CoreWeave, Inc. (CRWV) Earnings Call Transcript

Good afternoon, everyone, and thank you for being with us. CoreWeave had an exceptional third quarter. Revenue is expanding quickly. Margins are climbing faster than we had planned. Demand from customers is still very strong, and the newest agreements we have signed rank among the most profitable ones we have ever underwritten. Our initial fleet of cloud infrastructure is being re-contracted at premium rates, generating cash flow beyond what we assumed back when those early investments were made. Secured lending, customer prepayments, and operating cash flows form the anchor of our funding. Taken together, this momentum keeps our business resilient and well positioned to handle the current environment. Now for the details. Revenue came in at $3.7 billion, a rise of 168% year-over-year that exceeded the top end of our guidance; we added over $1 billion from one quarter to the next. Adjusted operating income was $379 million, comfortably ahead of our $200 to $260 million guidance, as margins entered the low double digits one quarter sooner than expected. Revenue backlog climbed to roughly $133 billion, and we increased active power to over 1.7 gigawatts, which doubles our active footprint since the year began. During the last quarter, we discussed the operating leverage that would show up once our infrastructure achieved scale and we broadened the business across short and long-term contracts. During this quarter, you can see that progress reflected in both the results we posted and the outlook we provided. We also discussed how strongly we believe in CoreWeave—that AI was going to be woven into the way organizations across the economy run, and that those creating that future would need a cloud of a different kind. That belief becomes more concrete quarter after quarter. Customers of ours are using AI to push biological research forward, assist in patient care, build autonomous systems, and aid people in learning. Through our AI cloud platform, they gain the capacity to chase those ambitions and carry their valuable applications into daily life. Bringing these promising ideas to life calls for tools that let them build and deploy AI on dependable compute at appealing economics. Our job is to provide that foundation and generate attractive returns on the capital we deploy. With the capital markets having been volatile lately, I'd like to walk through how these elements connect. The value customers receive from us is what earns their long-term commitments to CoreWeave. Those commitments underpin our approach to financing, and by investing with discipline we can keep growing the platform our customers rely on while also compounding shareholder value. Underlying all of this is a single core principle: in each decision, we look for the best economic outcome — how we fund, what we charge, and how fast we build. Operating an AI cloud at hyperscale requires significant capital and disciplined execution. From the very beginning at CoreWeave, we understood this, and we built both our business model and our financing approach around the long-term customer contracts that the capital would support. That approach allows us to assess each new funding source and then select the opportunities that maximize returns given prevailing market conditions. When Q3 ended, the weighted average cost of debt for the whole company was 7.9%, a decline of 250 basis points compared with a year earlier. Going forward, we will not fund our business through the unsecured debt market at current yields, and there is no need for us to do so. As has been our practice all along, we anticipate that most of the remaining CapEx needed to fulfill our backlog will be funded via delayed draw term loans aligned with particular customer contracts. Between now and 2027, more than 75% of committed CapEx is linked to investment-grade and enterprise customers, with the remainder linked to the world's top AI labs and AI-native companies. Pricing on these facilities mirrors the customer commitments underlying them as well as our demonstrated delivery track record. Their financing can cover between 70% and 100% of the CapEx involved, and from the time of our IPO they've priced in a range of 225 to 550 basis points over SOFR — investment-grade customers of high quality sit at the narrow end, while AI labs are at the broad end. At today's rates, that works out to under 10% on the high side. With these facilities, we're able to pull capital as equipment gets bought and delivered. Interest and principal on those contracts are covered by the cash flow they produce across the contract term, and CoreWeave gets additional cash flow as well. As we planned, this mechanism has already allowed us to retire over $4 billion in debt. That covers retiring close to 50% of our first two delayed draw term loans—the priciest facilities we have. With the facilities that back long-term commitments, once a customer contract wraps up, the debt has been paid off and CoreWeave owns that infrastructure outright, with no encumbrance. That fully paid-off infrastructure is itself turning into a substantial source of capital. Each year, more of it rolls off contract and brings in revenue without requiring fresh investment. The gear is already purchased, the data centers are already constructed—so our capital intensity declines as we expand. We likewise have a number of other significant capital sources, prepayments among them. We evaluate prepayments through an economic lens, balancing how they affect pricing against what financing would cost. Previously, that pushed us toward unsecured debt markets. Right now, prepayments look more appealing. So far this year, we've gathered prepayments on 70% of our deals, and customers are paying as much as 100% of contract value upfront. Over just the coming three months, roughly $9 billion in prepayments and deposits will come in from signed customer contracts. That figure includes over $3 billion tied to re-contracting existing cloud infrastructure, which we sold at meaningful premiums to their original price. We also anticipate EBITDA less interest—which leaves out the prepayment benefit—to more than double in 2027, boosting our capacity to fund capital needs internally. Additionally, financing from OEMs enables more deployments, especially as an efficient way to fund capacity that serves short-term opportunities. All told, we anticipate these sources will cover the significant majority of our capital needs through 2027, with only a limited amount expected to be raised at the parent company level. How disciplined we are in spending matters just as much as our access to capital. Beyond the commitments we have under contract, our success-based CapEx model allows us to slow down or speed up investment in line with customer demand for CoreWeave cloud and supply constraints, including regulatory impacts and the capital markets; that serves as a risk mitigant. The large majority of our capital is committed only once the contracts and infrastructure are in place. Nitin will cover this in greater depth soon; first, though, I want to connect that financing approach to the economics visible in our results. With more contracted infrastructure coming online, revenue rises on top of investments already made, and every additional deployment represents a smaller portion of a substantially bigger business. That is the operating leverage we talked about last quarter. Adjusted operating margin came in at roughly 1% in Q1, 5% in Q2, and was above 10% in Q3. The economics of new business are getting stronger too. During July we lifted pricing across SKUs by about 25%, and later in the quarter we added a further 10% increase. Pricing has kept climbing in Q4, more than offsetting higher costs.

We're also putting more capacity toward near-term deals where the pricing justifies it—selling capacity at above $40 million per megawatt each year—while rates for new infrastructure under four- and five-year agreements are moving toward $20 million to $25 million per megawatt. Having long-term commitments provides the visibility that supports our build-out. Near-term agreements let us respond to a favorable pricing market and capture premium returns; we weigh that trade-off contract by contract, looking at an investment's complete economics: the revenue it can bring in, its operating costs, and its financing costs. Robust demand gives us choices about the best way to create lasting value. We're seeing that same strength across our existing generation fleet. During Q3, Ampere and Hopper pricing increased by thirty to fifty percent versus the prior quarter, and those deals were struck above the levels set in their original contracts. Workloads vary in what they require from a performance and cost standpoint. Older generations remain economically attractive, particularly inside a dependable cloud at scale. We're entering multi-year commitments that run past the depreciable lifespan of the GPUs, which lifts margins further in the later years because price and operating expenses remain constant while depreciation falls away. Over 130,000 GPUs reach the end of their contracts in 2027, and roughly another 170,000 do so in 2028. A strong pipeline of customers is already competing for that capacity. If you look at where prices are right now, the renewals coming up in 2027 would bring in close to $4 billion of additional annualized revenue, and that requires no extra CapEx; once you add the 2028 renewals, that figure climbs to almost $10 billion of annualized incremental revenue. On top of that, should data center construction slow down across the sector, installed compute gets scarcer, and the follow-on effect is that GPU pricing could remain elevated for a longer stretch; in that kind of scenario, moving first is what counts. Our installed base exits its contracts fully paid off, years before most others in the market. That leaves us well positioned no matter which way market conditions go. If the industry builds fast, we grow alongside it. If supply slips, our installed, fully paid-off fleet gains value. Our plan isn’t based on that scenario. It’s grounded in demand we’ve already contracted, CapEx tied to success, and the agreements we’re signing right now. During the previous quarter, I talked about how training, inference, evaluation, and improvement are turning into a continuous loop for AI development. When an application is in production, it produces information about what works and what doesn’t. That information is used by developers to make it better, test the modifications, and return an improved application to use. Such a process calls for advanced software that helps people learn from each experiment, along with highly reliable cloud infrastructure for the entire life of the application. That’s the area where CoreWeave excels. Our platform is unified, enabling hundreds of thousands of GPUs to operate jointly as a full-stack, high-performance cloud that brings together networking, cooling, orchestration, and observability. That's the reason customers pick CoreWeave for their most critical workloads. In September, we introduced CoreWeave Forge, which consolidates our AI development capabilities into one environment; products and services we created together with leading AI labs are now accessible to enterprise customers, who can likewise build on their own data and expertise. In Q3, 70% of new cloud customers we onboarded were already using Forge. Wayve illustrates how a relationship deepens as the work becomes more demanding. They started out with our software and development tools to experiment and gauge model performance, and today they rely on CoreWeave Cloud to train their family of world models for autonomous driving. The Chan Zuckerberg Initiative relies on CoreWeave for computational biology research. Hippocratic AI is training and deploying agents that support patient communication and care navigation. These efforts aim to understand disease more deeply and help people get the information and support they need through their care. On the engineering side, our teams work with organizations such as Nissan and Aston Martin Aramco Formula One to cut testing cycles and lower the data needed to hit target accuracy. Physical AI holds a broader promise: assisting engineers in constructing superior systems and delivering beneficial capabilities into the real world more quickly. MasterClass relies on Forge to assess dialogues with its teaching agents and to enhance the experiences it operates on CoreWeave Cloud. The aim is to make learning more attuned to each individual, with support available whenever a person requires it. Such customers are performing the scientific, clinical, engineering, and educational work that combines human discovery with AI. CoreWeave gives them a platform for pursuing that work at scale and with speed. This is why the platform and the infrastructure of our AI cloud both matter. Whether a promising idea actually becomes a useful application depends in part on whether people are able to run the experiments, assess the results, and afford to serve that application dependably. More than half of the new customers we gained this year came from industries beyond AI labs and technology, and 10% of Fortune 500 companies are now CoreWeave customers. The number of organizations committing over $100 million to CoreWeave Cloud increased by more than 2.5 times year-over-year. As customers discover more valuable uses for AI, they require more from the platform. That broadens our opportunities and strengthens our relationships. Through the end of 2027, qualified demand stands at 10 times what we can actually supply, and even before we've formally brought 2028 to market, interest in that year already runs several times beyond our available capacity. Our active power now exceeds 1.7 gigawatts, and we still anticipate finishing the year above 1.85 gigawatts. Construction is proceeding at fully permitted sites, so we see no material risks to our 2027 deliveries. International expansion became a strategic priority for us this year. Of the 4.6 gigawatts we have under contract, over 1.3 gigawatts now sits outside the US — across Canada, Europe, and Asia — constructed in the places our customers asked us to operate. During Q3 we opened new markets including Indonesia and India, and we also expanded our presence in Denmark and Spain. Connecting our infrastructure in every region is our global AI backbone, which gives customers the same performance, software, and reliability no matter where their compute infrastructure sits. Delivering the capacity customers have committed to use, and earning strong returns on each deployment, stays our focus. Before handing things to Nitin, let me go back to where I began. CoreWeave was designed to last as a company; operationally, we're expanding with profitability and at significant size. Our approach to funding will continue tying capital to signed customer agreements, backed by appealing returns and drawn from a wide range of sources. By connecting these elements, we can adjust the pace of our investment in response to customer need, the regulatory environment, and market conditions for capital. We're bringing capacity online faster than planned, and that capacity keeps gaining value, as our initial fleet is being re-contracted at higher rates; we remain focused on earning the trust of the communities in which we invest and of the customers we serve. The combination of top-tier price-performance and dependability on our cloud platform, together with the innovation, software, and development tools that enterprises depend on to accelerate their AI strategies, means we've hit an inflection point: our business is growing more durable even as the fundamentals speed up.

We feel sure about what we can do for customers, shareholders, and the communities we're working with over the long term. On that note, I'll hand things over to Nitin.

Thanks, Mike, and good afternoon, everyone. Mike covered the momentum in our business — revenue growth, margin growth, demand, rising contribution margins, renewals on our original infrastructure, and our capital plan. Now I'll walk through the numbers behind that. In Q3, our model delivered the operating leverage we said it would as we scale. We grew active power past 1.7 gigawatts of critical IT load, lifted revenue sequentially by more than $1 billion, and pushed adjusted operating margins into the double digits — 5 percentage points above Q2. Three factors drove the quarter. First, we keep converting contracted power into active power and revenue-generating capacity at a rapid pace. Second, demand and pricing are getting stronger, and that's driven by the returns customers see on their AI spending with CoreWeave. Because our operating footprint has expanded, we have flexibility to capture more of that upside through short-term opportunities at premium prices. Third, platform adoption continues to broaden and deepen, as customers in a range of industries use a wider variety of our products and services. Our AI products and services keep scaling as customers consolidate more of their AI spending with CoreWeave. Customers are moving from experimentation into production, and the returns they measure are what drive our demand. Our managed inference platform, which we launched earlier this year, has already surpassed $250 million in booked ARR. The main constraint on growth in this business is the capacity allocated to it. All of these forces combined are boosting what we earn back from the money we put into products and capacity, and they make us more confident about the economics of this business over the long run; we know the macro backdrop we're working in is a dynamic one. We don't pursue growth regardless of cost. Conditions right now are quite profitable for businesses whose differentiated technology is in demand and that are able to execute. Under our success-based CapEx model, investment is tied to customer commitments. Every contract we sign is underwritten so that it covers the expenses tied to it and generates a profit. In each underwriting we factor in the GPUs, the data centers, the power, and the corporate overhead behind the contract. None of that has changed. As we've noted before, the cash flow we've contracted from commitments already made by customers should cover servicing and paying off our current outstanding debt, plus any extra debt we plan to take on in order to fulfill those contracts. To be clear, this covers secured debt at the asset level as well as unsecured debt held at the parent level, and it doesn't factor in renewals. Moving on to third-quarter results: revenue came in at $3.7 billion, a 168% increase compared with a year earlier, and it topped the upper bound of our guidance, which reflects solid execution along with favorable demand and pricing conditions for CoreWeave's AI cloud platform and services. Our backlog of revenue closed the quarter at roughly $133 billion—a 138% increase from a year earlier and yet another all-time high. More than half of that backlog sits within a single contract, and we anticipate delivery to the customer will begin before this year is out. Operating expenses totaled $3.5 billion, a number that includes $160 million of stock-based compensation. The increase largely mirrors the expense of scaling our active power and converting backlog into revenue, which lifts our cost of revenue as well as our technology and infrastructure spending. Sales and marketing expense continues to support our go-to-market strategy, the broadening of our customer base, and adoption across multiple products. G&A went up mainly because of investments we need to support our scale, and it continues to moderate relative to revenue growth. Adjusted operating income was $379 million, with an adjusted operating margin of 10%, which expanded sequentially from 5% in the second quarter and 1% in the first quarter of 2026. We hit low double-digit margins one quarter ahead of the plan we laid out at the start of the year. That margin outperformance comes from strong execution, a favorable demand and pricing environment, and the operating leverage that scale brings. We still see attractive opportunities in both long-term and short-term contracts. Since the platform combines scale with flexibility, when chances arise we can direct capacity that's available toward short-term contracts carrying better margins, in a deliberate way. That portfolio strategy added meaningfully to the solid results we posted in Q3. Adjusted EBITDA came in at $2.3 billion for Q3, compared with $838 million in Q3 of 2025 — a 169% rise year-over-year and 49% higher sequentially. Adjusted EBITDA margin reached 62%, our best showing across the past five quarters. Those strong numbers converted into $2 billion in operating cash flow during Q3. As we continue to scale, operating cash flow will keep expanding and will grow into a more and more significant source of funding for our growth. For the third quarter, we posted a net loss of $709 million, versus a net loss of $110 million in Q3 of 2025. Interest expense came in at $846 million for Q3, compared with $310 million in Q3 of 2025. Adjusted operating income was $379 million, nearly tripling from the second quarter, even as interest expense climbed roughly a third — that's the route to earnings. The same factor flows through adjusted net loss: $529 million versus $41 million a year ago; we booked an income tax provision even with a net loss, driven by a valuation allowance on net deferred tax assets. As we mentioned last quarter, barring significant discrete items or a shift in circumstances, our tax rate ought to stay broadly consistent through 2026. Now on capital expenditures. Q3 CapEx came in at $14.5 billion, which was above the top end of our guidance range. That reflects strong execution that let us pull forward some deliveries from Q4. The same execution brought construction in progress, or CIP, down to $9.3 billion from $11.9 billion in the prior quarter, as we put significant infrastructure to work against power that came online late in Q2. Year to date, we've brought more than 850 megawatts of power online, which takes our total live footprint past 1.7 gigawatts — doubling our active power footprint from year-end 2025 in only three quarters. That rate of buildout shows how strong and repeatable our execution model is, and it's driven by the full-stack solution we built to speed up how quickly we bring new capacity online. Now let's turn to the balance sheet and how solid our liquidity is. As of September 30th, cash, cash equivalents, restricted cash, and marketable securities came to $6.5 billion; beyond that, we have roughly $5 billion of undrawn capacity available through committed DDTL facilities, along with substantial customer prepayments and deposits that are set to arrive over the next three months. Since last year's IPO, we've demonstrated that we can continue to access the capital markets. That stems directly from our steady record of execution as we turn our $133 billion revenue backlog into cash flow. Over the next three months by themselves, we anticipate roughly $9 billion in customer prepayments and deposits, and we look for that figure to keep rising in 2027. As Mike mentioned, over $3 billion of those prepayments tie to existing cloud infrastructure where no extra CapEx is required. We anticipate EBITDA less interest, which leaves out the benefit of growing prepayments, will more than double in 2027. That gain adds to our expanding capacity to fund capital needs internally. We anticipate asset-level debt will cover the significant majority of the capital needs we have left.

On our balance sheet, there are two very different kinds of debt. One kind sits at the asset level, directly financing customer contracts and backed by committed cash flow and cloud infrastructure. The other sits at the parent company and is serviced by cash flow at that level. During the past year, we funded growth partly through parent-level unsecured debt, since those instruments were priced at levels that optimized our overall economics. As Mike mentioned, at today's yields we don't plan to issue more parent-level unsecured debt through 2027. The asset-level secured debt market, in contrast, has stayed stable, and our loans keep trading near par value. Lenders can see directly the cash flow we generate as we deliver on committed customer contracts. That market is still robust, and its depth for CoreWeave has increased since our IPO; in fact, we anticipate launching our next DDTL within the next few weeks. It will be an investment-grade facility, with multiple rating agencies signaling a high investment-grade rating. From our early discussions, we expect pricing on the facility at a spread implying a 7% to 7.5% yield at current SOFR, while covering more than 90% of the related CapEx. Of the committed CapEx for 2027 that remains and is not yet attached to an existing DDTL, more than 75% is tied to these kinds of investment-grade and enterprise customers. For the rest of our backlog, we expect to use similarly structured DDTLs, with customer prepayments and OEM financings serving as the other main capital sources for funding growth. Beyond the committed customer backlog, we see additional avenues available to us. CoreWeave Omni gives us another lever for growth with much lower capital intensity. By extending the CoreWeave platform to third-party infrastructure, Omni lets us monetize incremental customer demand without ourselves funding the purchase of the underlying assets. We have already signed our first customer, and many others have shown significant interest in running their workloads on CoreWeave while keeping direct ownership of their infrastructure. That is one more tool in our toolkit, letting us balance capital needs against customer demand given the prevailing environment. Now to guidance. Strong third quarter performance and continued execution give us confidence to raise full-year 2026 revenue guidance to $13 to $13.3 billion, and adjusted operating income guidance to $1.275 to $1.375 billion, a 26% increase midpoint-to-midpoint. We still expect more than 1.85 gigawatts of active power at year-end. We expect CapEx of $37 to $39 billion, raising the low end of our prior $35 to $39 billion range; this reflects our expectation to deploy infrastructure and deliver it to customers sooner than planned. For the fourth quarter, we expect interest expense of $1.05 to $1.15 billion. Given the capacity we plan to bring online and the contracted revenue already captured in our backlog, we're lifting our ARR guidance, and we now expect to finish 2026 with annualized run-rate revenue between $20 billion and $21 billion. Going forward, we anticipate adjusted net loss will shrink substantially in 2027, turning to adjusted net income positive during the first half of 2028 and for the whole of 2028. For 2027 in particular, we keep flexibility over the pace at which we carry out our growth plan. Capital will be committed only where demand is strong, financing is available on attractive terms, and expected returns clear our thresholds. As market conditions shift, we can adjust our future capital commitments to match; that has no bearing on our capacity to meet existing customer commitments or to surpass our 2027 ARR target by a meaningful margin. In other words, we hold levers to manage our CapEx. If we opted to concentrate solely on our committed backlog, we would anticipate 2027 CapEx staying consistent with our 2026 CapEx investment, even as revenue and EBITDA grow considerably. To bring that capacity online, we already have close to $5 billion we can draw under DDTLs already raised, and over $9 billion of the needed infrastructure already sits on our balance sheet as construction in progress, awaiting entry into service. Our spending is further backed by roughly $9 billion in prepayments and deposits we expect to collect over the next three months from signed customer contracts, by growing cash flows from EBITDA, and by additional DDTL and asset-backed facilities we expect to raise, among them the near-term transaction I referenced earlier. Taking these sources together, we expect to raise under $5 billion of incremental parent-level capital through 2027. Put simply, we are confident in our ability to fund the business for growth, even amid current macroeconomic conditions. Our results this quarter reflect deliberate choices made years ago about how to build and fund this business at attractive returns. We built differentiated technology, purpose-built for AI. We developed the operating capabilities to bring highly performant cloud infrastructure online quickly and at scale. We designed a success-based CapEx model that links investment to contracted demand, supported by a diverse set of capital sources that let CoreWeave finance its growth with discipline and at reasonable cost. We take our role in the AI build-out extremely seriously. This business is capital-intensive, and we are stewards of that capital. We deployed against committed customer contracts at targeted contribution margins, profitably and at scale. We can speed up or slow down that pace depending on market conditions. That discipline will keep guiding how we grow. The opportunity in front of us is extraordinary. AI is transforming how the world builds, discovers, and creates, and CoreWeave was built to power that transformation. With that, we will open up for questions.

Hi there, good afternoon, and thank you for all the extra detail you provided here. Mike, I'd like to begin with a broad industry question, which is that every one of those supply-demand figures you shared sounds really appealing. A concern I keep hearing from investors is that a wave of new supply is arriving in this industry over the coming one to two years, and our visibility into the demand algorithm isn't perfect. We could argue about when supply and demand normalize, whether that's next year or five years out. Could you walk us through what this implies for the CoreWeave model if and when the industry moves into a stretch of normalized or oversupply. Thank you.

Thanks for asking, and I appreciate everyone joining the quarterly call. We had an extraordinary quarter, and we're absolutely excited to share the results. On your question, the notion that infrastructure is arriving on the market more quickly than demand is growing has been a widespread view for several years now. From what we can see — and I believe this matches what you're hearing from other infrastructure suppliers and buyers — compute demand is growing more quickly than the capacity to supply it. The margin gains on every new contract back that up. You're correct that a lot of infrastructure is coming online, yet adoption of this technology and its spread through the economy is moving faster than that infrastructure can produce tokens, supply capacity, and provide the clusters needed for training and delivery; through 2027 we are essentially sold out. Demand for it is overwhelming. Beyond that, every bit of capacity we've brought to market has been taken — to be clear, it isn't the case that we're sold out through 2028, yet anything we've put out there has been bought. There's substantial demand out there that we could serve with that infrastructure.

Just a few points on that, Gabriella. What we're observing as demand moves across the industry is that this demand is turning into pricing power. For long-dated customer contracts, pricing is rising to $20 to $25 million per megawatt, and that applies to four and five-year contracts, while for shorter-duration contracts it reaches as high as $40 million per megawatt. That demand, we're seeing, converts into pricing power and into our capacity to sort of capture margin for our premium product infrastructure and cloud products.

All of that sounds good. Go ahead, please.

The last point is this: we could sell every bit of the capacity we have to the current buyers waiting in line. We are deciding not to do that. We are deciding to be patient. We are deciding to bring the infrastructure to market once it is further along, since we believe that will produce better economics output for the outstanding infrastructure we hold.

All of that is clear, and Nitin, I have a follow-up for you: I'm quite encouraged by that updated figure of 20 to 25 megawatts. What I want to ask about is the contracts you've already signed at committed megawatt monetization prices — how do you handle the risk in the gap between locking in that number on the revenue monetization side and the cost side of the equation, which can at times be short-lived and/or rise during the stretch between contracting the megawatts with the customer and actually bringing them online? How do you deal with that timing mismatch?

Many thanks for that question. Exactly — our approach there is that capital spending is tied directly to wins. Once a customer agreement is signed, we simultaneously issue the matching purchase orders and secure pricing going forward. On top of that, at the moment we sign those agreements, we already know which data centers the workloads will go into. We have a solid grasp of what those facilities cost and what it takes to run that infrastructure. So as each contract is signed, we're essentially ensuring the economics remain fairly stable all the way through to when we actually carry out those deployments.

Thanks for all the detail. I appreciate the updates.

Hey, thanks a lot, and there's a lot of great disclosure and plenty of comforting details. I wanted to ask how, if at all, some of the recent credit market concerns across the industry affect the way you approach contracting new capacity beyond 2027, perhaps as we think about time to power, customer types, duration, and when you line up the funding?

Thanks for that question. It's a great chance for us to make clear that we haven't brought 2028 to market because we're waiting for it to season further. As for how we think about power and capacity on an ongoing basis, we have a number of levers we use to make sure we stay in sync with the demand signals we have. You can see that in the pipeline of power we're bringing to market, and in the timing of that power relative to where we're providing guidance on what demand looks like. We're not getting out too far on the curve in terms of where we lock in power, because we're trying to coordinate our capacity roadmap with the clients' roadmap. One of the things you'll hear us speak about again and then again is our focus on maintaining a broad, diversified client base, which lets us have incredibly accurate information flow regarding who's going to need compute when, how much compute they need, across multiple industries, whether it's the AI labs, whether it's robotics, whether it's the S&P 500 clients; it's a really really robust information flow coming across our desk as we think about how we're going to enter into power contracts on a go-forward basis.

Thanks for that; it's helpful. Mike, could I just pose one more question to you? Some of the less-established AI clouds are dealing with various growing pains out there, so does that perhaps create an opportunity for you going forward, whether when it comes to customer choice, consolidation, or pricing power on your side? Thanks again.

That's a great question as well, I think. Right? One thing we've done is reach scale, and by doing so we've put ourselves in a very strong position to capitalize on this market as it goes through a certain amount of volatility. Ultimately, an AI cloud combines a software layer that lets people manage their infrastructure effectively with an infrastructure build itself. Our view is that we've done a wonderful job, arguably the best in the market, at delivering the software layer that's needed. When it comes to infrastructure, that's a scale game. It's a game in which you'll need to reach material scale to be an effective player. We've known that from the start. We've built our business around that idea, matching our long-term contracts with a diversified group of buyers while we sit here and try to keep driving scale. The end result is that our software is the best solution and the hardware is available at scale. Both of those give us a chance to be very selective about which clients we want to work with, which clients we want to bring onto our infrastructure and partner with as they start their AI journey or as they keep scaling it.

Thanks so much.

Great, thank you. Appreciate you taking my questions. Mike and Nitin, margins came in strong this quarter. Your guidance also points to a meaningful acceleration, or improvement, in margins as we move into the next quarter too. I was just wondering whether you could flesh out a bit the part that short-term contracts are playing in this margin upside you're seeing. You've talked historically about bringing capacity earlier in the quarter as well, which tends to drive margin upside, but how should we think about, maybe, execution versus the contribution from shorter-term contracts when it comes to the upside on margin, both in the third quarter and going into the fourth quarter guide? I have a follow-up.

Thanks, Samik, for the question. Margin performance in this quarter reflected our strong execution, along with both a favorable demand and pricing environment, as well as our ability to sell premium products to our end customers. All three of those vectors contributed to our overperformance this quarter, and we continue to expect those vectors to keep working in our favor through the course of the near term.

A major consideration for us is that we truly value the stability that long-term contracts provide. We also really appreciate the margin improvements that short-term contracts bring. And we genuinely like being able to deliver compute through offerings such as managed inference. Each of these three is expanding at a very fast pace, and each one gives us a tremendous chance to build and grow our business. In our view, our capacity to add more volume to the short-term business and more volume to our managed inference product stems from the scale of the business we have established on the foundation of the long-term contracts. You will keep seeing us go through the process of deciding how to distribute this valuable infrastructure among the three primary vectors of opportunity that lie ahead of us.

Understood. Mike, there's been a lot of worry lately after the elections regarding what the regulatory climate or regulatory pressures in the US might mean — is there any way you can help us understand how you're thinking about this, especially when you look at your long-term plans for 2030 and roughly 8 gigawatts, and what role your international expansion might play given the current macro backdrop you're facing? Thank you.