf, as you built it yourself.

In addition to using other third-party suppliers, it is these partnerships that will enable us to achieve as much success as possible in a challenging environment for a considerable period of time.

Thank you. The Q&A session for today ends here