CoreWeave conference call: CEO bluntly states that AI computing power demand is "ruthless and never-ending," holding $66.8 billion in orders, with future profits expected to stabilize at 25%

AI infrastructure provider CoreWeave's first-quarter guidance fell short of expectations, causing its stock price to plummet over 9% in after-hours trading, although the full-year outlook remains supportive. On February 26th, after U.S. stock market hours, CoreWeave reported a fourth-quarter adjusted loss per share of 56 cents, exceeding the market's general expectation of 50 cents. The net loss widened significantly from $51 million in the same period last year to $452 million. Despite the first-quarter revenue guidance range of $1.9 billion to $2 billion, which is far below analysts' forecast of $2.29 billion, market sentiment was pressured. CFO Nitin Agrawal explained that as the company expands on a large scale, data center leasing costs, electricity expenses, and depreciation will start before revenue recognition. Agrawal further added that once business and growth normalize, the company is confident in achieving a long-term profit margin of 25% to 30%. Company CEO Mike Intrator stated that the demand for AI computing power is "relentless and never-ending," leading to an average contract length with customers extending to 5 years, and the explosive demand for inference has caused A100 chip prices to rise instead of fall. By 2026, the company's capital expenditures are expected to double to over $30 billion, with projected revenues of $12 billion to $13 billion, aiming for an annualized revenue of over $30 billion by 2027. Additionally, for the fiscal year 2025, CoreWeave's total revenue is expected to reach $5.1 billion, a year-on-year increase of 168%. CEO Mike Intrator candidly stated in the meeting: 2025 will be a pivotal year for CoreWeave. This will be the fastest cloud service to reach $5 billion in annual revenue in history. Mike Intrator expressed optimism for the future: Our entire capacity for 2026 is essentially sold out, and we continue to sign new contracts that will be allocated once the new capacity comes online in 2027. On Thursday, after U.S. stock market hours, CoreWeave's stock price fell nearly 9%. Nevertheless, CoreWeave's cumulative increase this year still exceeds 36%, far surpassing the approximately 1% increase of the S&P 500 index during the same period. Key points from the conference call: Short-term profit margins under pressure, long-term outlook positive: Due to the early deployment of infrastructure and depreciation, Q1 2026 will be the low point for profit margins, with adjusted operating profit expectations of only $0 to $40 million, followed by a gradual increase each quarter. Management reiterated that as the business matures, the long-term operating profit margin target remains at 25% to 30%. Impressive order backlog: By the end of the year, the contracted order reserve reached an astonishing $66.8 billion, an increase of $11.2 billion quarter-on-quarter and over $50 billion year-on-year Diversification of Customer Structure and Extension of Contract Period: Demand has spread from leading cloud vendors (Hyperscalers) to AI-native companies and traditional enterprises. The average contract duration for customers has extended from about 4 years to approximately 5 years. Strong Demand for Older Generation GPUs: There has been a surge in demand for the previous generation GPU architecture (mainly used for inference). During Q4, the average price of the H100 remained above 90% of the beginning of the year, while the average price of the A100 even saw an increase in 2025. Significant Effectiveness of Product Cross-Selling: Customers are no longer just renting GPUs; 80% of customers with annual consumption exceeding $1 million have adopted CoreWeave's storage products; bundled sales with software like Weights & Biases contributed hundreds of millions of dollars in contract value in the second half of the year. Doubling of Power and Data Center Scale: By the end of 2025, there will be 43 active data centers with an active power capacity of 850 megawatts (an increase of 260 megawatts just in Q4). It is expected that by the end of 2026, the active power capacity will double to 1.7 gigawatts (GW). Becoming NVIDIA's "Exemplar Cloud": Q4 became the first platform to achieve NVIDIA's GB200 "Exemplar Cloud" status. It is expected to launch NVIDIA's next-generation Rubin GPU to the market in the second half of 2026. Aggressive Performance Guidance for 2026: Revenue is expected to reach $12 billion to $13 billion in 2026 (with a median year-on-year growth of about 140%). By the end of 2026, annualized revenue is expected to reach $17 billion to $19 billion; by the end of 2027, it will exceed $30 billion. Profit Margins Under Short-Term Pressure, but Future "Imagination Space" Exists CoreWeave's core business model involves deploying NVIDIA GPUs across dozens of data centers and renting computing power to enterprise customers and AI model developers for training and running large language models. The adjusted operating profit margin for Q4 was 6%, facing short-term profit margin pressure due to massive capital expenditures. CEO Intrator candidly stated: Our margins reflect the cost of building tomorrow's revenues. CFO Nitin Agrawal explained that as the company expands on a large scale, data center leasing costs, electricity expenses, and depreciation will start before revenue recognition. This will be particularly prominent in 2026, as the company will deploy approximately double the new capacity compared to 2025. He emphasized that the first quarter is the low point for profit margins for the year. Agrawal then added that once business and growth normalize, the company is confident in achieving a long-term profit margin of 25% to 30%. **In terms of future growth imagination space, CoreWeave is moving away from being purely labeled as a "GPU rental company" and evolving towards higher-margin software and ecosystems ** Nitin Agrawal disclosed that among CoreWeave's cloud customers with annual fees exceeding $1 million, approximately 80% have adopted at least one of the company's storage products. The company's storage business annual recurring revenue has surpassed $100 million in the third quarter. Meanwhile, through cross-selling with Weights & Biases, the company added several hundred million dollars in total cloud contract value in the second half of the year. On the software side, CoreWeave has developed a proprietary cloud technology stack and reference architecture, including Sunk and Mission Control. In January of this year, NVIDIA announced its intention to test and validate CoreWeave's platform for inclusion in NVIDIA's reference architecture for use by other cloud, enterprise, and sovereign customers. The CEO summarized: Distributing our proprietary cloud stack widely is expected to become a continuously growing source of high-profit revenue over time... This represents tangible long-term upside potential and is not reflected in the 2026 guidance we provided today. "Relentless" Demand and "Massive" Orders Regarding the market's most concerned issue of demand sustainability, CoreWeave stated that as of December 31, 2025, the company's contract revenue backlog has soared to $66.8 billion, an increase of $11.2 billion quarter-over-quarter and over $50 billion year-over-year. Company CEO Mike Intrator stated: The demand environment remains relentless, driving rapid adoption from hyperscale computing companies, AI-native enterprises, and traditional enterprise customers. Notably, there has been a change in customer behavior structure. CFO Nitin Agrawal pointed out that customers are entrusting their foundational AI workloads to CoreWeave, with longer commitments. Agrawal stated: The weighted average contract length has increased from about 4 years to about 5 years. Additionally, the company's customer structure continues to diversify, with the number of customers committing to spend at least $1 million on the CoreWeave cloud platform increasing by nearly 150% throughout 2025. New customers include AI-native and enterprise clients such as Cognition, Cursor, Mercado Libre, MidJourney, and Runway, while also expanding partnerships with its two existing hyperscale cloud customers. At the same time, regarding the market's previous concerns about "old chip capacity oversupply," CoreWeave's management revealed that due to the rapid spread of AI inference use cases, demand for the previous generation of GPU architectures is significantly increasing. Michael Intrator pointed out that relevant contracts have been signed in advance of capacity availability, attempting to dispel market concerns about weak demand for the older generation of GPUs. In terms of pricing, Intrator stated that GPU pricing overall remains stable for the entire year of 2025. The average pricing of H100 in the fourth quarter has decreased by no more than 10% compared to the beginning of the year, while A100 pricing is on the rise Clearing the Delay Shadow of Data Centers, Capital Expenditure to Double in 2026 At the execution level, CoreWeave has cleared the delay shadow of data centers discussed last quarter. The CEO revealed that the company has delivered over 50,000 Grace Blackwell to affected customers and has become the first cloud platform to achieve NVIDIA's GB200 exemplar cloud status. To absorb the massive backlog of orders, CoreWeave expects capital expenditure in 2026 to be between $30 billion and $35 billion, more than double the approximately $15 billion capital expenditure scale in 2025, and almost all directly linked to signed customer contracts. The CFO emphasized: (In 2026) nearly all capital expenditure is directly linked to customer contracts we have already signed, and we expect to double our active power capacity to over 1.7 gigawatts by the end of this year. For the 2026 performance guidance, the company expects full-year revenue to be between $12 billion and $13 billion (with a median year-on-year growth of approximately 140%). At the same time, the company expects annualized operating revenue to reach between $17 billion and $19 billion by 2026, and this figure will grow to over $30 billion by 2027. Transcript of CoreWeave's Q4 2025 Earnings Call, fully translated as follows (with AI tool assistance): Company Participants Michael Intrator, Co-founder, President, CEO, and Chairman Nitin Agrawal, Chief Financial Officer Unnamed Speaker Other Participants Amit Daryanani, Analyst, Evercore ISI Institutional Equity Ben Reitzes, Analyst, Melius Research Brad Zelnick, Analyst, Deutsche Bank Brent Thill, Analyst, Jefferies Gabriela Borges, Analyst, Goldman Sachs Josh Baer, Analyst, Morgan Stanley Mark Murphy, Analyst, JPMorgan Michael Turrin, Analyst, Wells Fargo Securities Meeting Statement Operator Hello everyone, thank you for your patience. I am Tiffany, and I will be the operator for today's call. Now, I would like to welcome everyone to the Q4 2025 and full fiscal year earnings call. To prevent background noise, all lines have been muted. After the speakers' remarks, there will be a Q&A session. If you would like to ask a question during the Q&A session, please press the star key followed by the number one on your telephone keypad. Now, I would like to turn the call over to CoreWeave. Please go ahead. Unnamed Speaker Thank you. Good afternoon, and welcome to CoreWeave's Q4 2025 and full fiscal year earnings call. Joining us today to discuss our performance are CEO Mike Intrator and CFO Nitin Agrawal 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 such results to differ materially are listed in today’s earnings press release and in our annual report on Form 10-K that we will file with the U.S. Securities and Exchange Commission (SEC). Any forward-looking statements made during this conference call are based on assumptions as of today, and we undertake no obligation to update these statements due to new information or future events. In this conference call, we will present financial metrics in accordance with Generally Accepted Accounting Principles (GAAP) as well as certain non-GAAP financial metrics. The reconciliation of GAAP to Non-GAAP metrics is included in today’s earnings press release. The earnings press release and the accompanying investor presentation can be found on our investor website. A recording of this conference call will also be available on our investor relations website. Now, I would like to turn the meeting over to Mike. Michael Intrator Good afternoon, everyone, and thank you for joining us. 2025 is a pivotal year for CoreWeave. We generated over $5.1 billion in revenue, a year-over-year increase of 168%. Our contract backlog grew to $66.8 billion, an increase of $11.2 billion quarter-over-quarter and over $50 billion year-over-year. As of December 31, our active power capacity reached over 850 megawatts, and the number of new reserved instance customers added in the fourth quarter was approximately double that of any quarter in our history. While achieving these results, we quickly addressed the data center delays discussed last quarter, delivering the affected deployments ahead of expectations from the third quarter earnings call. We are the fastest cloud platform in history to reach $5 billion in annual revenue. We are still in the early stages of the most transformative infrastructure expansion in history, at the forefront of building and operating some of the largest dedicated AI clusters for the world’s most demanding computing tasks. Stripping away the complexity, there are four fundamentals that define our current position: First, the demand environment remains strong, driving rapid adoption among increasingly diverse hyperscale cloud companies, AI-native enterprises, and enterprise customers. Second, by expanding our platform from pure GPU to more areas and recently enhancing our partnership with NVIDIA, we have unlocked new profit growth channels to monetize CoreWeave Cloud. Third, thanks to unparalleled execution and a strategic approach to capacity expansion, our data center scale is growing rapidly. Fourth, a disciplined financial model designed to invest ahead of revenue to meet contractual demands, supported by a backlog of $66.8 billion, provides high visibility for sustainable cash flow, attractive returns, and the ability to lower capital costs. We will discuss these aspects one by one today Regarding demand, as AI workloads become more complex, the pace of model scaling accelerates, and adoption rates continue to surge, the signals we are seeing among hyperscale computing companies, AI-native enterprises, and enterprise customers are continuously strengthening. This widespread demand has translated into deepening engineering collaborations with our largest customers and substantial progress in customer diversification. We are supporting the next generation of AI pioneers. As I mentioned in the fourth quarter, the number of new reserved instance customers we added was about twice that of any previous quarter in history, including AI-native and enterprise-level companies like Cognition, Cursor, Mercado Libre, MidJourney, and Runway. We have also expanded our relationships with some of our largest partners, including our existing two hyperscale cloud customers. The demand from these different types of customers is accelerating, while pricing remains stable for the full year of 2025. These trends are continuing as we enter 2026. Overall, this year, the number of customers committed to spending at least $1 million on CoreWeave Cloud has grown by nearly 150%. These are by no means one-time infrastructure deployments; they represent complex multi-product opportunities, are early chapters of lasting platform partnerships, and are growth engines that continue to generate compounding effects as AI becomes increasingly embedded in the operational ways of these companies. We are also seeing a significant increase in demand for the previous generation of GPU architectures, which remain in tight supply. The average pricing of H100 in the fourth quarter fluctuated within 10% compared to the beginning of the year, while the average pricing of A100 has risen in 2025. From our customers, we understand that the demand for such infrastructure is primarily for inference use cases, which are rapidly surging. We are signing these infrastructures into new reserved instance contracts before they are available, which strongly alleviates concerns about the demand for older generation SKUs. As our new capacity for 2026 is essentially fully allocated, we are continuing to scale up to meet the enormous demand from existing and potential customers in the near and long term. These trends reinforce our belief in the durability of demand and the lifecycle of these technologies running on CoreWeave Cloud. Given the insatiable demand environment and the ongoing signals we see from customers, we are accelerating our roadmap with the goal of adding over 5 gigawatts (GW) of data center capacity beyond what we have already contracted by 2030. Next, let's talk about new monetization avenues for CoreWeave Cloud. Our platform is continuously evolving, and we are unlocking pathways for profit growth through new products and services, as well as offering our proprietary cloud stack to a broader NVIDIA ecosystem outside of data centers. AI-native enterprises and enterprise customers are no longer just consuming our core GPU infrastructure; they are participating in our unified platform at a very high rate in terms of CPU, storage, software, and development tools As their AI workloads mature, the opportunities to add additional value for these customers are immense and represent significant upside potential over time. This has already manifested on our platform. For example, among CoreWeave Cloud customers who pay at least $1 million annually, about 80% have adopted one or more of our storage products. Additionally, we have seen strong cross-selling momentum with Weights & Biases, adding hundreds of millions of dollars in total cloud contract value (TCV) from Weights & Biases customers in the second half of the year. We have also accelerated the development of proprietary cloud stacks, reference architectures, and related software solutions, including SunK and Mission Control. These software solutions coordinate every layer of our dedicated cloud and are increasingly becoming a defining factor in the CoreWeave customer experience. In January of this year, we announced that NVIDIA intends to test and validate our platform, including our software and reference architectures, with the aim of incorporating these products into NVIDIA's reference architectures for cloud, enterprise, and sovereign customers. Currently, we have seen select customers authorize SunK as their default research cluster management platform across multi-cloud architectures. We expect the widespread distribution of proprietary cloud stacks to become a growing source of high-margin revenue over time. The ability to monetize our platform both within our own data centers and now outside of them through third-party licensing agreements greatly expands our potential market. This represents a tangible long-term upside potential, but it has not yet been reflected in the 2026 performance guidance we provided today. In terms of execution, our active power exceeded 850 megawatts by the end of the year, with approximately 260 megawatts added in the fourth quarter across 43 active data centers, up from only 32 at the beginning of the year. We signed additional power contracts close to 2 gigawatts in 2025, ending the year with over 3.1 gigawatts of contracted capacity, and we expect nearly all of this capacity to come online by the end of 2027. The capacity we have signed but not yet activated represents potential revenue potential, which we will monetize once constructed and delivered. We will continue to strategically procure land, power, and data center shell infrastructure. Particularly looking beyond 2026, we see strong opportunities in the current market for CoreWeave to grow contracted power capacity and will selectively leverage our expanded collaboration with NVIDIA to further accelerate our roadmap to better meet demand. Operating at such scale and speed is inherently complex. When disruptions occur, we take decisive action through disciplined coordination across teams and partners. We have rapidly cleared the delays discussed in the third-quarter earnings call. Overall, we have now delivered over 50,000 Grace Blackwell to affected customers, deploying servers in a rolling manner and completing deliveries within weeks of obtaining the necessary data center infrastructure. We are delivering at this remarkable pace across multiple different sites while handing over tens of thousands of GPUs to different customers We believe that CoreWeave is the only cloud platform capable of moving at this speed while delivering industry-leading performance and reliability, which has earned the trust of our customers and allowed us to capture more wallet share. The feedback and results we receive from our closest customers are exhilarating. Grace Blackwell's large-scale operations on our cloud are revolutionary. In the fourth quarter, we became the first cloud platform to achieve NVIDIA's GB200 Exemplar Cloud status, while still being the only AI cloud rated platinum by SemiAnalysis. We expect to continue to be at the forefront of execution and innovation in the AI cloud stack, and we will be one of the first companies to bring NVIDIA's new Rubin GPU platform to market in the second half of 2026, while expanding our product portfolio to include NVIDIA's Vera CPU and BlueField storage. The integration of these new technologies into our proprietary cloud platform will help provide customers with powerful new capabilities, including agentic workflows. Our speed of execution also explains why capital expenditures (CapEx) in the fourth quarter exceeded guidance. Our team was able to bring infrastructure online ahead of schedule, which we view as an acceleration of high-quality revenue capacity for 2026. To objectively illustrate our current scale, according to third-party estimates, CoreWeave's scale now exceeds the combined total of the 15 largest competitors in North America and Europe. Bringing over 260 megawatts online in a single quarter requires extremely perfect coordination, managing hardware, networking, storage, and dedicated software across over 100,000 GPUs and their millions of interconnect system components. This is one of the most complex operational projects in the tech industry. This is also where we excel more than anyone else. Finally, turning to our financials and business model. In 2026, we expect capital expenditures to reach at least $30 billion, more than double the capital expenditures of 2025. I want to clearly articulate this number. It reflects the extremely large contract demand that lies ahead of us. Our revenue backlog has grown to $66.8 billion, with the vast majority of planned capital deployment aimed at directly supporting these long-term contract demands, where we can foresee long-term profit margins through sustained cash flow. As Nitin will mention when providing guidance for 2026 and commenting on targets for 2027 and beyond, the dividends of these investments will compound. These backlog orders will continue to be primarily funded through the asset-level delayed draw term loans we bring to market. We hope to continue lowering our weighted average cost of capital in the process while unlocking broader industry participation in our financing tools. The demand for participation in our capital markets journey is increasingly diverse and growing. We have built excellent financing tools for the business, enabling us to scale at the speed of AI while remaining on track to achieve our investment-grade goals Before I hand over the time to Nitin, let me leave a few final thoughts. We have a backlog of $66.8 billion in contract revenue, and we expect every new capacity contract to start generating revenue by the end of 2026. We are delivering cloud infrastructure and converting it into revenue today. We are building and operating some of the largest dedicated AI clusters for the most demanding computing tasks in the world, with speed and quality that are second to none. The demand driving this construction is strong, diverse, and continues to grow as customers engage with our increasingly expansive product suite. The backlog of signed orders provides us and you with clear visibility into the trajectory of the future. We are expanding our collaboration with NVIDIA, making our platform a natural home for cloud, enterprise, and sovereign customers seeking the best AI infrastructure performance within our own data centers (now even extending externally). Our ability to foresee the future of AI innovation and meet its demands is unparalleled. We will continue to invest and expand this incredible market advantage with discipline and contracted cash flow, strategically deploying capital to increase capacity, deepen our product suite, and develop the AI cloud that customers need. Our top priority remains clear: to provide the highest performing, most reliable, and most efficient AI platform for customers globally. The market is accelerating, and we are ready to be both beneficiaries and drivers of the AI revolution. Next, I will hand it over to Nitin to speak. Nitin Agrawal Thank you, Mike, and good afternoon, everyone. Throughout 2025, since our IPO, we have executed the strategy we set for this year with discipline. We have significantly diversified our customer base, more than doubling our contracts and active power capacity, while strengthening our balance sheet by unlocking new sources of funding and lowering our weighted average cost of capital. We have also expanded our product portfolio through internal R&D and external acquisitions, successfully completing four strategic acquisitions to advance our roadmap. The pace of the current market and the scale of demand for CoreWeave Cloud create significant opportunities, and in 2026, we will make prudent investments to meet this demand and accelerate plans to further solidify our leadership position. Now let's take a look at the fourth quarter performance. Driven by strong customer demand and excellent execution, fourth-quarter revenue was $1.6 billion, a year-over-year increase of 110%. Full-year revenue was approximately $5.1 billion, a year-over-year increase of 168%. Demand for cloud continues to strengthen, with revenue backlog reaching $66.8 billion at the end of the quarter, growing more than fourfold this year alone. As Mike pointed out, we have made significant progress in diversifying our ultra-large customers, AI-native companies, and enterprise clients, which was a goal we set during our IPO last year. Additionally, customers are submitting their foundational AI workloads to us for longer periods, resulting in an increase in average weighted contract length from about four years to approximately five years We have a revenue backlog of $66.8 billion, with every new capacity contract expected to start generating revenue by the end of this year. We are fulfilling our commitments today. These commitments were made for current and past generations of GPUs and are part of a broader customer roadmap. Positive discussions have already begun regarding future SKUs (product models). Operating expenses for the fourth quarter were $1.7 billion, which included $157 million in stock-based compensation. We were able to deploy data center and server infrastructure faster than expected, with more capacity coming online this quarter than in any previous quarter in our history. This drove an increase in our revenue costs as well as corresponding increases in technology and infrastructure spending. Additionally, the increase in sales and marketing expenses was due to our investment in expanding our go-to-market organization to capture the rapid growth of AI opportunities. The increase in general and administrative (G&A) expenses was primarily due to professional service fees related to mergers and acquisitions and financing activities, public company costs, and additional staffing to support our growth. Adjusted EBITDA for the fourth quarter was $898 million, while for the fourth quarter of 2024 it was $486 million, nearly doubling year-over-year. Our adjusted EBITDA margin was 57%. Adjusted operating income for the fourth quarter was $88 million, while for the fourth quarter of 2024 it was $121 million. Our adjusted operating margin for the fourth quarter was 6%. Adjusted operating income was below expectations due to the earlier-than-expected deployment of infrastructure. The net loss for the fourth quarter was $452 million, while for the fourth quarter of 2024 it was $51 million. Interest expense for the fourth quarter was $388 million, while for the fourth quarter of 2024 it was $149 million, primarily due to increased debt to support our infrastructure expansion. Adjusted net loss for the fourth quarter was $284 million, while for the fourth quarter of 2024 it was $36 million. Turning to capital expenditures. Total capital expenditures for the fourth quarter were $8.92 billion, with total capital expenditures for the year at $14.999 billion, exceeding expectations as our team was able to bring infrastructure into service earlier than anticipated. As we forecasted last quarter, construction in progress saw significant growth, advancing to $9.4 billion in the fourth quarter, an increase of $2.5 billion quarter-over-quarter, reflecting the scale of infrastructure we expect to deliver in the near term. Just a reminder, construction in progress represents infrastructure that is not yet in use and has not yet been depreciated. When these assets come online, they will generate incremental revenue and corresponding depreciation. Our financing structure is designed to match this deployment model. The vast majority of our term debt is structured as delayed draw facilities, meaning capital is drawn only when data centers are operational. Despite the ongoing complexities of the global supply chain amid a persistent imbalance of supply and demand, we have continuously overcome these challenges through operational discipline and strategic sourcing. Our track record of scaling infrastructure gives us confidence in our ability to adapt and continue accelerating capacity deployment in 2026 and beyond Now let's take a look at our balance sheet and strong liquidity position. As of December 31, we have $4.2 billion in cash, cash equivalents, restricted cash, and marketable securities. We continue to make significant progress in optimizing our capital structure and reducing our weighted average cost of capital. In the fourth quarter, we raised approximately $2.6 billion through the initial issuance of convertible preferred notes, with investor demand significantly exceeding the issuance size, leading to an expansion of the issuance. We also expanded our revolving credit facility to $2.5 billion this quarter to manage liquidity and support various growth plans. For the full year of 2025, we secured over $18 billion in debt and equity financing, collaborating with more than 200 investment partners and financial institutions, reflecting the depth and breadth of capital committed to our growth. As Mike discussed, in January of this year, we announced an expansion of our commercial relationship with NVIDIA, accompanied by a $2 billion investment to support our platform, team, and shared vision for large-scale AI infrastructure. The efforts we made in the fourth quarter and over the past year to optimize our financial structure and reduce our weighted average cost of capital are reflected in a 300 basis point decrease in our weighted average interest rate this year, and a total decrease of nearly 600 basis points since 2023. To give everyone a more intuitive understanding, this 300 basis point improvement means an annual interest savings of nearly $700 million based on our fourth quarter debt balance. Looking ahead, we expect to continue reducing our weighted average cost of capital as capital providers and rating agencies increasingly recognize our top-tier execution capabilities, as well as the durability and visibility of cash flows supporting our "no questions asked" customer contracts. Aside from self-amortizing contract-supported debt and OEM supplier financing, we have no debt maturities until 2029. Regarding taxes, we recorded a non-cash tax benefit in the fourth quarter, primarily influenced by a significant piece of legislation (One Big Beautiful Bill Act). Our tax rate may experience significant fluctuations in the future due to similar factors. Demand across all customer categories continues to surge and diversify. We are accelerating investments to capture contracted demand, with these long-term commitments providing us with clear cash flow visibility to achieve top-tier cloud margins as deployed capacity matures. We expect capital expenditures for 2026 to be between $30 billion and $35 billion, more than double the investments in 2025. Almost all of this spending is tied to customer contracts we have already signed, and we intend to bring these contracts online this year, as we expect our active power capacity to double by the end of this year, exceeding 1.7 gigawatts. As I described earlier, when new capacity comes online, data center leasing costs (including power and depreciation expenses) begin to incur, while customer revenue gradually rises in the following months. In 2026, this effect will be amplified due to the scale of our deployment plans. We will bring online approximately twice the capacity compared to 2025, meaning the corresponding increase in depreciation will precede the related revenue recognition. For the full year of 2026, we expect revenue to be between $12 billion and $13 billion, with the median representing approximately 140% year-over-year growth We expect adjusted revenue to be between $900 million and $1.1 billion. We anticipate that the profit margin will rise from the low single digits in the first quarter, continue to expand in the second and third quarters, and return to the low double digits in the fourth quarter as the deployed capacity matures and revenue begins to scale relative to the existing cost base. Our profit margin trajectory for 2026 is the result of our prudent investments to meet the endless demand on the platform. As our business and growth normalize, we are confident in achieving a long-term profit margin of 25% to 30%. Our mature revenue contracts can generate a contribution profit margin in the mid-20s, coupled with the ongoing growth of high-margin products and services that we continue to unlock, which gives us strong confidence in achieving this goal. Our 2026 guidance excludes any meaningful revenue or profit benefits that may arise from further licensing our proprietary cloud stack to other NVIDIA clouds, enterprises, or sovereign customers. We expect this business to launch in 2026 and become more substantial in the coming years. This represents a tangible long-term potential upside. Specifically for the first quarter, we expect revenue to be between $1.9 billion and $2 billion. We anticipate adjusted revenue for the first quarter to be between $0 and $40 million. The first quarter represents the trough of our annual profit margin trajectory, as we expect infrastructure capital expenditures to reach $6 billion to $7 billion for the quarter, continuing to significantly increase new capacity based on the approximately 260 megawatts added in the fourth quarter. Our interest expense for the first quarter is expected to be between $510 million and $590 million. The long-term nature of our contracted revenue backlog provides us with visibility far beyond 2026. As we continue to grow rapidly, we expect to reach an annualized revenue run rate of $17 billion to $19 billion by the end of 2026, and by the end of 2027, this annualized revenue run rate is expected to grow to over $30 billion. We are not building this growth trajectory speculatively. The demand from signed customers, deep strategic partnerships, proactive infrastructure deployment, industry-leading capabilities, and a comprehensive strategy for the capital markets give us the confidence to present these numbers. We delivered strong fourth-quarter and full-year results, marking the end of a transformative 2025. Our contracted revenue backlog grew to $66.8 billion, while significantly diversifying our customer base; we raised over $18 billion in debt and equity capital at gradually decreasing costs and strengthened our platform through new products, new services, and strategic acquisitions. With 850 megawatts of active power distributed across 43 data centers, we enter 2026 with our heads held high, expecting to exceed 1.7 gigawatts by the end of the year, with each new capacity contract expected to start generating revenue this year. Our 2026 investment plan is fully supported by contracted demand. As we noted, our guidance excludes the potential upside from licensing the proprietary cloud stack, and we expect this business to begin contributing revenue in 2026 and scale in the years to come Thank you everyone, we look forward to answering your questions. Q&A Session Operator (Operator's Note) Please limit your questions to one during today's conference call. We will pause briefly to organize the Q&A list. Your first question comes from Keith Weiss at Morgan Stanley. Please go ahead. Josh Baer Hello, I am Josh Baer, asking on behalf of Keith. Thank you for taking the question. Congratulations on a strong quarterly performance. Your capital expenditures are significantly ahead, and it's also great to hear that the delivery delays are being resolved faster than expected. I tried to align these situations with the active power metrics and found that active power is more in line with expectations, while the revenue guidance is within the range, below the typical level of "significantly exceeding expectations." So I hope you can break down these dynamics: if you are progressing faster, why isn't this reflected in active power and revenue? Maybe it is. Thank you. Nitin Agrawal Thank you for your question, Josh. When we deploy capacity, most of it comes online close to the end of the quarter, so you will start to see it monetized in 2026. We continue to rapidly expand capacity. As we mentioned in our prepared remarks, we will continue to deploy this capacity throughout 2026, including the first quarter, which is the impact you are seeing. As for the first quarter numbers, we are actually providing guidance for 2026 and the first quarter for the first time now. Josh Baer Okay. Thank you. It's great to see this list of enterprise customers. I hope you can elaborate on what the contracts and deals with these types of enterprise customers look like. We have a clear understanding of what the large contracts with hyperscale cloud companies look like, but could you give a rough overview of the scale, duration, prepayments, and pricing related to these enterprise customers? Thank you. Nitin Agrawal We do not discuss individual contracts. But what you see is that the products we deliver (i.e., cutting-edge computing infrastructure delivered through CoreWeave Cloud) face very high competition in this environment. These contracts are largely similar in duration to those with hyperscale customers. Additionally, we work closely with each enterprise customer to design appropriate business models and structures for their end customers. Josh Baer Awesome, thank you. Operator Your next question comes from Amit Daryanani at Evercore. Please go ahead. Amit Daryanani Yes, thank you for taking my question. My question is actually about the cost of financing, especially considering that we have over $30 billion in capital expenditures this year. I would like to know, as you continue to expand capacity, what level do you estimate your blended cost of capital is currently at? How has it evolved over the past 12 months? When you negotiate with these data center operators, how do they assess your creditworthiness? Is it really tied to your customer contracts and customer identities, or does it depend on other factors? Additionally, in terms of financing, does NVIDIA's credit support or guarantor framework help achieve a significant reduction in your borrowing costs by 2026? Michael Intrator Thank you for your question. We have made incredible progress at the company level. As our enterprise business matures, and as we accumulate a richer track record in operating these infrastructures, collaborating with customers, and delivering infrastructure, you will see our capital costs decrease by 300 basis points over the past 12 months. Over the past two years, it has decreased by 600 basis points. We expect this trend to continue. This is mainly because our business is performing increasingly well within these delayed draw term loan (DTL) structures. When it comes to data centers, we added nearly 2 gigawatts of infrastructure in 2025. To give you a sense of scale, our portfolio had only 1.3 gigawatts at the end of 2024. Therefore, you will see a substantial improvement in our ability to acquire, build, and drive data center contracts. We are excited about this. It is an important stepping stone for us to continue driving business growth. Similarly, being able to sign such large-scale data center capacity contracts reflects the maturity of our business and the expanding reputation and scale of the company. As for the thoughts of data center operators, may I share my perspective? I believe data center operators are very willing to work with us. They want a diversified tenant mix within their data centers. They want exposure to a company like ours that is purely focused on driving AI infrastructure deployment. Therefore, they see us as a pure pathway to truly capture the scaling benefits of AI, and they are eager for this exposure. Regarding the impact of our relationship with NVIDIA on accelerating data center acquisition, I think you should view it this way: Clearly, working with investment-grade counterparties as purchasers will impact the capital costs or related costs of data centers. It is also evident that our selective (but certainly not exclusive) collaboration with NVIDIA in building out our data center footprint will have a positive impact on our data center land-related costs. Amit Daryanani Perfect. Thank you very much for sharing these insights. Operator Your next question comes from Mark Murphy of JP Morgan. Please go ahead. Mark Murphy Thank you very much, and congratulations on the extremely strong order performance. Mike, some AI models have shown tremendous progress—over the past few months, you have demonstrated quite astonishing leaps. The model code of Claude has been making headlines frequently. However, I believe we have yet to see those models that have been deeply trained on these massive Blackwell, GB200, or NVL data centers. And that is precisely the area where CoreWeave is pioneering and excelling. I am curious about what you are hearing in the market, particularly regarding the progress of these Blackwell-based models If we eventually see GPT-6 or any other model in the next three to six months, do you think they will represent a huge leap in capability, or will they appear more like a stable evolution of the system? Michael Intrator The Blackwell system is amazing. It represents the next leap function in computing power, enabling data scientists and the companies driving the models to build and scale infrastructure in ways they could never do before. My expectation is, and all the signs from the model companies indicate, that the pace of performance improvement for these models is just beginning. Of course, we are still in the early stages of the Grace deployment. There are not many clusters at this scale currently in existence, and we have already completed delivery. As these clusters come online in our portfolio and globally, I have no doubt that you will see these new technologies bring about a leap in performance. We are very excited about this. And I believe customers are extremely excited as well, because they understand what they can do once they have such high-performance technology—whether for training or for subsequent available inference. Mark Murphy Mike, thank you for your answer. Just to extend the question, since you mentioned the term "inference" earlier. How do you weigh the pros and cons when focusing on NVIDIA's reference architecture? Clearly, it is very powerful for large-scale training tasks and some subsequent work. But on the other hand, is there consideration for collaborating with other legitimate claims that can offer better inference cost-performance? And with NVIDIA acquiring Groq, I wonder if you think this somewhat reshapes the competitive landscape, to the extent that sticking with NVIDIA's reference architecture might also dominate in the inference space. I just want to understand how you foresee future developments. Michael Intrator Whenever there is a conference call and someone asks this question, I always talk about how we conduct our business. That is, we are customer-oriented. Customers come to us and tell us the infrastructure they need to drive their business. I want to be clear that when they say "infrastructure," they are not talking about training infrastructure or inference infrastructure, but "AI infrastructure." The reason they specifically come to us is that we can provide high-performance NVIDIA technology. They know we excel in this area. That’s why they come to us. Are they looking for other technologies from other providers? That’s to be expected. But what I know is that the demand signal for the product we deliver still far exceeds our supply capacity. Over the past three years, it has overwhelmed our and the entire market's ability to deliver infrastructure. Therefore, we will continue to focus on driving our exceptionally outstanding solutions and meeting this overwhelming demand. Mark Murphy Thank you very much. Operator Your next question comes from Brent Thill of Jefferies. Please go ahead Nitin Agrawal Brent, we can't hear you. Brent Thill Hey, good afternoon. Nitin, I have a brief question about the performance guidance. When I look at the revenue guidance, your revenue is in line with expectations, operating profit is slightly lower, while your capital expenditures are much higher. I think this indicates that even within the guidance you provided, some metrics may have significant fluctuations. I'm curious how you view this guidance moving forward? Have some variables you observed in the fourth quarter been excluded? Or has your guidance changed in some way? I know it's hard to estimate, but some numbers have indeed exceeded the range you initially provided us. Nitin Agrawal Thank you for the question, Brent. From a guidance perspective, let me break it down through a few variables. The capital expenditure numbers we discussed are essentially to serve our signed customer backlog, which we disclosed this quarter to be $66.8 billion. It is this that is driving our investment in the platform. When you consider the revenue uptick, we also mentioned that nearly all of our backlog contracts will start generating revenue this year. That’s the upward slope you’re seeing. We delivered 850 megawatts of power capacity in this fiscal year, and we expect power capacity to reach 1.7 gigawatts for the full year of 2026. In terms of margins, when all these factors reflect on margins, our margin progress is actually a result of our prudent investments to meet the endless demand on the platform. We talked about the fourth quarter. Just in the fourth quarter, we brought in new capacity equivalent to 30% of the active power base. This naturally causes some recent margin compression, as the rising costs of capacity precede the full maturation and recognition of revenue. As I mentioned in my opening remarks, the first quarter represents the low point we see in margins. Since then, as we expand into deployed capacity, our margins will expand quarter by quarter, recovering to low double digits by the fourth quarter. We also discussed some aspects regarding the long-term trajectory of this business. Our strategy and management philosophy remain to continue investing based on customer demand for backlog contracts. In the long run, when business growth returns to normal, how does it manifest in our business? We are confident in achieving a profit margin of 25% to 30%. The factors supporting this are that if you look at our mature, fully deployed contracts, the contribution profit margin from that portfolio is in the mid-20s. We are also continuously enhancing products and services in our portfolio that improve margins. For example, we announced in the third quarter that our annual recurring revenue (ARR) from storage business on our platform has surpassed $100 million. Today, we also discussed that the attach rate for storage among large customer groups has now reached 80%. Although not included in our 2026 guidance, we see tangible long-term upside from further monetizing the proprietary cloud stack, licensing it to other NVIDIA clouds, enterprises, and sovereign customers. Therefore, what you see in 2026 reflects the accelerated growth driven by our existing backlog orders, which continue to increase with customer demand Michael Intrator I would like to add a few points here. Our profit margins reflect the "cost of building tomorrow's revenue." As Nitin mentioned, the fundamental profit margin of a stable facility is in the mid-20s. This will become evident as we continue to build infrastructure and bring more online. The fluctuations you see are due to the relative proportion of the infrastructure we are building compared to the installed capacity. We brought 260 megawatts of power online in the fourth quarter, which accounted for a third of our installed capacity. Therefore, the fluctuations you see will be larger. As we continue to scale the company, the fluctuations you see will decrease as the proportion of new data center capacity relative to the total becomes smaller. This is what acceleration looks like. To serve our customers, we made the decision to invest early to lock in tomorrow's revenue, which is a fundamental strategic decision. And we execute this very responsibly because we are not "blindly building," nor are we "waiting for customers after building." We are relying on the backlog of contracts we have already secured to make these investments. Brent Thill That's great. Thank you for the information. Operator Your next question comes from Gabriela Borges of Goldman Sachs. Please go ahead. Gabriela Borges Hey, good afternoon. Thank you for taking my question. Nitin, I want to ask about the diversity of your customers on the platform. I'm curious if you could share your observations regarding the differences in unit economics or the attractiveness of using the CoreWeave platform among different types of customers on the platform? This is a relatively broad question. I know your pricing model is based on "dollars per GPU" and the duration of committed contracts, but I'm curious if you could share your observations on customer behavior across different cohorts? Thank you. Nitin Agrawal This largely depends on several variables we discussed earlier regarding how contracts are structured. These include the length of the term, the amount of upfront payment, the generation of GPUs, and the demand for capacity at that moment, all of which play a decisive role. Fundamentally, as Mike described, aside from the scale considerations we take into account when facing large versus small customers, the contract structure across our entire customer base looks relatively similar. However, across the entire customer configuration, as market dynamics change, we strive to generate similar economic benefits from the infrastructure we create. Mike mentioned that for Hopper architecture chips, we continue to see incremental demand, and the average selling prices (ASPs) for renewing these Hopper contracts fluctuate around 10% of the initial contracted price; whereas for A100, we actually see an increase in ASP when signing new contracts. These dynamics reflect broader market dynamics, but the economic benefits across our entire customer base appear relatively similar Michael Intrator The only thing I want to add is that I think one very exciting thing is that many enterprise customers, many smaller AI-native customers are coming to our infrastructure, and they have the capability to use H100 and A100. Their ability to leverage this infrastructure to build products and provide inference services is a very good testament to the persistence of the search for computing power. These are all brand new use cases, and people are doing things we have never seen before. This is really exciting. Gabriela Borges Thank you for your thoughts. Operator Your next question comes from Ben Reitzes of Melius Research. Please go ahead. Ben Reitzes Hey, guys, thank you very much. I want to ask Brent the other side of that question. One situation in the market right now is that there is a lot of capital expenditure, but there may not be enough margins or cash flow in the near term. I think what he really wants to express is that you talk about long-term margins being between 20% to 25% or 25% to 30%, and your confidence in that. So how confident are you in the growth rate of capital expenditures? We understand that you are spending now to gain future revenue, but will this growth rate be more tempered than we might expect as margins rise? Do you need to continue investing at this pace of capital expenditure, which is above Wall Street expectations, to achieve those numbers? Any additional details on the capital expenditure side to balance Brent's earlier question about margins would be very helpful for us to understand this trajectory. Thank you very much. Michael Intrator Thank you, Ben, that's a good question. I think it's important to break this question down into at least two parts. The first part is that our business is built on a model based on success, where customers come to us and sign long-term contracts for the infrastructure they need to build their businesses. So when a hyperscale cloud provider comes to us to purchase five years of infrastructure, they are buying five years of infrastructure from us at a fixed price. This is a very stable way to build revenue and margins. This is also where Nitin's confidence in margins comes from. Because we not only know how much we will earn now, but we also know how much we will earn as these contracts are fulfilled over the five-year period under the "take-or-pay" commitment. Therefore, this is a very stable way to build a business and also a very stable way for us to obtain financing in the capital markets to support the build-out. This has also been a fundamental component of how CoreWeave has built its business from the very beginning. The second part contained in the question, I think, is: how confident are we in the ongoing demand for computing infrastructure? This is a great opportunity for me to talk about what we are seeing in the demand stack because this demand stack is actually very fascinating We are not only seeing demand spreading throughout the economy, initially mainly limited to hyperscale cloud services and foundational models, but now you see it explosively entering the enterprise sector, entering the software field, and seeing all these new participants starting to come in and acquire the infrastructure they need. You also see an extremely fascinating aspect, which is that demand is gradually expanding from the pure GPU level that initially opened our business to levels such as storage, CPU, etc. This is actually a reflection of the customer portfolio using our infrastructure truly extending to the application layer. Therefore, in terms of demand, we are grasping the pulse of the market, the depth of which very few companies in the world can achieve. We are receiving information from the entire economic system, and everyone is trying to acquire the necessary infrastructure. So we are very confident about the contract situation based on the customer portfolio and are very excited about the different types of customer portfolios. There are a few data points I would like to add that we discussed in our remarks. Regarding capital expenditures for 2026, the vast majority is tied to the signed customer contracts we intend to go live with this year, and we expect the power capacity deployed this year to double. That’s one data point. The second data point I want to ensure you understand is, from the perspective of EBITDA profit margins, I know you asked about cash expenditure-related issues. But from the perspective of EBITDA profit margins, we generated a 57% margin this quarter. Considering we have a portfolio of long-term contracts and large-scale customers, we previously talked about a contribution profit margin in the mid-20s. But from the EBITDA profit margins of these projects, they are in the range of about 70%. This gives us confidence that once our contracts scale up, they will bring us abundant cash flow. Operator Your next question comes from Brad Zelnick of Deutsche Bank. Please go ahead. Brad Zelnick Awesome. Thank you very much for taking my question, and congratulations on an absolutely amazing year. I just wanted to ask, as we look forward to future contracts, do you see similar demand for five-year contracts? How do these deals differ in terms of pricing structure, expected upfront amounts, or other aspects? How should we view the return on investment (ROIC) of these deals compared to the previous generation of contracts and the economic model you outlined in your initial S-1 filing? Thank you. Michael Intrator Actually, it’s not even a year yet, although it feels like a long time. It’s actually been about 11 months. In our written statement, we talked about a trend we are seeing where the average term of contracts has extended from four years to an average of five years. Obviously, this is very favorable for us, making our business more stable and providing us with great confidence to build and expand our infrastructure. So I want to emphasize first that the customers entering and using that infrastructure are increasingly confident in the lifespan and utility of this infrastructure, which again reflects many important elements we need to build our business For infrastructure pricing, we consider it from the perspective of profit margins. As infrastructure prices change, we are adjusting the pricing we deliver to customers to target the profit margin levels mentioned by Nitin. These are some of the trends you see. As for advance payments, it has always been a lever we use to adjust the economic terms of contracts with customers. One exciting thing for us is that as we continue to lower capital costs, our reliance on advance payments is decreasing. Brad Zelnick Okay. Thank you. Operator Your last question comes from Michael Turrin of Wells Fargo Securities. Please go ahead. Michael Turrin Hey, great. Thank you very much for taking my question. This question is for Nitin or Mike. Can you talk about what gives you confidence in reaching a $30 billion revenue run rate by 2027? How much of that is already booked, and how much is business that your team needs to go after? As a second part, Mike, you mentioned certain segments of the market earlier; can you discuss any changes in demand you see in those segments and how you prioritize within them? Thank you very much. Michael Intrator I’m prepared to answer that in reverse, if I may. One thing we are very focused on as a company is ensuring we have a comprehensive and diverse perspective on compute usage. So we are really doing everything we can to work with everyone who consumes compute resources because we feel that gives us the best understanding of where future demand will come from. As I’ve said before, one thing that will happen is that as compute becomes more accessible, you will see business ideas that don’t even exist yet become a reality. They will become our new customers, and we hope to work closely with them from the very beginning as they build these incredible new businesses. Regarding the $30 billion run rate question, what we are doing is forecasting when the existing contracts we have sold will go live based on the contracted power capacity we have. As I mentioned, our capacity for 2026 is actually sold out, and we are continuing to add contracts that will be allocated to the capacity coming online in 2027, and our customers have shown tremendous and sustained interest in acquiring more capacity to bring more compute power. These are the largest and most reputable companies in the world. These are the most important AI labs in the world. They are the ones building the future of AI, and they are trying to ensure their infrastructure through us. It’s really exciting. After going through this forecasting process, we are very confident in the $30 billion number we have announced. Michael Turrin Thank you. Okay. Michael Intrator As I said, in the 11 months since we went public, I appreciate each and every one of you for being part of our journey in building this company. As we conclude the meeting, I want to thank the CoreWeave team and our partners Without you, these achievements would not have been possible. From product development speed and innovation to excellent operational performance and rigorous financial discipline, I feel incredibly proud and humble about the execution demonstrated by the entire organization. It is the focus and intensity of the whole organization that enables us to continue our trajectory of ultra-high-speed growth and embrace the incredible opportunities ahead. Thank you all for your participation today. Thank you for your support, and I look forward to continuing to update you on our progress in the future. Operator Thank you. Ladies and gentlemen, this concludes today's conference call. Thank you all for participating. You may disconnect now