管理層發言
Good day, everyone, and welcome to the SharonAI Second Quarter 2026 Conference Call. Please note, this conference is being recorded. It is now my pleasure to hand the floor over to your host, Ross Barrows, Head of Capital Strategy and Investor Relations. Sir, the floor is yours.
Good afternoon, and welcome to our earnings call to discuss SharonAI's operating results for the quarter ended June 30, 2026. Joining me today is James Manning, SharonAI's Chief Executive Officer; and Tim Broadfoot, SharonAI's Chief Financial Officer. I'll now take a moment to read the safe harbor statement. During the course of this conference call, we may make certain forward-looking statements within the meaning of the federal securities laws, including statements regarding our expectations, plans, prospects, strategies, future operating results and financial performance. Although they may reflect our current expectations and are based on our current view of the industry and our business, they are not guarantees of future performance. These statements are subject to risks and uncertainties that could cause our actual results to be materially different from those expressed in these statements and speak only as of the date of this call. For more details on factors that could affect these expectations and cause these differences, please see our most recent Form 10-K and Form 10-Q and other SEC reports filed with the Securities and Exchange Commission and available on the SEC's website and in the Investor Relations section of our website. SharonAI undertakes no obligation to publicly update or revise any forward-looking statement, whether as a result of new information or future events. In addition, during this call, we may discuss certain non-GAAP financial measures. Reconciliations to the most directly comparable GAAP measures and related disclosures are available in today's earnings release and/or on our Investor Relations website. I'll now turn the call over to James.
Hello, everyone, and welcome to SharonAI's Second Quarter 2026 Earnings Call. I'm James Manning, CEO and Co-Founder of SharonAI. I'll begin with the highlights from the quarter and an overview of our market position. I'll then cover some of our recent customer wins, and I'll talk about some additional capacity and our capital strategy moving forward. The central message from the quarter is that we have materially increased each of the three inputs required to scale this business: AI factory capacity, contracted customer demand and capital. Let me give you the headline numbers first, then I'll unpack them. As of today, we have 212 megawatts of total secured AI factory capacity across Australia and New Zealand, which is an upgrade of 80 megawatts from our last guidance of 132 megawatts. 120 megawatts are contracted through multiyear take-or-pay agreements, and I'll expand further on this updated capacity shortly. We expect to have more than 64,000 NVIDIA GPUs deployed by mid-2027. We've raised approximately $2.2 billion of capital since December 2025, and we've executed roughly $8.8 billion of total contract value year-to-date. Three months ago, our portfolio was comprised of 100 megawatts capacity and $2.2 billion of TCV. So the contracted book has grown by roughly 4x and our secured capacity has more than doubled since. That demonstrates both the strength of demand and our ability to expand our supply to meet it. On customers, the standout is a six-year strategic compute collaboration with NVIDIA worth $4.9 billion in total contract value. Alongside that, we have signed a five-year take-or-pay agreement with a global AI lab worth $1.32 billion and a five-year take-or-pay agreement with a global technology company worth $950 million. A few days ago, we secured a five-year take-or-pay agreement with a global AI platform with $373 million in TCV. Notably, this is a B300 deployment with a record price of over $4 per GPU-hour. On platform, we have a growing pipeline beyond our announced capacity, and we've expanded our partnership with VAST Data to 600 petabytes of storage commitment, providing sufficient back-end infrastructure to support continued growth of up to 100,000 GPUs. On capital and governance, we completed a $1.6 billion oversubscribed financing round in June, which followed a $350 million convertible note in April. We've made three significant leadership appointments. Anuj Goel, formerly of Macquarie Group, joins as our CFO; Melissa Anastasiou joins as our Chief Legal Officer; and Andrew Penn has been appointed as the Non-executive Chairman of the Board. Bringing in senior leadership of Andrew, Anuj and Melissa's caliber strengthens our governance and ability to execute SharonAI as we enter our next phase of growth. I'm delighted to welcome the multiple new team members we have added across the organization, including technical operations and sales to the team. SharonAI is a leading Australian NeoCloud and trusted AI infrastructure partner. SharonAI is purpose-built to power the next generation of artificial intelligence and high-performance computing. We do so through our partner-led ecosystem, enabling our customers to confidently build, train and deploy AI that drives productivity, innovation and growth for their customers and themselves. What that practically means is we design and operate AI infrastructure optimized for large-scale training, inference and high-performance compute. We deliver GPU as a Service, AI platform layers and high-performance storage as one integrated solution. And we serve enterprise, government, hyperscalers and AI natives. I'm often asked why are we well positioned? And I'd like to think of it this way. Our NVIDIA cloud partner status supports our prioritized access to NVIDIA's latest generation of GPUs. Our networking, storage and orchestration are purpose-built for AI and HPC workloads. Our Australia and New Zealand hosted sovereign infrastructure is particularly relevant to regulated and sensitive customers in the region. Our capital-efficient deployment model is built around partnering with leading data center operators to deploy their Tier 3 and Tier 4 facilities. And by co-locating with improving data center infrastructure, we accelerate our deployment, reduce capital requirements and minimize the development risk associated with greenfield builds. Finally, while we're headquartered in Australia, our customers are global, and our contract wins this year emphasize just that point. I said last quarter that we solve for one thing and that's scarcity. And using that framework, which hasn't changed, I'd argue this quarter has validated it on all four fronts. From a GPU allocation standpoint, timely access to NVIDIA's GPUs remains one of the most critical constraints in this market. Manufacturing constraints and demand from hyperscalers continue to limit the supply available to everyone and emerging providers are facing long lead times. Our NVIDIA cloud partner status and our six-year collaboration with NVIDIA puts us in a unique position to provide access to AI compute. Power: High-density GPU clusters need substantial reliable power. However, ready data center sites with source power are becoming increasingly scarce due to grid constraints and long regulatory queues. Our multi-site data center relationships underpin our secured capacity, which has now grown to 212 megawatts. On the regulatory front, data residency and sovereignty requirements are becoming increasingly important across a number of markets. That trend supports our locally hosted model, and we extend our footprint this quarter with our first New Zealand facility. And finally, on capital and talent. Executing in this market takes significant capital and highly specialized HPC talent. Our successful capital raisings to date address the first issue. And our senior hires, as I mentioned earlier, address the second in addition to our ongoing technical team buildout. So let me spend a bit more time on NVIDIA and our relationship. This is a first-of-its-kind partnership: six years, an initial 72 megawatts, 40,000 GB300s and $4.9 billion of minimum revenue, or an average of $817 million of revenue per annum at implied base rates. But this partnership does two things. It expands our ability to provide compute access to the broader AI ecosystem, namely AI natives and enterprise customers, and it reinforces supply certainty at scale through the NVIDIA Cloud Partner program. But the other thing we've seen it do is reaffirm to our partners globally that SharonAI is a regional leader in AI compute. We are well positioned to expand our megawatts and GPU opportunities throughout the region with the support of all our partners, including NVIDIA. Next, I want to be clear about how this works commercially because I think it's been misunderstood based on some of the commentary we've seen. Under the agreement, NVIDIA provides a six-year anchor commitment. That commitment helps derisk the capital investment by providing NVIDIA a guaranteed minimum revenue stream for the initial six-year period of the hardware deployed. This is viewed very favorably by debt providers who help fund the substantial capital investment in the GPUs and the associated infrastructure as they can bank the guaranteed revenues in their models. But the pricing under this agreement is guaranteed as a minimum only. That is, it provides a floor, not a ceiling. We expect to secure customers for a significant portion of the GPU capacity at prices above the guaranteed minimum. In those cases, we retain 100% of the anchor price and then share the incremental revenue above it. Importantly, NVIDIA will share in this incremental revenue, too, which creates a new strategic alignment with NVIDIA who are incentivized to support us to both deliver a premium GPU service and to source and secure higher rate-paying customers to maximize the share of incremental revenue. Importantly, if we perform successfully under the initial 40,000 GB300 allocation, we believe there may be an opportunity to expand the program over time. On the contracting model itself, not much has changed from what I described last quarter, but it's worth reiterating. Here is an example showing what a contract might look like. In month one, the customer contracts and prepays an amount. That prepayment lets us submit the purchase orders for the specific GPUs and network infrastructure in a way that reduces our upfront capital outlay. Over months one to four, we receive and install the hardware. The GPU and the other hardware is delivered within three to four months and final payment lands on delivery and installation and configuration takes two to four weeks. From month five onwards, we recognize monthly revenue on reserved capacity for the full term. For take-or-pay contracts, we are paid irrespective of whether they use the compute 100% of the time or 40% of the time, which gives us real clarity on the expected revenues. And at the end of the term, depending on tenure, there might be several years less of useful economic life, so we can recontract or sell to the on-demand market. The question we get asked the most is whether the customers actually recontract. And I'd like to point out a few things. Data gravity, moving petabytes between clouds, is a real switching cost, not moving compute, and the 600 petabytes committed under the expanded VAST Data partnership is there for customers to grow into. Second, the platform itself. Because networking, storage and orchestration are tuned to each workload, switching means rebuilding and revalidating their stacks. Third, the time to compute because redeploying elsewhere means a multi-month hardware and deployment lead time all over again for the customer. And finally, the upgrade path. Because as an NVIDIA cloud partner, we have priority access to generational upgrades of future GPU allocation, we can save the customer from joining the queue for scarce supply. So who are our partners? We see our partner ecosystem as a unique differentiator. We orchestrate a best-in-class ecosystem around a single AI cloud platform: compute, data, networking, data centers, procurement and installation and hardware lifecycle support. We don't need to own every layer. Instead, we combine leading technologies and infrastructure partners within a single SharonAI platform. That model is designed to support faster deployment and more capital-efficient growth. To name a few, NVIDIA is our primary supplier of compute. NEXTDC is our primary supplier of data center capacity. And recently, our agreement with VAST has notably strengthened our storage strategy, and we cannot forget World Wide Technology, which is our exclusive APAC procurement, testing and implementation partner. It's also worth calling out that this partnership approach has had two big impacts. One is that this results in lower operational risk, greater market validation and credibility and two, that internal technical headcount does not need to scale as fast as some others as they internalize these capabilities. And now to capacity. And this is a piece of news I want to make sure it doesn't get lost today. Since our last capacity update, we have secured an additional 80 megawatts in Australia, taking our total secured AI factory capacity to 212 megawatts. To put that trajectory in context, we had 54 megawatts at the start of the year. We have, therefore, increased our secured capacity roughly fourfold year-to-date while accelerating customer wins. Demand has consistently run ahead of what we can supply. So having 92 megawatts of secured available capacity heading into the back half of this year is exactly the strong position we wanted to be in. The pipeline isn't just a number, it's a commitment to deliver compute online. And so I'm pleased to confirm that we have successfully handed over a B300 cluster to one of our customers this month as well. We are actively focused on our next deployment of both B300 and GB300 equipment into the balance of this quarter and in early quarter four. If you look at how the contracted revenue book has built throughout the year, it's a fairly steep line. We started Q1 with Canva, GMI and ESDS with $1.3 billion of total TCV. In May, we announced a global technology company with a major Asia Pac presence for a further $950 million. In June NVIDIA for $4.9 billion and in July, the global AI lab for $1.32 billion. And just a few days ago, we signed another agreement with a global AI platform for $373 million. That takes us to roughly $8.8 billion of total contracted value for the 120 megawatts of contracted capacity, which leaves us with 92 megawatts available to sell. Finally, it's worth turning to our capital strategy. And we secured approximately $2.2 billion of cash since December 2025. That includes the recent $1.6 billion strategic financing closed in the second quarter. The June financing was oversubscribed and led by a cohort of top-tier institutional funds, and we remain grateful to their ongoing support. Many of you will have joined the call today, and we appreciate your continued support and suggestions as we work to deliver our GPUs to customers. I'll now close with four points on our outlook. First, demand continues to materially outpace supply, and we secured 212 megawatts of capacity for deployment by the end of 2027, while our contracting visibility now extends out through to 2031. Second, we're well funded for our near-term build-out following the $1.6 billion financing and other capital raises to date. Third, we expect the first material revenue to commence in the fourth quarter of 2026 as large-scale B300 and GB300 deployments come online. And fourth, we are targeting more than 64,000 GPUs deployed by mid-2027 across our footprint in Australia and New Zealand. We've made significant progress in a short period of time, but the hard work is still ahead of us. Contracted revenue becomes recognized through execution, delivery and operating at the high standards our customers expect. That's what the next 12 months is about and I'm confident in our ability to deliver. Finally, on a personal note, I wanted to take this opportunity to thank Tim Broadfoot, our CFO, for his work in getting SharonAI where it is today. This will be Tim's last 10-Q, and we look forward to Anuj joining our team and leading the next call. Tim will continue consulting with the company for a period, and we wish him all the best in the future. Operator, please open the line for some Q&A.
分析師問答
Your first question for today is from Darren Aftahi with Lucid Capital Markets.
Congrats on all the progress. Just two, if I may. The additional capacity, the 80 megawatts you added this morning in the release, is that source coming from the same partner you're working with? Or is it a new partner? Second question on the NVIDIA partnership, the 72 megawatts, any updates on releasing that? And with that question on the release, in our conversations with customers, I assume it's in the ballpark of where your latest contract was, north of $4 per GPU-hour?
Thanks, Darren. So new partner solution for the additional 80 megawatts. We're fairly confident around some early megawatts potentially as early as late this year, but definitely in Q1 next year. So good to unlock some capacity there and delivery through 2027 from that perspective. So the 80 megawatts is in Australia, and there's some strategic activities we're focused on around that capacity. And at this time, it's probably not appropriate to give you much more detail on it. But as we've been through the whole history to date, it's been about adding consistent megawatts across partners and delivering modules to get them online and get those programs working. With respect to your second question on the customer demand for the AICP program that we've been running for the 40,000 GPUs, I'd point you to the announcement just this week: we sold that capacity for record dollars per hour or price per GPU-hour, depending on which way you're thinking about it for both B300 and GB300. And that's the demand profile we're seeing and the pricing mechanisms in the discussions we're having with our current customers. So we are seeing quite a constrained market ultimately for access to GPUs. And with those constraints, we're being able to incrementally increase the price per hour that we're getting. Based on the customer demand profiles we're seeing, we'd expect that strong pricing to continue throughout the year.
Your next question is from Brett Knoblauch with Cantor Fitzgerald.
Related to the NVIDIA contract, I know it's quite unique there, and congrats on adding the additional capacity in Australia. What is your priority to resell the potential backstop capacity from NVIDIA or to sell the remaining capacity or the remaining 92 megawatts that you have? Is there a preference for what would come first or what would NVIDIA want first? How should we think about that?
Great question. We've often talked about our sales cycle, Brett, and that's probably the way we think about this. So when I talk about the program that we've got currently going to resell the space in Melbourne, that's compute that's very well designed. We have a very clear path about how we're going to build that out, what the compute form is going to be when it's coming online, all the RFS dates are done. And so we know with that knowledge, we can start giving customers RFS dates and contracts. So short term, we're very focused on the resale of that NVIDIA capacity because there's a lot of deals there for AI natives, and we're seeing a lot of demand in there. And the program really put us on the map globally for a lot of other customers that we didn't historically have relationships with. And so we've got some great relationships, which is giving us really good insight into the other capacity that we've just announced. Quite often, as I've spoken about on several calls, key to us is when we get capacity online and we know we've got energy or white space, we then have to go through a design process to get the right form factor of compute to then be able to take that out to customers. So we're early in the journey on the additional megawatts, but we are already having those conversations with customers. One of the great things we're seeing out of the resale process on the AICP is we're talking to these AI natives and they're looking at what's the rest of your capacity? What are you saying to our sales? What are you seeing for '27? What are you going to have online for '28? And so we're getting a lot more further out insight as to what customer demand profiles are looking like. They're all asking for it. They're like, can we guarantee if we get 5,000 GPUs out of the 40,000 on this, can you guarantee some 5,000 or 10,000 in your next bit of capacity that you're going to be building out. And so that's amazing from a forecasting perspective. It gives us a lot of confidence, but it also enables us to start to talk to those customers about specifically what they're looking for. Are you looking for a cluster with more storage next time? And so we can do a bit more planning. Having released that additional capacity publicly and now being able to talk to customers about where we see that pipeline and what's publicly available as pipeline and then when we talk to them about what's not publicly available as pipeline, it's very helpful overall from an organizational perspective about planning overall capacity and how we're thinking about growing the business.
Very helpful. On contract duration, if I look at all the contracts you've signed, maybe absent some of the really small ones, it's been five years, except for NVIDIA at six. Is there a target duration you're looking for when you ultimately resell NVIDIA capacity? Like is it more one, two years? Is it shorter? Is there a target duration that we're thinking of?
We're largely being driven by customers on that component and that conversation. I think every customer — it's a bit of a balance between price and duration. Every customer would like to have the longest term they can. The demand we're seeing is in the three- to five-year range. But they all want to lock up as much as they can. So we're trying to find a balanced book where we take the limited resource of 40,000 GPUs and split it between a mix of three- to five-year contracts, but also, depending on what we see that customer's forward demand profile or curve is, think about how we match those things across future demand as well. We want to try and find those customers that we can expand, because once we've landed a customer, it's a lot easier to expand those relationships.
That makes sense. On the storage with your partnership with VAST, it's about 100,000 GPUs, which is about similar to how much megawatts you've now secured from the 80. At what point would you look to expand that just ahead of additional capacity ramp in the future? Are you thinking about that yet? Or is that still a bit of a ways out?
We always like to leave a few breadcrumbs in an announcement. I think those early indications of where we're thinking as we sign those deals like the one we did with VAST was a good indicator about where we thought the business is going and where we thought we'd be announcing our megawatts as we came into this period. We're always in discussions with VAST. They've been an amazing partner. We are looking at how we expand that storage. The other thing I'd say more broadly on storage is we've seen huge customer demand and shifts in those storage dynamics. How we design a facility and how we turn on a facility is changing as we're realizing with customers we need to take more storage into a design beyond the standard three petabytes per 1,000 GPU reference architecture. Customers are looking for more storage. As we think about that, the recent $1.32 billion contract was 10 petabytes of storage for 1,000 GPUs. That's a material upgrade from three. That means you have to think about storage capacity and the additional loads, traditionally air-cooled loads attaching to our GB environment. So there are mixes here we have to start considering as we're seeing these shifts in storage.
Your next question is from Fedor Shabalin with B. Riley Securities.
My question is kind of a follow-up to the first two questions. On the NVIDIA partnership and the GB300 capacity under the management, what kind of customers are you targeting to fill that capacity? And can you frame how much of it you expect to be contracted take-or-pay versus sold on demand? Is there a preference here? Related to that, does the mix skew differently by customer type, like hyperscaler versus enterprise? How does that affect the GPU hour rates you're underwriting? And can you comment on what the deployment schedule looks like for these 40,000 GPUs?
Thanks for the question. For the AI natives that we're seeing on the AICP program, you can expect the vast majority will be AI natives on take-or-pay. They are various model builders and inference providers, and we'll be deploying that over the first half of 2027, and that compute will be online. There will be a little bit of spot, but the vast majority will be those three- to five-year terms on a take-or-pay basis. We're very focused on those customers that we can grow with. The great thing is they give us early conversation about what they need elsewhere in our capacity pipeline for '27. We're seeing AI natives wanting to lock up as much compute as possible for as long as possible. It's easier to have those customer relationships rather than a purely spot market approach because egressing a customer with thousands of GPUs is significant work. So it makes sense to keep them locked in for longer terms versus short-term spot for that sort of deployment. If it's inference stuff, you can see inference come and go faster, but a lot of AI natives are looking for longer-term solutions with more storage deployment. We're targeting deployment over the first half of 2027 with full billing in Q3 across that cluster.
That's helpful. And my follow-up is you've guided to revenue ramping materially from third quarter this year through 2027. What's the biggest swing factor that could push that ramp into Q4? For example, if I recall correctly, service start date is September 16 for one of the customers. What could potentially happen or can you reassure us that this is the starting date?
Great question. We're currently tracking RFS dates from a data center perspective. We're relying on our data center partners to make sure they do their delivery. We are carefully tracking and monitoring our supply deliveries, such as Supermicro. If you ask what the risk factors are: hardware delivery and data center readiness are the two main risks, plus third-party supplier relationships. We believe we'll have a very solid Q4. We expect an end of Q3 turning on, so Q4 will be where you materially see that revenue ramping. We've delivered the B300 to that customer, so we're mid-quarter now. You'll see a month and a full month and a bit of billing in Q3 for the B300 as well. As we get the hardware deployed and handed over to the customer, revenue recognition follows. Physically, the data centers look like they're in good order; we're getting through those deployment processes. So I think you'll see a solid Q4 result on those numbers.
Thank you very much. And I promise this is the last one. You're targeting mid-2027 GPU deployment and you've upgraded secured capacity multiple times since June. Is that upward revision being driven by signed contracts pulling capacity forward or by anticipated demand ahead of signed paper? What conditions make you confident to post a contracted megawatt target if you will?
Customer demand is definitely there, and the data center delivery piece is something we can see can be built and delivered in the timeframe. The 64,000 GPUs we're talking about are contracted demand by mid-2027, so we're very comfortable with that. We upgraded to 80 megawatts today, and we'll look to update additional megawatts in the future. We've been slowly building this out. We moved from 132 megawatts to 212 megawatts by the end of 2027 today. We won't promise to upgrade at that same rate indefinitely, but we are ambitious about adding capacity. We'll do it in a measured way. We need to ensure customer signing contracts and financing economics are in place before we announce more capacity. We won't announce capacity without the right dynamics in place, around both customers and financing.
Your next question for today is from Michael Donovan with Compass Point.
This is Ian Generes calling in for Michael Donovan. Congrats on the continued progress and signings. My first question: your partnerships now include NVIDIA, Dell, VAST and a number of data center operators. Can you talk about how those relationships support the growth strategy from here, whether that's validating next-gen GPUs, and what kind of line of sight they give you into future demand?
We see demand cycles through our partner network. Partners refer business and that's helpful. When we talk to our supply side on demand, we're definitely hearing about supply constraints in the market. When we speak with Dell, Supermicro and Lenovo, the story is overwhelming large demand. With storage partners like VAST, we hear what's happening in storage for NeoClouds and others. By using this partner network, it's all about lowering our execution risk. Everyone needs a relationship with an OEM, but when we have a relationship with WWT and data center operators, it lowers our overall operating risk. We also get a lot of customer referrals through those channels. That partner-led model gives us insights into changes in customer profiles and how we need to adapt before implementing changes in our business. For example, the 10 petabytes per 1,000 GPUs requirement we discussed will inform how customers and partners plan beyond our own environment. That information flow through the network is very valuable over time.
That's very helpful. As those conversations extend into the next generation, how are you observing pricing dynamics on Vera Rubin? Are customers engaging on Rubin commitments today for late '27, '28 deliveries? How do you see pricing trending relative to GB300s, for example?
We haven't started pricing Vera Rubin yet, but we are seeing extraordinary demand for it. We're actively having capacity and design conversations with NVIDIA around specific capacity and compute workloads, and that's when we can have customer conversations for specific compute demand in those locations. There's a lot of early demand for Vera Rubin. Customers we're talking to on AICP are already asking about late '27 VR capacity: can we get some of that, can you promise deliveries? We're working through where Vera Rubin deployments will be for us in late '27 or early '28. Customers are already looking for us to secure and lock in those deliveries for them, so we're conscious of that in data center procurement and design for implementation.
Your next question for today is from Jonathon Higgins with Unified Capital Partners.
Congratulations on the momentum. First, on capacity, you're averaging about $1 billion in TCV being signed, if not more, every month and deal frequency is getting better. How do you strategically think about that capacity? You've raised it today to above 200 megawatts. How should we think about that into 2028 and what you're seeing on the demand side?
Capacity is a great question. We're thinking about how we grow, and we haven't provided guidance beyond '27 for additional megawatts beyond what we've announced. We've taken an approach where once we announce capacity, we're focused on designing and delivering that capacity and allocating it to customer contracts. There is momentum and we're contracting at a faster rate, and we're focusing on customers that can grow with us and bring on high-quality customers that will take up that capacity. You'll see us announcing customer contracts attached to the capacity we've already locked under AICP. We're starting to work through outlook capacity coming up. There are a few larger classes and some smaller classes we're focused on for deployment. I'm not promising the same pace indefinitely, but we have many customer conversations materially advanced for the existing AICP cluster and early conversations about larger announced capacity coming online. We will finalize designs, form factors and RFS dates with OEM partners and balance what we deploy for GB300s and later for Vera Rubin in 2028. We'll need additional capacity and are ambitious about growing, but we'll match capacity to customer demand and ensure we can finance and deploy appropriately.
I understand that. And second, you talked about the sovereign capability of the group and demand in Australia, New Zealand and Asia Pacific. Can you give us an idea of how demand or shortages out of ANZ and Asia compare to what you're seeing in the U.S.? Are they having a greater inability to source compute than the U.S. market?
The entire market is constrained. Whether we're talking to customers in North America, Asia, or Australia, demand is strong and supply constrained. For the subset of customers we're talking to, for hundreds of thousands of GB300s there's extreme demand. That's why releasing additional capacity and announcing it is useful: we can tell customers we can give you 5,000 GPUs now and work with you on additional capacity for '27 and '28. Those conversations give customers a pipeline and access to compute. We're not only focused on large customers; smaller 1,000- or 500-GPU clusters matter, because as you land and expand those customers it's important for growth. Demand from both the U.S. and Asia is equally strong, but we're prioritizing a balanced customer book and customers with strong growth profiles so we can expand them over multiyear terms.
One more on the financial side: you've got more dry powder than you had at the start of the year. With the NVIDIA deal and recent financing, can you talk about how you're seeing IRRs in the business, how the cost of finance has moved for you, and the ability to access financing over the last several months?
What we did in the last quarter was phenomenal, and we're grateful for our ongoing shareholder support with the $1.6 billion raise and the earlier convertible note. That support has been instrumental. On debt markets, we're advanced on debt facilities across the business and expect to be coming to market to explore and explain some of those solutions near term. I won't call out specific pricing on this call. But you can see strong top-line pricing per GPU-hour and per megawatt. We're seeing very strong pricing on the compute side reflective of strong customer demand, which supports strong IRRs and is attractive to lenders. We've recently concluded our full technical diligence to lenders in short order and in very good order. So we're very comfortable about delivering on that program. Great. Thank you very much. So I think that concludes our call today. I just wanted to say thank you to our shareholders and everyone on the call. Importantly, I want to thank our team. They continue to execute. We got that B300 on just recently, and we're moving to deliver the next batch of compute. It's really important that our team hears my thanks for delivery because 2026 and 2027 will be the years about delivery for us. I want to reiterate that we're both well positioned financially and operationally to continue to grow the APAC story across Australia, New Zealand and Asia Pacific markets across the balance of the year and beyond. We just want to thank everyone for their support and time today. Thank you.
This concludes today's conference, and you may disconnect your lines at this time. Thank you for your participation.