CoreWeave (CRWV.US) Q3 Conference Call: CEO Responds to High Debt Concerns and Says “Unsecured Debt Market Will Not Be Used”, 75% Promises Capital Expenses to Link Investment Level Customers

The Zhitong Finance App learned that AI cloud computing service provider Coreweave (CRWV.US) held a results conference call after announcing the results for the third quarter of 2026 (as of September 30) after the market on November 10 EST. At a time when revenue surged 168% year over year and the revenue backlog was approaching $132.5 billion, management spent a lot of time responding to the market's core concerns about high debt and financing costs. Chairman and CEO Michael Intrator called this quarter a “breakout quarter” (a breakout quarter), emphasizing that the company adopted “success-oriented capital expenditure”, that more than 75% of committed capital expenditure was linked to investment-grade and corporate clients, and made it clear that “no unsecured debt market” financing will be used at current yield levels. At the same time, he made a key judgment: the computing power was “basically sold out” in 2027, and the infrastructure to be introduced to the market in 2028 has also been pre-ordered.

Revenue for the quarter was US$3.661 billion, up 168% year on year, and increased by more than US$1 billion month-on-month, higher than the market's agreed expectations of US$3.54 billion to US$3.56 billion. The adjusted operating profit of US$379 million (profit margin of 10%) greatly exceeded the company's previous guidance range of US$200 million to US$260 million; adjusted EBITDA reached US$2,256 million (profit margin 62%). Under GAAP, the company's net loss increased from US$110 million in the same period last year to US$709 million, a loss of US$1.27 per share, but it was better than the market's expected loss of US$1.46 per share. The company raised its annual revenue guidance to US$13 billion to US$13.3 billion, and the adjusted operating profit guidance to US$1,275 billion to US$1,375 million. After the financial report was announced, CRWV rose about 8% after the market, before closing down 0.51% to $48.73 in regular trading.

Focus and Response: High Debt Financing Questions vs. Strong Demand

The difference that received the most attention from the market at this meeting was that while the company handed over impressive growth rates, net losses still widened due to high financing costs — net interest expenses for the quarter nearly doubled year-on-year to US$846 million, which is already higher than the adjusted operating profit of US$379 million. Reuters also reported that corporate bond yields once rose to about 15.2%. In response to this concern, Intrator systematically responded in its statement: the company's weighted average debt cost has dropped 250 basis points year over year to 7.9%; 70% of this year's transactions have received advance payments from customers; approximately $9 billion in advance payments and deposits will be collected over the next three months; the company has already repaid more than 4 billion US dollars of debt through delayed term loans, and “will not use the unsecured debt market at current yield levels.” He stressed that the debt was paid off when the contract expired and the infrastructure was “fully owned and debt-free” by CoreWeave.

Qualitative Q3 performance: profit margin entered double digits one quarter ahead of schedule

Intrator called this quarter a “breakthrough quarter,” pointing to rapid revenue growth, higher profit margins than planned, and extremely strong customer demand. The adjusted operating margin rose from about 1% in the first quarter and 5% in the second quarter to more than 10% in the current quarter, entering the low double-digit range one quarter ahead of schedule. According to management, the third quarter results and the increase in the fourth quarter guidance were mainly driven by three factors: the rapid transformation of contracted electricity into active electricity and revenue generation capacity, the strengthening of demand and pricing environments, and the continued expansion of platform adoption across industries.

Demand and backlog: Revenue backlog of $132.5 billion, largely sold out in 2027

By the end of September, the company's revenue backlog of orders reached about US$132.5 billion, a further increase from about US$104.2 billion at the end of June. According to Intrator, computing demand is growing faster than infrastructure delivery capacity. The company “basically sold out” in 2027, and the infrastructure introduced to the market in 2028 has also been sold; by the end of 2027, qualified demand was 10 times the available capacity. Active electricity has surpassed 1.7 gigawatts, doubling since the beginning of the year, and is expected to exceed 1.85 gigawatts by the end of the year.

Financing and capital allocation: success-oriented capital expenditure, rejection of unsecured debt issuance

According to Intrator, the company uses a “success-oriented capital expenditure” model, which can adjust the pace of investment according to customer needs, regulatory influences and capital market conditions, and most of the capital is only committed once contracts and infrastructure are in place. By 2027, over 75% of committed capital expenditure was tied to investment-grade and enterprise customers, and the rest was tied to the world's leading AI laboratories and AI native companies. He expects EBITDA interest reduction (excluding advance payment proceeds) to “more than double” in 2027, thereby enhancing internal financing capabilities. The company expects the adjusted net loss to narrow significantly in 2027, and to improve the adjusted net profit in the first half of 2028.

Pricing and renewal: 30% — 50% premium for older GPUs re-signing

Intrator revealed that the pricing of Ampere and Hopper (older GPUs) increased 30% to 50% month-on-month in the third quarter, higher than the original contract price, and signed a multi-year commitment extending beyond the depreciation life. More than 130,000 GPUs will expire in 2027, and renewing the contract at the current price can generate incremental annualized revenue of nearly US$4 billion; in 2028, an additional 170,000 GPUs will expire, with cumulative incremental annualized revenue close to US$10 billion, all “without additional capital expenses.” CFO Nitin Agrawal added that this increase is not in the current revenue backlog and is an available opportunity after re-signing.

Customer Diversification and Forge: 10% Fortune 500 Customers

Management emphasizes diversification of customer concentrations. 10% of the company's Fortune 500 companies are already its customers, and 10% of the backlog orders come from financial services; more than half of the new customers added this year come from outside of AI laboratories and the technology industry. According to Intrator, CoreWeave Forge, which was launched in September, integrates AI development capabilities. 70% of the new cloud customers added in the third quarter were already Forge customers, making it a key entry point into the enterprise customer base.

Full year guidance: Revenue increased to $13 billion to $13.3 billion

The company raised its full-year guidance. It expects revenue of US$13 billion to US$13.3 billion in 2026 (previously the upper limit was US$13.2 billion), adjusted operating profit of US$1,275 billion to US$1,375 billion, an annualized operating rate of US$20-21 billion at the end of the year, and capital expenditure of US$37 billion to US$39 billion (lower limit increase).

Full Meeting Minutes

Meeting time: November 10, 2026 at 5:00 p.m. EST (after hours)

Company spokesperson:

Michael Intrator — Chairman and CEO

Nitin Agrawal — Chief Financial Officer

Participating analysts: Gabriela Borges (Goldman Sachs Research), Brad Zelnick (Deutsche Bank Research), Samik Chatterjee (J.P. Morgan Chatterjee), Karl Keirstead (UBS Research), Ben Reitzes (Melius Research), Amit Daryanani (Evercore ISI).

Good afternoon, host. Welcome to CoreWeave's third quarter 2026 earnings call. After the official statement, we will move on to the question and answer session.

Michael Intrator (CEO) Good afternoon, and thank you all for attending. CoreWeave handed over an impressive report card for the third quarter: revenue grew rapidly, profit margins rose faster than expected, customer demand continued to be strong, and the signing of new contracts was also the most profitable batch of orders in our history. Our early investment in the cloud infrastructure is being renewed at a premium, generating more cash flow than was estimated at the time of the initial investment.

The company's capital side is supported by guaranteed loans, customer prepayments, and operating cash flow. The combination of multiple benefits makes our business resilient and able to calmly cope with the current market environment. Here is the specific data: Revenue reached 3.7 billion US dollars, up 168% year over year, exceeding the upper limit of the guideline range; additional revenue exceeded 1 billion US dollars month-on-month. The adjusted operating profit of US$379 million was significantly higher than the guideline of US$200 million to US$260 million, and the operating margin moved into the low double-digit range one quarter ahead of schedule. Revenue and the volume of pending orders increased to about US$133 billion; the amount of computing power used increased to more than 1.7 GW, and the scale of computing power has doubled since the beginning of the year.

Last quarter, we mentioned that with the formation of the scale of infrastructure and the balanced layout of the business between long and short term contracts, the operating leverage effect will become apparent. This quarter's results and our guidance have confirmed this. We have also always believed that AI will penetrate all walks of life, and building this future of practitioners requires a new form of cloud infrastructure. This confidence comes true every quarter. Customers use our AI computing power platform to advance biological research, support clinical diagnosis and treatment, develop autonomous driving systems, and enable personalized learning. Our AI cloud platform helps them to implement innovative ideas and promote applications with practical value to the public.

Implementing these ideas requires a stable and cost-effective computing power base for the development and deployment of AI models. Our core mission is to build this foundation while creating a significant return on capital. In view of recent capital market fluctuations, I would like to explain how this set of business logic is self-balanced: the value we create for our customers is exchanged for long-term customer cooperation commitments; these promises are the foundation of our financing framework; and prudent investment allows us to continue to expand the computing power platform our customers rely on while continuing to increase shareholder value. Behind every decision, a core principle is followed: every choice pursues optimal economic returns — including financing methods, pricing strategies, and the pace of production expansion.

Building a large-scale AI cloud requires huge capital investment and strong enforcement and control capabilities. CoreWeave has known this since its inception. Our business model and financing strategy are designed around long-term customer contracts from the beginning to the end, and capital investment serves contract requirements. This strategy allows us to evaluate each additional capital and select financing channels that maximize returns based on the market environment. By the end of the third quarter, the weighted average cost of company-wide debt was 7.9%, down 250 basis points year over year.

Looking ahead, at current rates of return, we will not be using unsecured debt financing, and there is no need for that. Continuing our consistent approach to supporting the remaining capital expenses required for orders to be executed, most of which will be completed through deferred withdrawal term loans tied to specific customer contracts. By 2027, more than 75% of capital expenditure has been committed to investment-grade enterprises and large enterprise customers, with the remaining portion going to the world's leading AI laboratories and native AI companies. The pricing of this type of financing is anchored in the underlying customer contract, and our proven delivery capacity can cover 70% — 100% of the corresponding capital expenditure. Since the IPO, interest rates for this type of financing have generally increased by 225 to 550 basis points compared to SOFR; interest rates for high-quality investment-grade customers are close to the lower end of the range, and AI lab customers are biased towards the upper end of the range.

Based on current interest rates, the actual financing cost corresponding to the upper limit of the range is less than 10%. The pace of withdrawal of this type of loan matches the pace of equipment procurement and delivery; in addition to covering interest and principal repayment at maturity, the cash flow generated by the contract can also bring additional cash flow to CoreWeave. According to the plan, we have repaid more than $4 billion of debt through this model, including two of the most expensive first deferred withdrawal term loans, and have paid off nearly half of it.

For financing projects that service long-term contracts, when the customer contract expires, the corresponding debt has been paid off, and the infrastructure is entirely owned by CoreWeave. This unindebted stock computing power has itself become an important source of funding. Every year, a number of existing contracts expire, and no additional investment is required to continuously generate revenue. Equipment procurement and data center construction have all been completed, and as the scale expands, our capital intensity will continue to decline. Advance payments are another important source of funding. We adopt a market-based pricing approach for advance payments, balancing pricing discounts and financing costs.

In the past, we used the unsecured debt market to finance; in the current environment, prepayments are more attractive. This year, we have collected advance payments for 70% of customer transactions, which can cover the entire contract amount at once. In the next three months alone, we will collect approximately $9 billion in advance payments and security deposits from signed customer contracts, of which more than $3 billion will come from renewals of existing cloud infrastructure, with a significant premium over the original pricing.

Furthermore, profit before interest, tax, depreciation and amortization (EBITDA less interest) excluding advance payment income is expected to double in 2027, improving the ability of internal funds to cover capital requirements. OEM financing can also support additional computing power deployment, which is particularly suitable for capacity investment in short-term demand scenarios. Taken together, we expect that the above funding channels will cover the vast majority of capital expenditure requirements in 2027, and only a small amount of additional financing will be required at the parent company level.

Strict control of capital expenditure is as important as the ability to obtain capital. In addition to investments corresponding to locked contracts, our “success-oriented capital expenditure” model can flexibly adjust the pace of production expansion according to customer computing power requirements and supply chain constraints (including supervision and capital market changes). This is an important risk hedging tool. The vast majority of capital investment will not commence until the contract is completed and the infrastructure conditions are in place. Nitin will explain this later.

Next, I linked the financing model with the profit logic reflected in performance. As more in-contract computing power is put into operation, revenue continues to grow based on capital investment that has already been implemented, and the share of new deployments in the overall business continues to decline — this is what we called operating leverage last quarter. The adjusted operating margin was about 1% in the first quarter, 5% in the second quarter, and surpassed 10% in the third quarter. The profit level of newly signed businesses also continues to improve.

In July, we raised the price of each hashrate SKU by about 25%, and then increased it by 10% in the third quarter. Prices continued to rise in the fourth quarter, and the increase completely covered rising costs. At the same time, we are also allocating more computing power to attractively priced short-term orders. Short-term computing power can be priced at $40 million per megawatt per megawatt; while new long-term contracts for 4-5 years are priced in the range of $20 million to $25 million per megawatt.

Long-term contracts provide performance visibility for our production expansion plans; short-term contracts allow us to seize high pricing windows and earn premium returns. We weigh trade-offs on a case-by-case basis to evaluate the full financial account of an investment: it can generate revenue, operating costs, and financing costs. Strong demand has given us the option to create sustainable value. The existing first-generation computing power clusters are also in high demand: in the third quarter, the Ampere and Hopper architecture's computing power increased 30% to 50% month-on-month, higher than the original contract price. Different loads have different trade-offs in computing power performance and cost. The older generation of computing power is still very economical, especially after it is connected to a stable and large-scale cloud platform. We are signing multi-year renewals, and the contract period even exceeds the GPU depreciation period; in the later stages of the contract, pricing and operating costs will remain stable, but depreciation will gradually return to zero, and profit margins will rise further.

More than 130,000 GPU contracts will expire in 2027, and around 170,000 GPU contracts expire in 2028. Currently, a large number of customers are already queuing up to compete for this batch of computing power. According to current pricing estimates, the 2027 renewal will bring in nearly $4 billion in annualized incremental revenue without additional capital expenses; when combined with computing power due in 2028, the annualized incremental revenue is close to $10 billion. Furthermore, if the construction of data centers across the industry slows down, the supply of existing computing power will become scarce, and the high-cost GPU cycle is expected to prolong.

In this scenario, the first-mover advantage is essential: our stock of computing power will complete debt settlement and contract expiration earlier than most of our peers in the market. Regardless of the market trend, we are in an advantageous position: the expansion of the industry is accelerating, we are growing at the same time; supply implementation is delayed, and the value of our debt-free stock computing power has further increased. Of course, our business plan does not rely on this scenario; the planning is based on locked contract requirements, successfully directed capital expenditure, and new orders that are being placed.

Last quarter, I mentioned that model training, reasoning, evaluation, and iteration are forming a closed loop of AI development in a continuous cycle: online applications will feed back on their strengths and weaknesses. Developers will iterate on the model, test new versions based on feedback, and then re-launch the optimized application. This process requires not only professional software to support each round of experiments, but also highly stable cloud infrastructure throughout the application life cycle.

This is CoreWeave's core strength. We have created a unified platform where hundreds of thousands of GPUs work together to form a high-performance full-stack cloud, integrating networking, cooling, orchestration, and observability capabilities. This is why our customers are handing over the core load to us. In September, we launched CoreWeave Forge to integrate the entire suite of AI development capabilities into a unified environment. We open products and services developed jointly with leading AI laboratories to enterprise customers, so that companies can independently develop AI based on their own expertise and data. Of the new cloud customers added in the third quarter, 70% are already Forge users themselves. Wave is a typical example: they first used our development tools for model testing, and now they are using the CoreWeave cloud to train models of the autonomous driving world.

The Chan Zuckerberg Initiative uses our computing power to carry out computational biology research; Hipponomic AI trains and deploys intelligent devices to help patients communicate and guide patients, help research diseases, and help patients obtain diagnosis and treatment information and support.

In the field of engineering, we cooperate with organizations such as Nissan and the Aston Martin F1 Team to shorten the simulation test cycle and reduce the amount of data required to achieve target accuracy. The long-term vision of physical AI is to help engineers develop better systems and implement practical capabilities faster. MasterClass uses Forge to evaluate the effectiveness of teaching agent conversations and optimize the course experience running on the CoreWeave cloud.

The ultimate goal is to personalize learning and support users when they need it. These customers are innovating in science, clinical, engineering, and education, combining human exploration with AI. CoreWeave provides them with a platform to support large-scale, high-speed R&D. This is why our AI cloud platform is as important as the underlying computing power: whether a good idea can be turned into a usable application depends largely on whether developers can conduct experiments, evaluate results, and stably launch services at low cost. More than half of the new customers this year came from outside of AI labs and the tech industry; 10% of Fortune 500 companies have already become CoreWeave customers. The number of customers invested more than $100 million in a single commitment was more than 2.5 times higher than the previous year.

Customers continue to explore AI implementation scenarios, and demand for platforms is becoming more and more abundant, which broadens our business space and deepens customer stickiness. By the end of 2027, we have received 10 times the amount of eligible computing power currently available; the scale of demand in 2028 is already several times more computing power available at that time. Before it was officially released to the outside world, we had already used 1.7 GW of computing power. We maintain our expectations of using over 1.85GW of computing power by the end of the year.

The site under construction has obtained a full license, and the construction progress is stable. We judge that there is no substantial risk in the 2027 delivery target. This year, international expansion became a strategic priority. Of the 4.6 GW of contracted computing power, the scale of computing power outside the US has exceeded 1.3 GW, covering Canada, Europe, and Asia. The landing locations are all regions where customers request it. In the third quarter, we entered new markets in Indonesia and India, while expanding our presence in Denmark and Spain. Our global AI backbone network connects regional computing power, and can obtain consistent performance, software capabilities and stability no matter where the customer's computing power is deployed.

Our core goal is to deliver the computing power that our customers have signed up for, and get significant rewards for every deployment. Before handing it over to Nitin, I'll go back to the original point: CoreWeave's goal is to build a long-term sustainable business. We have achieved large-scale, profitable growth. The financing strategy will continue to bind funds to customer commitment contracts, rely on stable returns, and obtain funds from multiple channels. This mechanism allows us to flexibly adjust the pace of production expansion according to customer needs, the regulatory environment, and the pace of the capital market.

Our computing power delivery schedule is ahead of schedule, and the computing power value continues to rise. While renewing early stock computing power premiums, we continue to maintain the trust of the community and customers where the project is located. Our cloud platform takes into account price, performance, and reliability, and supports the software and development tools necessary for enterprises to implement AI strategies. The company has reached an inflection point, and the sustainability of the business continues to increase.

Fundamentals continue to improve, and we are confident that we will serve our customers, shareholders and partner communities for a long time. Let's leave it to Nitin.

Q&A session

The moderator question-and-answer session will now begin. Each questioner is limited to one main question+one follow-up question. The first question came from Goldman Sachs Gabriela Borges.

Gabriela, your line is connected, please talk.

Gabriela Borges (Goldman Sachs Analyst) Good afternoon Mike, thanks for bringing so many details. Let me first ask a macro question about the industry: the supply and demand data you gave is very impressive. However, investors generally have a concern: in the next 1-2 years, the industry will have a large supply of additional computing power, and it is difficult to accurately predict the pace of growth on the demand side. The market is always debating whether the balance between supply and demand will arrive next year or not until five years from now.

My question is: if the industry enters a phase of balance between supply and demand or even oversupply, how will this CoreWeave business model be affected? Thank you.

Michael Intrator (CEO) thanks for the questions and for attending this quarterly meeting. This quarter's results have been impressive, and we're excited to share them with you. First, let's talk about your concerns: There has always been a mainstream opinion in the market over the past few years that computing power infrastructure will be implemented faster than demand. However, we have observed that the growth rate of demand for computing power, including other computing power vendors and computing power buyers, is higher than the rate at which infrastructure is being put into production and delivery. The increase in profit margins on every new contract also confirms this.

Yes, it is true that the market will launch a large amount of computing power one after another. However, the rate at which AI technology penetrates the entire economy exceeds the rate of expansion of computing power supply, cluster building, and model training capabilities. Our 2027 computing power is almost sold out, and demand is extremely strong.

2028 will also be in high demand. Not that all of the computing power has been signed up for 2028, but all of the computing power resources we put on the market have been subscribed. Massive demand awaits the implementation of computing power supply.

Gabriela Borges (Goldman Sachs Analyst) got it. To follow up on a question: I'm concerned about the updated pricing of $20 million-$25 million per megawatt. For contracts that have already been signed and the revenue level per megawatt is locked in, how can you manage the risk of income-side pricing lock-up and mismatch between the cost side? From the signing of a customer contract to the official launch of computing power, costs may fluctuate or even rise during this period.

How do you manage the contradictions caused by this time difference? Thanks for updating the information.

The host's next question came from Deutsche Bank Brad Zelnick. Brad, the line is connected, please talk.

Brad Zelnick (Deutsche Bank Analyst) Thanks. The disclosure was very thorough, and a lot of information dispelled our concerns. I want to know: Will recent industry credit market concerns change your signing strategy for adding new computing power after 2027? This includes aspects such as the computing power delivery cycle, customer structure, contract duration, and financing matching pace.

Michael Intrator (CEO) thanked for the question. Just to use this to clarify: We haven't launched the 2028 computing power on a large scale; we hope to wait for the electricity supply and computing power planning environment to become more clear. We have multiple adjustment grippers to ensure that supply rhythms are aligned with demand signals. You can see this in our schedule of computing power to be implemented and in the demand guidelines.

We do not lock in long-term computing power too early, but try to align the production capacity roadmap with the customer's development plans. One point I've mentioned over and over again is maintaining a broad and diverse customer pool, which allows us to get very accurate demand signals: who needs computing power, when and how much, covering multiple industries, including AI laboratories, robotics circuits, and S&P 500 companies. This information is an important basis for us to develop forward computing power contracts.

Brad Zelnick (Deutsche Bank Analyst) Thank you. Mike, one more follow-up question. Many emerging AI cloud vendors are now experiencing growth pains. Will this bring opportunities to CoreWeave? Like customer migration, industry consolidation, or pricing authority level?

Michael Intrator (CEO) That's a great question, too. We have completed large-scale implementation and are in an advantageous position in a volatile market environment. At the end of the day, an AI cloud consists of two layers: a software layer for managing infrastructure, and the computing power hardware itself. We believe the software layer we've built is at the top of the industry.

Hardware computing power itself is a game of scale, and players must reach sufficient size to compete effectively. We understood this from the beginning. The business model is a long-term contract with a diverse buyer base. Now we continue to expand our scale. On the one hand, software products are competitive, and on the other hand, hardware can be supplied on a large scale. The combination of the two allows us to select customers, select suitable partners, and accompany them to start AI construction or continue to expand.

Brad Zelnick (Deutsche Bank Analyst) Thank you so much.

The host's next question came from J.P. Morgan Samik Chatterjee. Samik, please speak up.

Samik Chatterjee (J.P. Morgan Analyst) Thanks. Mike and Nitin showed strong profit margins in the current quarter, while leading to continued improvement in profit margins in the fourth quarter. I want to know what role short-term contracts play in this round of rising profit margins? As you mentioned before, the computing power of early production during the quarter will boost profits.

How should we split the third quarter and fourth quarter guidelines. Do profit increases come from execution and implementation, or short-term contract contributions, respectively? I have another follow-up question.

Michael Intrator (CEO) We will continue to make internal trade-offs: long-term contracts bring business stability, short-term contracts increase profit margins, and value-added services such as escrow inference also contribute to growth. All three businesses are growing rapidly, supporting the company's expansion. We can take on more short-term orders and escrow reasoning requirements. The foundation is a computing power chassis built on long-term contracts. Therefore, in the future, we will continue to optimize the distribution of computing power and allocate scarce computing power resources between these three opportunities.

Samik Chatterjee (J.P. Morgan Analyst) Follow-up question: After the election, the market is very concerned about the changes in the US regulatory environment and regulatory pressure. Can you share your judgments, particularly the long-term plan for 8GW 2030 and the significance of international expansion in the current macro environment? Thank you.

Michael Intrator (CEO) concluded first: We don't expect major changes in the regulatory landscape to have a real impact. The licensing and construction progress of each data center corresponding to the pending order is relatively stable, and we are very confident in the pipeline.

Regulatory changes essentially put forward requirements for construction methods. We have always insisted on responsible development, communicated and collaborated with the project site community for a long time, and maintained a good neighborhood relationship with the territory. This is our consistent practice. Therefore, we are confident that we will continue to expand in the new policy environment.

As you said, CoreWeave prioritizes the global layout, and has added 1.3 GW of computing power outside the US. We have landed in Indonesia, India, Spain, Denmark and Canada. On the one hand, these regions meet the diverse computing power layout needs of customers. On the other hand, we can build new infrastructure and spread layout risks in communities that welcome the implementation of computing power projects.

The host's next question came from UBS Karl Keirstead. Karl, please speak up.

Karl Keirstead (UBS analyst) Thanks. Mike, you mentioned the goal of customer diversity. Recently, many cutting-edge AI laboratories have heard that they plan to build their own computing power and reduce leasing. We want to know how CoreWeave will respond? And what percentage of the additional computing power do you plan to distribute to non-lab customers, that is, enterprise customers? Can you explain this customer restructuring plan, and at the same time introduce the customer composition of orders in line with the increase in orders to be executed in the third quarter?

Michael Intrator (CEO) No matter which part of the business we discuss, our core idea is to reduce risks such as customer concentration and data center concentration. One piece of data is compelling: 10% of Fortune 500 companies are already CoreWeave customers, and the financial industry accounts for 10% of our pending orders. Diversification has actually been implemented. More than half of the new customers added this year are not part of the AI circuit.

From the beginning, we knew that to do cloud business, we must serve long-tail enterprise customers. They need a computing power base and development software like Forge to build their own AI tools.

We will maintain long-term cooperation with AI laboratories, large public cloud vendors, and social media companies; at the same time, the company will unify the direction and vigorously expand other types of customers.

Nitin Karl, let me add: Forge is a very good vehicle for us to enter enterprise customers. We have seen significant growth, and the results have been good. 70% of the new cloud customers added in the third quarter were originally Forge users. Forge is an entry point for enterprise customers to start deploying computing power, helping us open up the enterprise market space.

Karl Keirstead (UBS analyst) thanked them both. That's all I asked, and left it to my other colleagues to ask.

The host's next question comes from Melius Research Ben Reitzes. Ben, please speak up.

Ben Reitzes (Melius Research) Mike and team, thanks for sharing. You mentioned recent communication with the newly elected governor of Texas. Can you explain the communication situation and the extent of our exposure in Texas? You say delivery delays are not expected; what supports this judgment? One more short follow-up question.

Michael Intrator (CEO) On the Texas side, the vast majority of our computing power has either been put into operation or fully licensed and under construction. This is also the basis for our judgment that data centers that handle pending orders will not be impacted.

At the same time, we believe it is important to establish communication channels with new governors and regulators, just as we have maintained communication with successive local decision makers in the past. The core of my letter is to convey the message: we maintain a non-partisan position and hope to continue to advance the construction of computing power; we are a compliant and reliable operator, willing to cooperate with the territorial government and community, and look forward to continued implementation.

Ben Reitzes (Melius Research) Thanks Mike. Follow up: I'd like to know about the arrival of the Vera Rubin video card and the progress of the climb, as well as the state of the Vera CPU rack. How important are these products to recent increases in performance guidance and next year's recurring revenue?

Michael Intrator (CEO) CoreWeave is proud to be the world's first cloud vendor to launch and deliver Vera Rubin to Cognition, and the world's first customer to put Vera Rubin into formal production. This is significant for Cognition, and for the industry as a whole. This is a new, extremely powerful architecture that will be one of the core hardware for the next stage of AI development.

We are also really looking forward to deploying Vera CPUs independently. It has huge application space on the AI infrastructure circuit, and related demand is growing rapidly. This is our key layout direction for the next few years, and it is also the focus of the entire industry. We are honored to be the first to complete online production.

Nitin Ben, let me answer your second question about this quarter's results and the drivers behind the fourth quarter upward guidance. Three core points: First, the computing power within the contract continues to be rapidly transformed into in-use and revenue-generating computing power; second, the demand and pricing environment continues to strengthen, and the return on investment continues to improve; third, the platform penetration rate continues to increase, and customers across industries use more of our products and services. Together, these three points will boost this quarter's results, and will also continue to support the fourth quarter guidance after the increase.

Ben Reitzes (Melius Research) thanks Nitin.

The host's next question comes from Evercore ISI Amit Daryanani. Amit, please speak up.

Amit Daryanani (Evercore ISI) Thanks for sharing the details. I have two quick questions. First, you mentioned that under the same hardware utilization rate, short-term contracts are annualized at $40 million per megawatt, and long-term contracts are $20-25 million. After deducting financing costs, what is the difference in return between the two? Also, Mike, can you tell me what percentage of computing power are you willing to spend on short-term contracts?

Nitin Mike mentioned earlier that our business consists of three pillars: long-term contracts bring stable cash flow, high-value-added value-added services such as escrow inference, and short-term contracts. We assess every opportunity internally and calculate returns, computing power financing costs, and supply and demand environments to allocate computing power.

Judging from the payback cycle, the payback period for short-term contracts is faster, about 16-18 months; according to current long-term contract pricing, the hardware capital expenditure payback period for long-term contracts is in the 1-2 year range.

Michael Intrator (CEO) Another thing worth watching: Early contracts expire one after another, with 130,000 GPUs in 2027 and 170,000 GPUs in 2028. This batch of GPUs is ideal for short-term orders. Moreover, the loan for this infrastructure was fully paid off at the end of the original contract. This is the design of a deferred withdrawal loan. After full amortization, it can significantly increase cash flow.

Amit Daryanani (Evercore ISI) got it. Following this thread, you mentioned that 2027 renewals would bring in $4 billion in incremental revenue and close to $10 billion in 2028. Is this figure gross revenue or an incremental premium? Have these contracts been renewed and finalized, or are they based on current price expectations? These numbers seem very impressive, and this kind of computing power would also have a high EBITDA profit margin.

This part of Nitin is not currently among the orders currently locked for execution. Orders pending execution are computing power that has now been contracted; this figure assumes that after the stock computing power expires, the revenue that can be generated by re-signing the contract at the current market price, and there is no need for additional capital expenses.

Michael Intrator (CEO) added clarification: Of the $9 billion advance payment and security deposit we mentioned, more than $3 billion comes from stock computing power renewals that have already fully paid off the debt. This capital inflow does not require any debt principal or interest. The corresponding infrastructure has completed the full cycle of the original contract and deferred withdrawal loan. That's why it's so valuable.

Amit Daryanani (Evercore ISI) was clear, thank you so much.

This concludes the moderator's Q&A session. Next, I'll return to Michael Intrator for the closing remarks.

Michael Intrator (CEO) Before I close, I'd like to thank the team, customers, partners, and everyone for trusting CoreWeave and working with us to advance this cause. All results are inseparable from you. I'm proud of the progress we've made and our next goal—we're building the cloud platform that's essential for the AI era. Thank you all for attending the conference today, thank you for your support, and we will keep up with you next quarter.

This concludes the host's conference call. Thank you all for attending the conference. You can disconnect.