目录
I. Opening Remarks
II. Management Remarks
Chief Executive Officer Mike Intrator
Demand and Customer Adoption Continue to Broaden
AI Development Becomes a Continuous Iteration Loop
Rapid Growth in Managed Inference
Power, Supply Chain, and the Foundation for Long-Term Growth
Chief Financial Officer Nitin Agrawal
Second-Quarter Financial Performance
Capital Expenditure and Balance Sheet
Third-Quarter and Full-Year Guidance
III. Q&A
Renewal Opportunities for Prior-Generation GPUs
Long-Term Supply Chain Agreements
Managed Inference and Capacity Allocation
Drivers of Higher Margins on New Contracts
Annualized Revenue Run-Rate Target
Data-Center Approvals and Power Roadmap
Training and Inference Infrastructure
Edge AI, Cloud Platforms and the Competitive Landscape
Revenue Contribution from New Capacity and Vera Rubin Commercialization
Redeploying GPUs as Contracts Expire
Financing Shorter-Term Contracts
IV. Closing Remarks
本内容基于公开资料和研报数据整理，不构成任何投资建议，不代表任何个人观点，仅供学习参考，请理性阅读
I. Opening Remarks
Operator: Hello, and thank you for joining CoreWeave, Inc.’s fiscal second-quarter 2026 earnings call. Following management’s prepared remarks, we will open the call for questions. To ask a question, press star followed by 1. To withdraw your question, press star followed by 1 again. I will now turn the call over to CoreWeave.
Moderator: Thank you. Good afternoon, and welcome to CoreWeave’s fiscal second-quarter 2026 earnings call. Joining us today to discuss the company’s results are Chief Executive Officer Mike Intrator and Chief Financial Officer Nitin Agrawal.
Before we begin, please note that today’s remarks will include forward-looking statements. Actual results could differ materially from those contemplated by these statements due to factors described in today’s earnings release and in our forthcoming Form 10-Q quarterly report to be filed with the U.S. Securities and Exchange Commission. Any forward-looking statements made on this call are based on assumptions as of today, and we undertake no obligation to update them for new information or future events.
During this call, we will discuss GAAP financial measures as well as certain non-GAAP financial measures. Reconciliations between GAAP and non-GAAP measures are included in today’s earnings release. The earnings release and accompanying investor presentation are available on our investor relations website at investors.coreweave.com. A replay of this call will also be available there.
I will now turn the call over to Mike.
II. Management Remarks
Chief Executive Officer Mike Intrator
CEO: Good afternoon, everyone, and thank you for joining us.
CoreWeave delivered an exceptional second quarter. We outperformed our plan across the board, and the operating leverage we have spent years building is now clearly evident in our results. Strong execution across the organization drove record financial performance, rapid capacity growth, broader customer demand, and continued platform innovation.
We generated record revenue of $2.6 billion, up 112% year over year, while our revenue backlog increased to $104 billion and enterprise adoption accelerated. This figure does not include more than $25 billion of additional net customer commitments secured during the first few weeks of the third quarter.
We continued to execute our power strategy, reaching 1.5GW of active power capacity after adding nearly 500MW in the quarter—more than in any other quarter in our history and over three times the capacity added in the prior-year period. We remain highly confident in our ability to reach at least 8GW by 2030.
Adjusted operating income increased to $128 million. Margins expanded significantly as our growing scale increasingly translated into operating leverage.
We continued to broaden our technology stack, launching seven new AI platform capabilities and achieving several industry-first milestones that help customers build, deploy, and operate AI faster and at greater scale.
Our extraordinary progress reflects the collective effort of the entire organization, and our business is only getting stronger. Customer contracts signed in the second quarter are expected to generate margins 5 to 10 percentage points higher than contracts added in recent quarters.
These are not isolated milestones. They validate the thesis on which we founded CoreWeave: the AI market is developing as we anticipated. We believe the AI era has arrived and will ultimately reshape every part of the global economy. Every organization is undergoing transformation, creating opportunities both to redefine existing markets and to build entirely new ones.
The future will be led by first movers that seize this opportunity—both AI-native companies at the technological frontier and change agents within established enterprises. They are learning and iterating at extraordinary speed, which has become a prerequisite for leadership in the AI era. We believe these pioneers require an entirely new platform to unlock AI’s potential at scale.
These convictions are the core assumptions underpinning our strategy, shaping our product roadmap, capital allocation, partnerships, and ultimately how we serve customers.
Today, I want to discuss four areas where our vision is translating into tangible results:
First, demand and customer adoption continue to broaden.
Second, our platform provides the tools required for continuous AI development.
Third, our AI-native architecture delivers superior performance and economics.
Fourth, we have built the power and supply-chain foundation to support years of growth.
Demand and Customer Adoption Continue to Broaden
AI is transforming every organization. Debate over the future demand for AI cloud infrastructure may continue, but our customers’ actions provide a clear signal. Demand continues to strengthen as the market expands across more industries, geographies, workloads, and generations of GPU architecture.
AI is no longer confined to frontier model laboratories. It is increasingly embedded in software, industrial systems, financial markets, enterprise workflows, and defense, security, and intelligence missions.
This breadth is evident in our backlog, new commitments, platform utilization, and pricing environment. Pricing and margins for Blackwell and Vera Rubin SKUs are reaching new highs, while pricing for prior-generation SKUs remains at or above levels seen several years ago. Our near-term capacity remains effectively sold out.
As a result, a growing range of customers is willing to enter into committed contracts on increasingly attractive terms, positioning CoreWeave to continue gaining market share for years to come.
Every organization must rethink what is possible. AI does more than accelerate existing processes; it enables enterprises to redesign core functions, create new products, and enter markets that previously did not exist.
Caterpillar is a compelling example. Together, we will deploy the NVIDIA Vera Rubin platform to support industrial-scale training and inference for physical AI. Caterpillar will use CoreWeave’s AI cloud infrastructure as its data factory, training specialized models to improve the intelligence and productivity of automated construction equipment.
Life sciences is emerging as another important growth area for CoreWeave. Organizations working to solve some of the world’s most complex scientific challenges are increasingly adopting our platform. We recently welcomed Isomorphic Labs as a new customer and are proud to support its mission to address a broad range of diseases.
Financial services is also a major growth area. Flow Traders and IMC recently joined our expanding roster of systematic trading customers. These firms are using CoreWeave to develop and deploy the next generation of quantitative-trading AI models. They chose our platform because we can orchestrate high-performance workloads with the speed, reliability, and efficiency these applications require.
In the public sector, our partnership with Leidos marks an important step in the development of CoreWeave’s federal business. Together, we are accelerating the delivery of secure AI capabilities for defense, security, and intelligence missions.
The future of AI is being built by a new class of innovators. Some are AI-native companies operating at the frontier; others are change agents within established enterprises who are willing to challenge the status quo. CoreWeave serves both.
Descartes is using our platform to develop Oasis 3, the first physical-AI world model accessible through an API. IBM is using CoreWeave to securely conduct experiments in reinforcement learning, agentic tool use, and model evaluation. Through Monolith, our specialized field engineers work directly with customers including Nissan and ZF to accelerate the development of enterprise-ready AI applications.
Customer demand now extends beyond infrastructure. CoreWeave Omni is generating significant interest from sovereign, enterprise, and cloud customers. Over the past few weeks, we signed our first contract, with the project scheduled to begin scaling in 2027.
These examples span different industries and use cases, but the pattern is consistent: AI is moving from experimentation into core operations. Organizations acting decisively are building competitive advantage.
Deployment is no longer the endpoint. As AI enters production, the way applications are built is changing, and the participants that learn and iterate fastest will lead.
AI Development Becomes a Continuous Iteration Loop
Over the past several years, many organizations treated a model as a deliverable: train it, deploy it, and move on to the next project. Enterprises no longer operate that way.
Training, inference, evaluation, and improvement now form a continuous loop. Models and agents in production generate real-world data, which can be used for evaluation and new experiments. Those experiments improve the model or application, which is then redeployed into production. The loop repeats, and capabilities compound over time.
This shift fundamentally changes the demand curve and economics of AI. Compute is no longer a one-time requirement concentrated early in a model’s lifecycle. It becomes an ongoing need that grows with every application placed into production and every round of improvement.
Our AI-native platform was built for this model. It spans frontier cloud infrastructure, a rapidly growing managed-inference business, leading developer tools and agentic solutions, and a best-in-class orchestration and observability layer powered by Mission Control.
Together, these capabilities provide customers with an integrated environment. They can deploy applications through CoreWeave Inference using our models or models customized through serverless capabilities; monitor performance with Weights & Biases; evaluate applications in production; test new models; optimize performance through serverless reinforcement learning or sandbox environments; and validate every change for quality, performance, and cost before returning an application to production.
CoreWeave’s AI research and iteration agent further accelerates this process. It can analyze thousands of evaluation runs, surface insights, and recommend the next experiment within minutes, dramatically shortening the path from an idea to an experiment and then to production improvement.
In the second quarter, we launched seven new AI platform capabilities and achieved several industry firsts. These innovations were co-developed with customers and partners to solve real production challenges. That is why platform adoption is gaining such strong momentum.
This week, the number of model-training runs tracked on our platform surpassed 1 billion. Behind that figure are millions of experiments, thousands of research breakthroughs, and a growing community of engineers, researchers, and institutions building the next generation of AI.
Our AI development services carry higher margins and have already been adopted by a broader customer base than our core cloud business. This is becoming a natural path for customer expansion: the developers building AI applications today are tomorrow’s AI cloud infrastructure customers. By serving them early, we establish relationships that naturally expand as their AI workloads grow.
Rapid Growth in Managed Inference
Only a few months after launch, our managed-inference platform is experiencing explosive growth, currently constrained only by near-term capacity.
From serverless offerings to dedicated deployments, CoreWeave monetizes Token usage while giving customers flexibility in how they use the platform. Companies including Grammarly and You.com are moving from experimentation to live production traffic, running AI coding agents, serving their own fine-tuned models, and deploying open-weight models at scale.
Customers should not have to choose between speed and cost. On CoreWeave, they do not. They choose our platform because it delivers advantages in total cost of ownership, quality, service breadth, and performance.
On benchmarks such as Artificial Analysis, we consistently lead in cost per Token and time to first Token for open-source models including Kimi K2.6, K2.7, Code G 5.2, and MiniMax M3.
This leadership is translating directly into revenue. Within just a few months of launch, the managed-inference platform’s annualized revenue run rate increased from $1 million to more than $100 million. We expect it to reach at least $250 million by the end of 2026.
Pioneers need a different kind of platform. The complete AI lifecycle cannot be supported simply by adding GPUs to a general-purpose cloud. It requires an entirely new approach spanning power, cooling, rack design, networking, orchestration, observability, developer tools, and managed services.
That is why CoreWeave was purpose-built for AI. Our platform is differentiated by the depth, breadth, and sophistication of its technology.
In the second quarter, we became the first cloud service provider to deploy and complete validation of NVIDIA Vera Rubin NVL72. Innovations in software-defined liquid cooling and rack management extended our track record of being first to market.
We also set new MLPerf training and inference records using open-source models running on the NVIDIA Grace Blackwell platform, achieving the lowest inference cost per Token in the benchmark.
Performance alone, however, is not enough. Customers require enterprise-grade security, observability, reliability, and compelling economics. According to Signal 65, CoreWeave can deliver total cost of ownership up to 47% below the hyperscaler average.
Customers also need a platform that integrates these capabilities while offering a broader portfolio of storage, CPU, and networking services across globally distributed data centers.
In July, Gartner named CoreWeave a Visionary in its 2026 Magic Quadrant for Cloud AI Infrastructure. We believe this recognition provides further independent validation of our approach to building an AI cloud.
These achievements are not isolated technical milestones. They translate directly into faster deployment, higher utilization, better application performance, and lower customer costs. Combining deep technical capabilities with best-in-class performance, quality, market leadership, and total-cost-of-ownership advantages is a winning formula for both customers and CoreWeave.
Power, Supply Chain, and the Foundation for Long-Term Growth
CoreWeave is the foundational platform for AI at scale. This market requires infrastructure on an unprecedented scale. We must therefore secure power, sites, cooling, hardware, storage, networking, and supply-chain resources well before demand materializes—and operate them as one integrated system.
As I noted at the beginning of the call, we ended the second quarter with 1.5GW of active power capacity after adding nearly 500MW during the quarter alone. According to third-party estimates, the power capacity we added in the second quarter exceeds the total capacity currently operated by any individual emerging cloud service provider.
More importantly, scale is creating competitive advantages, and our economics should improve further over time. Each new deployment is added to an installed base far larger than it was a quarter earlier. As that base expands, each new build represents a smaller share of the overall footprint, while existing deployments continue to generate contracted revenue.
This is how we convert scale into operating leverage. It drove second-quarter margin expansion and underpins our expectation for further sequential margin improvement in both the third and fourth quarters.
We are also securing the resources required to sustain growth for years to come.
Contracted power capacity increased to 3.7GW in the second quarter. Since quarter-end, we have added approximately 500MW, bringing contracted power capacity to 4.2GW as of today. These figures exclude more than 1.5GW of potential capacity under option, including expansion options at existing sites and signed letters of intent.
Our first several self-build projects are now fully underway, with the first site expected to begin operations later this year. Power resources provide the foundation for further vertical integration, giving us greater operational control and supporting long-term margin expansion.
We are also expanding our global footprint, with more than 1GW of power capacity now under contract outside the United States. This includes our recent entry into Asia-Pacific through a 360MW commitment in Indonesia. The project is expected to begin operations in approximately 18 months.
As we provide services in the regions where our customers and their end users are located, we expect international markets to become a meaningful growth driver. Taken together, we have strong visibility into achieving our target of at least 8GW by 2030.
We expect demand to materially exceed supply for years to come. In this environment, securing power is only one part of the equation. Equally important is securing every critical component required to deliver AI cloud infrastructure at scale.
Building on our close relationships with NVIDIA, OEM, and ODM partners, we recently entered into a long-term agreement with Core Scientific. This partnership demonstrates how we are mitigating supply risk for critical resources to meet customer demand.
Our investments in technology, capacity, vertical integration, supply chain, and global expansion all reflect a single vision: AI will become increasingly pervasive; the fastest-moving pioneers will emerge as leaders; and they will need a platform capable of supporting continuous learning and deployment at unprecedented scale.
Before turning the call over to Nitin, I want to reiterate that CoreWeave enters the second half with stronger business momentum than at any point in its history.
AI is reshaping every industry. The pioneers building the future need more than compute—and that is why they choose CoreWeave. Our AI cloud is purpose-built for the full AI lifecycle, supporting every workload from frontier training to rapidly scaling inference.
Demand continues to exceed supply across industries, geographies, and infrastructure generations. We are serving a diverse customer base at extraordinary scale while steadily improving operating leverage, with clear visibility into the power and critical components required to support growth for years to come.
We have a generational opportunity ahead of us. CoreWeave is an indispensable cloud platform for the AI era. We have never been more confident in our strategy, and our continued execution only reinforces that conviction.
I will now turn the call over to Nitin.
Chief Financial Officer Nitin Agrawal
CFO: Thank you, Mike, and good afternoon, everyone.
The second quarter was exceptional for CoreWeave, marked by robust customer demand, a substantial increase in active capacity, and continued execution against our product and financing roadmaps.
Perhaps most importantly, the second quarter marked an inflection point in margins, with sequential expansion as we have discussed over the past several quarters.
Before reviewing our results in detail, I would like to spend a moment on how demand dynamics are evolving in the current environment and what that means for cash flow and the value of our rapidly expanding infrastructure asset base.
Demand for the CoreWeave cloud remains exceptionally strong across our customer base. Every GPU we bring online attracts demand from multiple customers. Given the scarcity of cloud capacity, we remain highly disciplined in allocating it—prioritizing strategic opportunities, adding new customers while deepening existing long-term relationships, and generating attractive returns that continue to improve.
Our AI products and services beyond GPUs are also scaling materially as customers consolidate more of their spending on CoreWeave. These margin-accretive businesses—including storage, CPU, networking, and software—exceeded $400 million in ARR as of the second quarter. We expect them to continue expanding rapidly as customers increasingly recognize the value we provide and increase their spending on CoreWeave.
The improvement in operating margin occurred before our July pricing adjustments. In July, we raised prices by approximately 25% on average across all SKUs, reflecting the current demand environment and the higher returns customers realize from their CoreWeave platform investments as they shift toward inference. We are also passing through higher component costs to customers.
From a cash-flow perspective, as previously discussed, a typical five-year contract generates strong and improving unit economics over its life, although those economic benefits are not realized evenly.
Costs are primarily front-loaded as capital expenditure and funded through a combination of debt, customer prepayments, and other corporate capital. Once a cluster is delivered, contract revenue begins to ramp, gradually creating a predictable and highly cash-generative revenue stream.
We incorporate all these factors when underwriting expected margins before signing a contract. Once deployed, the assets generate attractive returns, fully repay the asset-level debt used to support capital expenditure, and produce substantial additional free cash flow.
At the end of the initial contract, the cluster no longer carries the associated leverage, and we can either recontract that cloud infrastructure or offer it to the market.
Even without assuming further monetization of this cloud infrastructure, we already earn attractive returns. Every subsequent resale or renewal represents incremental upside beyond the returns realized during the initial contract term.
We are now seeing tangible evidence of this recontracting upside. In addition to current-generation SKUs, our previous-generation NVIDIA GPUs are also largely sold out. As older-generation equipment completes its initial contracts, it is positioned to continue generating strong returns in subsequent years.
Ampere and Hopper systems are representative examples. We recently signed an attractively priced A100 contract extending through 2029. As a reminder, this SKU was introduced in 2020.
Previous-generation architecture clusters are already installed, powered, production-ready, and operating at scale. Customers have validated their returns on these investments. In a market where new capacity is constrained and costs continue to rise, these clusters have substantial value.
Production-grade AI cloud infrastructure is a scarce and valuable asset. Our business economics do not depend on recontracting after the initial customer agreement expires, but we are now seeing longer equipment lives and higher pricing, creating the potential for significant additional upside.
Supported by these favorable factors, we are more confident than ever that investments in products and capacity enabled by industry-leading AI cloud services can generate attractive long-term returns.
Second-Quarter Financial Performance
Turning now to our second-quarter results.
Second-quarter revenue was $2.6 billion, up 112% year over year and 24% sequentially, driven primarily by continued strong execution and customer demand for the CoreWeave AI cloud platform.
Revenue backlog at quarter-end was $104.0 billion, up 46% year over year. As Mike noted, this figure does not include more than $25.0 billion of incremental net customer commitments secured early in the third quarter.
More than 50% of the current backlog relates to contracts under which delivery to customers has already begun. By year-end, we expect that proportion to exceed two-thirds of the second-quarter backlog.
Second-quarter operating expenses were $2.6 billion, including $165 million of stock-based compensation. The increase directly reflected the expansion of active power capacity required to convert backlog into revenue, which also drove corresponding increases in cost of revenue and technology and infrastructure spending.
Sales and marketing expenses increased primarily because we expanded investment in our go-to-market organization as we further diversified our customer base and entered new products and markets.
General and administrative expenses increased mainly because of the personnel costs required to support our growth, although they continued to grow more slowly than revenue.
Second-quarter adjusted EBITDA was $1.5 billion, double the $753 million reported in the second quarter of 2025. Adjusted EBITDA margin was 59%.
Second-quarter adjusted operating income was $128 million, compared with $200 million in the second quarter of 2025 and $21 million in the prior quarter. The result was significantly above the high end of our guidance range, demonstrating the emergence of operating leverage as the business scales. Adjusted operating margin was 5%.
Although we continue to absorb substantial capacity ramp-up costs, margins have already improved as we scale.
Second-quarter net loss was $626 million, compared with a net loss of $290 million in the second quarter of 2025.
Second-quarter interest expense was $640 million, compared with $267 million in the second quarter of 2025. The increase primarily reflected the expansion of our debt balance to support continued infrastructure growth and fulfill contracted customer commitments.
Despite reporting a net loss, we recognized income tax expense because we recorded a valuation allowance against deferred tax assets, as I explained last quarter. Absent significant discrete items or changes in circumstances, we expect the 2026 tax rate to remain broadly stable.
Second-quarter adjusted net loss was $567 million, compared with an adjusted net loss of $130 million in the second quarter of 2025.
Capital Expenditure and Balance Sheet
Total second-quarter capital expenditure was $9.4 billion, slightly above the high end of our guidance range. The elevated spending reflected accelerated customer deliveries and increased construction in progress.
Construction in progress rose to $11.9 billion from $9.6 billion in the prior quarter, reflecting the substantial power capacity we expected to deploy early in the third quarter after securing significant additional power late in the second quarter.
In June alone, we added more than 300 MW of active power capacity—more than in any full quarter in the company’s history.
As Mike noted, the global supply chain remains complex. We continue to maintain operating discipline and leverage our partnerships, including new relationships with partners such as Solidigm, to procure required resources strategically.
Turning to our balance sheet and strong liquidity position, as of June 30, we held more than $6.9 billion in cash, cash equivalents, restricted cash, and marketable securities.
During the second quarter, we made substantial progress in strengthening our balance sheet and broadening and deepening our capital sources, raising approximately $18.0 billion through debt, convertible securities, and equity financing.
These financings included several firsts, including our inaugural European bond issuance and our first publicly issued delayed-draw term loan backed by high-performance computing infrastructure.
Our recently completed second public syndicated term loan marked another important milestone: it was our first financing transaction to include shorter-duration customer contracts.
The transaction was priced during one of this year’s most volatile periods in the credit markets, yet attracted substantial investor interest. Given the market environment at the time, we elected to complete the transaction at full size. The associated spreads have since tightened.
Perhaps more importantly, the transaction demonstrates growing confidence in the credit markets regarding the long-term value of NVIDIA infrastructure operating on the CoreWeave cloud.
This financing is important because it enables us to serve critical enterprise demand at scale while accelerating the expansion of our managed inference platform and increasing our exposure to shorter-duration contracts, which typically carry higher average selling prices and margins.
These transactions attracted broad and deep investor participation, underscoring the market’s sustained interest in supporting CoreWeave’s growth.
Including these transactions, we have raised more than $32.0 billion in debt and equity capital to date.
Over the past year, our weighted-average cost of debt has declined by nearly 300 basis points. Based on our debt balance at the end of the second quarter, that equates to approximately $1.1 billion in annual interest savings.
Third-Quarter and Full-Year Guidance
Given our continued strong execution, we now expect active power capacity to exceed 1.85 GW at year-end, above our previous guidance of more than 1.7 GW.
For the second half specifically:
Third-quarter revenue is expected to be between $3.45 billion and $3.6 billion.
Third-quarter adjusted operating income is expected to be between $200 million and $260 million. As margins continue to expand sequentially, we expect the fourth-quarter adjusted operating margin to reach the low-double-digit range above 10%.
Third-quarter interest expense is expected to be between $860 million and $940 million, reflecting the higher debt balance used to finance accelerated deployments.
Third-quarter capital expenditure is expected to be between $11.5 billion and $13.5 billion, primarily to deliver substantial new capacity to customers.
For the full year, disciplined execution and strengthening momentum across our customer base give us confidence to raise our 2026 revenue guidance to $12.4 billion–$13.2 billion and our adjusted operating income guidance to $960 million–$1.15 billion.
Because we have increased our expectations for capacity deliveries to customers this year and recently secured several major contracts, we now expect 2026 capital expenditure of $35.0 billion–$39.0 billion.
Finally, we are also raising our year-end annualized revenue run-rate guidance to $18.5 billion–$19.5 billion.
The long-duration and attractive margins of our contracted revenue backlog provide clear visibility into future performance. We are highly confident in the targets we have just outlined.
The second quarter demonstrated strong customer demand for CoreWeave’s full technology stack and the discipline of our operating model. We are strategically expanding our customer base while supporting the next wave of enterprise AI applications at increasingly attractive margins.
Customers are increasing their spending with CoreWeave to leverage the full depth of our AI-native platform. We remain on track to deliver continued sequential margin expansion throughout the remainder of the year.
We also made substantial progress on our capital structure, securing the financing required to support our long-term growth plans while reducing our weighted-average cost of capital.
We look forward to seeing you at our annual developer conference, Fully Connected, in September. Our management team and customers will demonstrate how our platform is accelerating the adoption of AI in production environments.
Thank you. We will now begin the question-and-answer session.
III. Q&A;
Operator: We will now begin the question-and-answer session. Please limit yourself to one question and one follow-up. To ask a question, press star followed by 1 on your telephone keypad. To withdraw your question, press star followed by 1 again. Please pick up your handset before asking your question. If your device is locally muted, please unmute it. Please stand by while we compile the question queue.
Our first question comes from Samik Chatterjee of JPMorgan. Please go ahead.
Renewal Opportunities for Prior-Generation GPUs
Analyst: Hi, everyone. Thanks for taking the questions, and congratulations on the strong overall performance. This is Brent.
I would like to discuss two topics. First, you mentioned shorter-duration contracts and the opportunity to renew some legacy contracts as they expire, taking advantage of the current pricing environment.
In renewal discussions, what contract duration do customers typically prefer? Approximately how much of the installed base could come up for renewal over the next few years? Could you help us frame the potential scale of this opportunity? I have one follow-up. Thank you.
CEO: Thank you for the question. I appreciate the opportunity to discuss what was truly an exceptional quarter. We delivered outstanding results across infrastructure, software and solutions, as well as sales and contracting with both new and existing customers.
One important trend we are increasingly recognizing—and for which the market is now providing clear evidence—is that prior-generation infrastructure retains significant value across many AI use cases.
We have discussed this for years. We integrate NVIDIA’s most advanced solutions into our cloud platform and deliver them to customers with the most demanding requirements. These products are indeed critical for certain frontier use cases.
However, the broader environment and ecosystem contain many other use cases that can run on earlier-generation SKUs. We have signed a fully priced contract extending through 2029 for a GPU based on a 2020 architecture. This provides an important reference point for what could happen as contracts on our existing infrastructure expire.
CFO: Capacity approaching renewal currently represents only a very small portion of our equipment portfolio. Average selling prices for prior-generation products remain at or above levels from roughly a year ago.
Another particularly attractive opportunity is that, when the original contracts expire, we can redeploy those assets into managed inference and offer customers a compelling product. As Mike noted, this is a rapidly growing part of our business.
We expect this business to continue scaling quickly and reach an annualized revenue run rate of approximately $250 million by year-end.
Long-Term Supply Chain Agreements
Analyst: Thank you for the detail. My follow-up relates to the supply chain.
You are clearly managing supply chain constraints well and continuing to bring capacity online. Building on your existing agreements, how do you think about entering into long-term agreements across more parts of the supply chain to secure additional supply and continue executing against your planned power and capacity ramp? Thank you.
CEO: That is a great question. Ultimately, my responsibility is to ensure that the company can deliver the products our customers need.
Doing so requires active management of a complex supply chain spanning land, power and data-center shells, as well as GPUs, networking and memory. The growth and expansion of AI are creating challenges across every one of these areas.
Over the past several years, we have built deep, long-term relationships with partners including ODMs, OEMs, NVIDIA and memory suppliers.
We continually assess what actions are required to secure the infrastructure, components and capital needed to deliver our products at prices and quality levels acceptable to customers.
This capability is deeply embedded in CoreWeave’s DNA. It is what we do every day: maintain these relationships and secure everything required to deliver our product—NVIDIA infrastructure delivered through the CoreWeave Cloud.
CFO: One additional point is worth emphasizing: the value generated by output from the CoreWeave Cloud has increased faster than the cost of supply chain inputs, resulting in margin expansion.
As Mike noted in his remarks, the contracts we signed last quarter had typical contribution margins 5 to 10 percentage points above the levels observed in recent quarters.
Analyst: Great. Thank you, and thanks for taking my questions.
Managed Inference and Capacity Allocation
Operator: Our next question comes from Brad Zelnick of Deutsche Bank. Please go ahead.
Brad: Great. Thank you very much, and congratulations on the strong execution.
My first question concerns the managed inference offering, which is off to a very strong start. What have you learned so far? What factors will determine how you allocate capacity between managed inference and traditional take-or-pay contracts going forward? I have one follow-up.
CEO: Thank you. This was indeed an exceptional quarter for us, and we are very excited about it.
We take a holistic view of the product portfolio. Over the past several years, we have made tremendous progress scaling through long-term take-or-pay contracts. Having reached hyperscale, we recognized the need to broaden our portfolio to offer the full range of products customers require, including higher-margin products, software solutions and CPU resources. All these capabilities are essential to customer success.
Managed inference has scaled in a remarkable and distinctive way, growing from $1 million to $100 million in a single quarter. One key lesson is that this represents both a substantial opportunity to deliver frontier compute to customers and a way to redeploy GPUs as contracts expire, maximizing their value to the company over time.
This is a very deep market. Because we control the underlying chip resources, we believe we have an inherent advantage and expect to be highly successful in this market.
CFO: Brad, one point worth noting is that the company announced the completion of DDTL 5.5 financing yesterday. This demonstrates strong capital-markets interest in CoreWeave and a willingness to support our products, including underwriting shorter-duration contracts. That is clearly advantageous as we enter these markets and address customer demand.
Drivers of Higher Margins on New Contracts
Brad: Thank you, Nitin, and thank you, Mike. That leads directly to my next question.
Margins on recently signed contracts improved by 5 to 10 percentage points. Could you break down the drivers? How much reflects shorter contract durations, and how much comes from strong competitive differentiation or other factors? More broadly, what pricing environment are you seeing in the market? Thank you.
CEO: Several factors are at work, and it is difficult to isolate them individually.
Infrastructure delivered through the CoreWeave Cloud creates greater value for customers than alternative solutions. The quality of our platform, infrastructure reliability, security and total cost of ownership collectively drive customers to return repeatedly and expand their usage of the CoreWeave Cloud and infrastructure.
One reason is that customers recognize that the same infrastructure delivers greater value when provided through CoreWeave.
A second factor is that many customers have begun commercializing their own products. They are procuring compute more aggressively and are willing to accept higher margins for us because access to that compute is critical to their success.
This trend is occurring across the infrastructure market but is especially pronounced within our ecosystem. We deliver a premium product, and users of this compute are willing to pay premium prices. That is highly encouraging.
Brad: Thank you.
Annualized Revenue Run-Rate Target
Operator: Our next question comes from Amit Daryanani of Evercore ISI. Please go ahead.
Analyst: Thank you. This is Ervin Lou calling on behalf of Amit. I have one question and one follow-up.
My first question is that pricing appears to be benefiting from several tailwinds, including the greater value you provide customers, pass-through of higher component costs and upcoming recontracting opportunities.
Given these factors, should we still view ARR of $18 billion to $19 billion as a reasonable target for year-end 2027?
CFO: We have just raised our year-end annualized revenue run-rate guidance. Our updated 2026 guidance is $18.5 billion to $19.5 billion, which is incorporated into the guidance we just provided.
Analyst: Understood. Thank you.
Data-Center Approvals and Power Roadmap
Analyst: My follow-up concerns the increasingly complex regulatory environment for data centers, including reports of opposition to data-center construction in certain regions.
Against that backdrop, could you discuss your confidence in deploying more than 3GW of active power capacity by the end of next year and your roadmap to reach 8GW by year-end 2030?
CEO: This is a very timely question and an important one for the entire AI and data-center industry.
First, we believe some regions may move forward with construction moratoriums, but that will not change the market’s demand for infrastructure. It will only affect where that infrastructure is ultimately located.
Our approach to stakeholder engagement is to work closely with the communities that will host this infrastructure, with transparency as the foundation of that collaboration.
Companies must work with local governments, utilities and policymakers to ensure that the facilities they build integrate effectively into local communities. Ultimately, being a good neighbor benefits both us and the community.
A critical element is ensuring that we bear the cost of grid upgrades rather than passing those costs on to local electricity customers. Projects create substantial construction employment during the build phase and permanent jobs once the data centers become operational. Data centers also contribute to the local tax base.
All these considerations are essential when entering a community, engaging with its residents and building the infrastructure required to maintain U.S. competitiveness in AI.
The figures we have provided reflect our current position and published guidance. As of today, they have not been affected by regulatory resistance, and we remain confident in them.
We will continue expanding and engaging with stakeholders. Our data centers are best-in-class, and we expect to be held to that standard as we continue building infrastructure globally.
CFO: Let me add some specific figures.
We currently have 4.2GW of contracted power capacity. In addition, we have approximately 1.5GW of power-secured land options over which we hold exercise rights. Together, that brings us close to 6GW, and it is only mid-2026.
We therefore remain fully on track to achieve our previously stated target of more than 8GW of active power capacity by year-end 2030.
Analyst: Understood. Thank you for the clarification.
Training and Inference Infrastructure
Operator: Our next question comes from Raimo Lenschow of Barclays. Please go ahead.
Raimo: I would also like to discuss training and inference infrastructure...
CEO: That is a great question and one we have discussed for several quarters.
We do not believe infrastructure should be built separately as “training infrastructure” and “inference infrastructure.” We build AI infrastructure.
AI infrastructure must include every component required to support the full AI cycle—from training to inference and continuously back again through each iteration needed to serve customers and their users.
The infrastructure we are building can transition seamlessly to inference workloads over time.
Edge AI, Cloud Platforms and the Competitive Landscape
Raimo: Great. My follow-up is that, following Meta’s announcement yesterday, we have received many questions today about edge AI, and related concerns have resurfaced.
How do you see the market evolving across edge deployments, smaller clouds, emerging cloud service providers and hyperscalers? Thank you.
CEO: One point should not be underestimated: CoreWeave sits at the center of an enormous flow of information across the industry.
Hyperscalers use us, model labs use us and enterprise customers are now beginning to scale on our platform. The feedback from all these customers—and the signals it provides about the market’s future shape—has always been highly important in determining how we position ourselves and allocate compute to serve customers.
Ultimately, we believe some workloads will run at the edge, while others will not have the same latency requirements. Our cloud platform is designed to serve both categories effectively, and we will continue building it accordingly.
Customers continually tell us whether they need more edge resources or more large-scale compute that is less latency-sensitive. That feedback loop remains constant.
Yes, we see edge workloads, and we also see workloads that do not need to be deployed at the edge. We are highly confident that the scale of CoreWeave’s infrastructure and our ability to flexibly allocate it across use cases will become a competitive advantage over time.
CFO: On the increased competition you mentioned, we continue to see demand, pricing and margins rise across the board despite intensifying competition. That is another indication of sustained growth in demand for CoreWeave’s offerings.
We are growing within an already enormous addressable market.
Raimo: Yes, exactly. Understood. Thank you.
Revenue Contribution from New Capacity and Vera Rubin Commercialization
Operator: Our next question comes from Michael Turrin of Wells Fargo Securities. Please go ahead.
Michael: Great. Thank you very much. I appreciate there may be some specific timing factors, but the company added an impressive 500MW of active power capacity this quarter, while sequential incremental revenue was roughly in line with the prior quarter. We have heard management’s comments regarding the revenue uplift to come.
Could you help us understand the timing of the new capacity additions within the quarter? When will the 300MW added in June begin contributing revenue at a relatively steady level?
Also, could you discuss the early market signals you are seeing for Vera Rubin commercialization and the potential uplift relative to prior generations? Thank you.
CFO: In the second quarter, we added approximately 500MW of power capacity, including approximately 300MW in June alone—more than CoreWeave had ever added in any full quarter historically.
Most of this power capacity came online late in the second quarter, so you will begin to see its contribution to the business in the third and fourth quarters.
CEO: Let me spend a moment on Vera Rubin.
Vera Rubin is driving margin expansion from the outset, which is extremely encouraging. Market demand for the Vera Rubin platform is exceptionally strong, and CoreWeave has significant pricing power in delivering this infrastructure through its cloud platform.
A substantial portion of the 5 to 10 percentage-point step-up in contract margins we discussed comes from Vera Rubin SKUs.
We are very excited about Vera Rubin’s prospects. We believe it will be a highly successful product generation for both CoreWeave and its customers.
Michael: Thank you very much.
Redeploying GPUs as Contracts Expire
Operator: Our next question comes from Brett Knoblauch of Cantor Fitzgerald. Please go ahead.
Brett: Hi, everyone. Thank you for taking my question, and congratulations on a very strong quarter.
Mike, based on management’s remarks, the pricing environment appears to be at record levels for both prior-generation and current-generation GPUs.
For GPUs approaching the end of existing contracts, how do you decide whether to recontract them, place them in the spot market or deploy them into your inference offering? How far ahead of contract expiration do you typically make that decision?
CEO: That is a great question and one we continue to evaluate.
It is important to recognize that the inference market is highly dynamic and expanding extremely rapidly. We are still working to keep pace with the buildout of new infrastructure, so the flexibility created as some infrastructure rolls off its original contracts allows us to continue scaling the inference offering while we assess the ultimate size and breadth of the managed inference opportunity.
For some infrastructure where the original contracts expire, we will enter into new term contracts if we believe the economics support doing so.
In assessing the economics, we consider the available contract duration, the company’s long-term stability, and the long-term obligations required as we continue to build and scale.
At the same time, we recognize that, in the near term, we can generate higher margins from this infrastructure by selling capacity under shorter-term contracts.
The market has faced a structural supply-demand imbalance for years, and we expect that to persist for the foreseeable future as supply globally struggles to catch up with demand.
Financing Shorter-Term Contracts
Brett: That leads into my next question.
DDTL 5.5 enables you to finance shorter-term contracts. Combined with growing local resistance and political concerns around data center development, CoreWeave’s execution capabilities appear to position it well to capture the benefits of rising prices.
How has the success of DDTL 5.5 changed your view of target contract durations going forward? Given the useful life of the equipment, do you intend to use this structure more extensively to capture higher margins through shorter-term contracts?
CEO: Executing DDTL 5.5 allows CoreWeave to structure contracts across different durations and pursue the lease portfolio that we believe offers the greatest profitability.
We want to sell compute capacity under long-term contracts, while also using shorter-term contracts to capture incremental margin. We have been pursuing this strategy very actively.
We pioneered DDTL 5.5 in the market, giving the company meaningful access to these opportunities.
There is another very important aspect of shorter-term contracts that I want to highlight. Enterprise customers generally do not want to sign five-year agreements; they tend to plan over shorter cycles.
By enabling the financing markets to support contracts included in DDTL 5.5, we can diversify contract durations, accommodate a broader range of agreements, and open new markets for the company.
These customers want to purchase compute capacity for two or three years. Before bringing DDTL 5.5 to market, that segment was not readily accessible to us.
We can now provide compute capacity on timelines that align with how customers use, procure, and contract for it. We expect this market to accelerate significantly.
Brett: Great. Very clear. Thank you.
IV. Closing Remarks
Operator: This concludes the question-and-answer session. I will now turn the call back over to Michael Intrator for closing remarks.
CEO: Before we conclude, I want to thank our team, customers, and partners for their trust in CoreWeave, their hard work, and their unwavering commitment.
We could not have achieved this without your support. I am incredibly proud of what we have accomplished together and of the work ahead.
We are building the indispensable cloud platform for the AI era. Thank you for joining today’s call and for your continued support. We look forward to updating you on our progress over the coming quarters.
Operator: This concludes today’s call. Thank you for participating. You may now disconnect.