SharonAI Holdings, Inc.

SharonAI Holdings, Inc.

20.13
USD
-2.43
(-10.77%)
-30.89EPS
3.03BMarket Cap

Q3 FY2026 · Earnings Call TranscriptNovember 16, 2026

Operator

It is now my pleasure to hand the floor over to your host, Ross Barrows, Head of Capital Strategy and Investor Relations at Sharon AI. Sir, the floor is yours.

Ross Barrows

Good afternoon, and welcome to our earnings call to discuss Sharon AI's operating results for the quarter ended September 30, 2026. Joining me today is James Manning, Sharon AI's Co-Founder and Chief Executive Officer, and Anuj Goel, Sharon AI'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 reflect our current expectations and are based on our current view of the industry and our business, they're 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. Sharon AI 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.

James, over to you.

James Manning

Hello, everyone, and welcome to Sharon AI's third quarter 2026 earnings call. I'm James Manning, Co-Founder and CEO of Sharon AI.

For Sharon, 2026 has been the year we build the foundations of this business: capacity, customers, and capital. And 2027 is the year we expect those foundations to materialize into meaningful revenue.

Execution is our number one priority, and we should be measured by clusters being delivered, acceptance of those clusters by customers, and the associated revenue generation. Today, we want to show you that we are delivering and how that delivery converts into growth.

I'll start with why we believe Sharon AI is in position to win, including the role Australia plays. I'll then cover what we delivered in this quarter and our achievements post-balance date.

Finally, how that all builds into recurring revenue, our contracted book, the NVIDIA relationship, and our capacity. Anuj will then take you through our capital strategy, and I'll return with the summary before we open the call for questions.

I'll start with a brief overview for those that are new to the story. Sharon AI is a leading Australian Neocloud.

We design, own, and operate the infrastructure that AI runs on. We enable large-scale training, inference workloads, and other high-performance computing, which we deliver through our platform, combining GPU as a service, an AI inference layer, and high-performance storage as one.

Our target customers span hyperscalers, AI natives, leading AI labs, enterprise, and government. We've disclosed some of our customers, which include NVIDIA, ESS, Canva, and GMI Cloud.

We also describe the profiles of other customers we serve, including a global services company, a global technology, a global AI lab, and a US-based inference provider. How will Sharon AI lead the pack?

I'd point to four things. Our NVIDIA Preferred Partner status gives us priority access to GPU allocations, and we've been working with NVIDIA since 2021, which gives us a natural advantage in our deeper understanding of why and how AI compute workloads work.

Our networking, storage, and orchestration are purpose-built for AI, which supports higher density than traditional cloud environments. Our model is capital-lighter and faster to market.

We deploy into grid-connected tier 3 and tier 4 partner data centers under multiyear contracts, which avoids build costs and long-lead items like substation and power transmission. Finally, our infrastructure is hosted onshore in Australia and New Zealand with reliable and increasingly renewable energy.

Each of those advantages matters more as this market matures, and that is the theme of the next few slides. It's worth looking at the market opportunity and why we are building Sharon the way we are.

Structured research shows Neoclouds like Sharon, which specialize in GPU-based services, gaining a share across every layer of the AI infrastructure. On IT infrastructure, Neoclouds made up about 2% of the total IT infrastructure in 2025.

However, by 2031, that share is projected to reach over 12%. Within the cloud infrastructure space, the Neocloud share is forecasted to roughly triple to about 16% by 2031, while the value of that share, as shown in the middle chart, is forecasted to grow roughly 18 times from about $27 billion to over $480 billion.

Finally, on AI capacity, Neoclouds held about 14% of AI capacity in 2025. By 2030, however, that share is projected to reach approximately 33%, while hyperscalers' share falls from about 81% to below 60% over the same time period.

Put simply, demand for AI compute is growing faster than hyperscalers and legacy market can build on their own, and growing out of share of that capacity is moving to Neoclouds. That is the market Sharon AI is built to serve.

The constraint out there is the ability to deliver this performance-focused infrastructure for the AI age: the power, data center, and GPU capacity all coming together, which is what we are focused on at Sharon AI. Our framework for what takes to win in this market hasn't changed.

We think about GPU allocation, power, regulation, and capital and talent. This quarter, I want to spend some time on power because it's the constraint we hear the most about from customers and investors, and it's where Australia and our model stands apart.

Globally, one of, if not the hardest parts of delivering AI compute is connecting it to power. In most markets, new grid connections, substations, and transmission lines carry multiyear lead times.

Our model seeks to avoid these by design. Every site we deploy into is either an established data center facility that is already connected to the grid or a new build or the expansion of an existing site where power is already established in that location.

Recently, David Burns, our COO, set some of this out in a blog post available on our website. He talked about grid power already being in place, the remaining step to get operational generally sitting at the data hall level, and where our colocation partners are energizing individual halls.

We work with a portfolio of data center providers who are commissioning tier 3 and tier 4 data center facilities across multiple sites so that no single energization determines our overall capacity. What does this translate into?

At all of our sites, we know our contracted customers and when power is on site, with no reliance on future grid utility connections or upgrades. This helps us have confidence in our customer delivery dates, and in a market where this is constrained, a credible delivery date is a competitive advantage.

On other constraints, our NVIDIA partnership continues to support GPU allocation, and we continue to extend our sovereign footprint in Australia. Regarding capital and talent, we have secured approximately $2.6 billion since December last year, and we continue to strengthen our board and leadership team with high-quality additions.

Why Australia? Australia is so well positioned to capitalize as a global AI infrastructure build-out.

Global demand for AI compute is outpacing the physical infrastructure needed to deliver it, and timely access to power is increasingly the binding constraint. We believe Australia has the foundation to become a global exporter of AI compute and tokens, and Sharon AI is built to take advantage of that.

Australia has abundant land, significant energy resources, and available power, strong water security, and strong potential for renewable energy at scale. We are close enough to the rest of APAC, but far enough away to be safe and secure.

Finally, we have strong connectivity, which ensures low-latency delivery as capacity is built out. With this, we are also seeing a change in where compute locates.

Traditionally, we saw a focus on data center proximity to metro regions, and broadly, cloud computing could be defined by the saying that you bring the power to the compute. AI factories are increasingly characterized by the compute going to the power, however.

This shift sees the future AI factories colocated closer to power assets, partly due to the scale requirement, but also due to potential impacts on infrastructure and the local environment. These strategic advantages matter as much to our customers.

They know they have a stable government and supportive policy environment, a skilled workforce, and world-class research institutions, and strong global connectivity and trusted international relationships, including Australia being part of the Five Eyes alliance. For sovereign and regulated workloads, that is a combination that's difficult to replicate.

Domestic demand is also growing quickly. Australian operational data center capacity is forecasted to grow from about 1.6 GW in 2026 to about 3.9 GW by 2030, according to Data Centers Australia, and Sharon AI stands to provide a meaningful amount of this AI factory capacity over time.

Finally, on social license, we are also clear that social license is key. Building AI infrastructure in regional Australia depends on meaningful community engagement and on those local communities sharing the benefits, such as jobs and investment.

That is the opportunity Sharon AI was built for: rapidly deploying trusted, sovereign AI infrastructure in a country with the power, land, and stability to do so at scale. Now to the quarter ahead.

Let me give you the headline numbers first, and then we'll unpack them: 212 MW of total secured AI factory capacity across Australia and New Zealand, of which 129 MW of that is contracted under multiyear take-or-pay agreements; 68,000 NVIDIA GPUs across 2026 and 2027 to be deployed. This quarter was all about converting that position to delivered capacity.

Most importantly, we are delivering. The initial deployments of our global technology customer, a 1k cluster was accepted in August, and the first tranche of that has been billing since the third quarter.

Approximately 30 MW of our contracted capacity is expected to be billing from January 2027, representing $400 to $450 million of annualized run rate revenue. That customer acceptance matters beyond the revenue it brings.

It's independent confirmation from a global technology company that we can deliver at scale and is also the reference point for today's conversations with future customers. $2.6 billion of capital since December 2025, including our first GPU-backed debt facility just last month.

This was our inaugural $356 million fixed-rate GPU-backed debt facility, which shows our ability to execute in debt capital markets. And $9.6 billion of cumulative total contract value, which is up around seven times since the first quarter of this year.

During the quarter, we have signed a $1.32 billion 5-year take-or-pay agreement with a global AI lab. Since the quarter end, we have signed a $170 million 4-year take-or-pay agreement with a US-based inference provider and a $1 billion 5-year take-or-pay agreement with a global services company with a 30% deployment.

We also strengthened the underlying business environment. On the security front, we achieved a SOC 2 Type II attestation.

In operations, we completed NVIDIA infrastructure services testing, or NVIS, verification on all live deployments. With our partners, we announced partnerships with VAST Data and Rafay, which we'll come back to shortly.

Finally and importantly, our team: We strengthened the team to ensure that the business behind all of this grows. David Burns joined, formerly from Telstra Enterprise, as our COO.

Anuj Goel, former Head of Technology APAC for Macquarie Capital, joined as our CFO. Melissa Anastasiou, former General Counsel of Spark, joined as our CLO.

Guy Wiley, former Head of Finance and Treasurer of Telstra, joined as our Head of Capital and Strategy. What does this all translate into?

Today, we'll introduce an annualized run rate revenue, or ARR, into our reporting metrics. We believe that this starts to reflect material change in our business and gives investors better insight to what's coming for Sharon AI.

In September 2026, our ARR was approximately $50 million. With the going live of the ESS and global technology company deployments underway and expected to be fully deployed and billing in January 2027, we anticipate that ARR will step up to between $400 and $450 million, or eight to nine times the September level.

That step-up relates to about 30 MW, or 23% of our contracted 129 MW capacity. It is the first of many meaningful steps, and the balance of our contracted capacity comes into billing in stages throughout 2027.

We expect a project-level contribution margin of more than 80% at full deployment. That is the earnings power of this business: contracted capacity with known billing dates, take-or-pay terms, and strong margins.

Every deployment we bring through the customer acceptances adds to it directly, which is why delivery is our focus. That's why we are focused on execution.

In doing this, we plan to disclose more to point a timing for investors to understand where we are operationally. For our global technology customer, for instance: Tranche 1, which is around 20% of the GPUs in that contract, has been installed and tested, accepted by the customer, and commenced billing in the third quarter.

Tranche 2, which is the remaining 80% or so of the GPUs, the data hall is ready for service and installation is complete. Testing is in progress, with customer acceptance expected in December, and billing is expected to commence from January.

For ESS, the single contract tranche is on the same timetable as the Tranche 2 delivery for the global technology customer, with customer acceptance expected in December and billing is expected to commence from January. This $1.25 billion contract over 5 years was data center ready for service in October this year, and we worked with changes to ensure delivery occurs as fast as possible.

We've also provided a deployment roadmap. It is designed to give investors clear understanding on when and where things will be executed within our business plan.

Canva and GMI have been billing since the second quarter of 2026. As mentioned on the previous slide, Tranche 1 of the global technology company contract since the third quarter.

We're deep in deployment operationally this quarter, ensuring both ESS and that global tech company go live. It all translates to a large first quarter of 2027, where billing commences on the second tranche of the global technology company and on the ESS contract.

The contracts we just announced with the US-based inference provider and a global services company follow, with billing expected for both of them in the second quarter of 2027. The global AI lab deployment follows, with billing commencing across the second and third quarters, and the NVIDIA AICP program from the third quarter, which I'll explore in more detail on following slides.

Turning to contracted revenue, this chart shows the cumulative contract value, or TCV, at the end of each quarter this year and as of today. We had $1.3 billion under contract at the end of the first quarter.

That grew to $7.1 billion at the second quarter, $8.8 billion at the end of the third, and as of today, November 16, contracted TCV stands at $9.6 billion, about seven times where we started the end of Q1, with this TCV only including the floor price of NVIDIA AICP program for 40,000 GB300s online in 2027. That $9.6 billion is spread across a diverse customer base, from AI labs to hyperscalers, large technology companies to inference providers and enterprise.

We'll walk through those relationships on the next slide. Together, these contracts give us multiyear revenue visibility as we bring new capacity online.

Looking at the revenue by customer, we'll now take a breakdown of the total contract value, or TCV, by customer and in the order we signed them: Canva and GMI, around $40 million; ESS at $1.25 billion; a global technology company at $950 million; the NVIDIA AICP program anchor at $4.88 billion; a global AI lab at $1.32 billion; the most recent US inference provider at $170 million, as well as a global services company for another $1 billion. Let's take some insights from those contracts.

The contract we signed with the US-based inference provider in November was at $19 million per megawatt per year. This matched the highest pricing in our book to date.

Then the global services company contract, which also signed since quarter end, is priced at $18 million per megawatt per year. What this shows us is sustained contract wins with long-term duration at these higher prices.

We've also seen customer deposits increase from 10 to 20% of TCV at the beginning of the year to up to 30% of TCV or 60% of CapEx this quarter. We are focused on the quality of the contracts.

The size of our book matters, but so does its quality, and in particular, pricing, which has strengthened throughout the back half of this year. Over the course of 2026, we've seen pricing on long-term contracts, which average approximately 5 years in duration, strengthen and move to more than $19 million of annualized revenue per megawatt.

As you can see the contract-by-contract on this slide, but the biggest upside is in pricing yet, with the NVIDIA AICP program capacity, we expect to sign shorter duration contracts at over $30 million per megawatt. This is the benefit of a book with an ability to take on large portion of shorter-term contracts.

The view of the broader market is that pricing reflects two things: Demand for large clusters with near-term delivery continues to run well ahead of supply, and we can offer customers capacity where power is already secured, which is increasingly scarce. Pricing at these levels, combined with our margins above 80%, is what drives those returns for shareholders.

We expect paybacks on our deployments to be well within the life of the contracts that support them. That discipline we apply to every deployment: We commit to GPUs against signed take-or-pay contracts, and we finance them against those contracts.

With approximately 155 MW of data center capacity in Australia available for sale throughout 2027, including both the uncontracted capacity and NVIDIA AIC capacity to be on-sold, we have room to sell into this strong pricing environment. It's worth spending some time to update the market on our AICP program.

The NVIDIA AICP compute program was signed in Q2 of this year. It's 72 MW and approximately 40,000 GB300s with a 6-year term.

The data center is expected to come online and begin commissioning in first quarter of 2027, with the data hall ready for service and NVIS testing in the second quarter, and customer billing to commence as part of the capacity of the third-quarter build and building throughout the fourth. As a reminder of how the anchor works commercially: NVIDIA's 6-year take-or-pay commitment underwrites the 72 MW and approximately 40,000 GB300s.

That contracted and anchored revenue gives lenders greater certainty, which lowers our cost of capital throughout the build-out. We can then on-sell or upsell that capacity to end customers seeking shorter duration compute at the prevailing market rates.

Where we do, we retain 100% of the anchor price and share in any upside with that alongside NVIDIA. When you look at the NVIDIA contract on a per-megawatt basis, the anchor price is the minimum we will earn on that capacity.

Given the shorter-term pricing I described moments ago in the market, we expect to place much of it into materially higher prices. This is the large upside in the TCV which isn't in the market yet.

Capacity: Turning to capacity as a whole, we announce AI factory capacity, which has grown roughly four times this year from 54 MW in February to 212 MW today, with 129 of those megawatts contracted to customers. We've talked about not waiting on new grid infrastructure to deliver contracted capacity.

The sites we announce have secured power. The remaining work sits at the DC build, the data hall level, and is spread across both partners and sites, all designed to reduce and mitigate our deployment and developing risk.

I touched on the SOC 2 achievement earlier on, but security is the other foundation that enterprise and government customers look for. This quarter, we achieved SOC 2 Type II attestation, and an independent auditor has tested our security, availability, and confidentiality controls across our AI cloud platform, which operated over a sustained observation period.

Many of these customers require third-party attestation before they can contract, this matters commercially as well as operationally. Next, we are working towards our ISO 27001 certification, the international standard that government and global customers expect.

All of this is about ensuring that we have developed a world-class platform designed to deliver regulated and secure workloads for our customers. The last slide for me, for now at least, on our partners.

Our partner ecosystem has been critical to our success to date, and it's what makes our capital-lighter model possible. We orchestrate compute, data centers, networking, the data platform, integration, hardware life cycles around a single Sharon AI platform, rather than owning every layer ourselves.

Our two platform partnerships are worth highlighting this quarter because both bear directly on how we scale. The first is VAST Data.

In June, we committed to deploy 600 petabytes of VAST AI operating system as the data foundation for our AI cloud. This is one of the largest sovereign AI data deployments in Asia-Pac.

On VAST's benchmark of around 6 petabytes per 1,000 GPUs, that is the data center capacity to support the needs of around 100,000 GPUs. Storage should not become the bottleneck as our GPU fleet grows.

This quarter, we also launched Data Enclave with VAST, which brings leading AI models into a customer-controlled infrastructure. The second is Rafay.

In September, we signed a 5-year strategic agreement under which Rafay's platform becomes the central orchestration and operational layer across our AI factories. Rafay standardizes how we provision, govern, monitor, and deliver capacity across sites, tenants, and workloads.

The architecture is designed to support orchestration of up to 150,000 GPUs over the term of the agreement. For execution, that gives us one consistent operating model as we bring more than 68,000 GPUs online throughout '26 and '27 and is expected to speed up delivery and improve utilization.

The model works for our partners as well: Colocation partners gain a tenant that scales inside the facilities that they've already built. I'll now hand over to Anuj to take you through our capital strategy.

Anuj Goel

Thanks, James. As this is my first results call with Sharon AI, I'll take the opportunity to say it's a privilege to join the team at this stage of the build-out.

Since December 2025, we have secured approximately $2.6 billion of capital. This includes $104 million in convertible notes in December, $74 million from the sale of our 50% stake in the TCDC JV in January, $125 million from our NASDAQ IPO equity raising in Feb, $350 million in convertible notes in April, a $1.6 billion strategic financing in June, and most recently, our $356 million GPU-backed facility in October.

In addition, we have and/or will be receiving customer deposits of approximately $800 million as we deploy. Our approach is to match each source of funding to the stage of our company's growth.

Equity and convertible debt funded the platform and our early deployments. Now, GPU deployments are funded with asset, SPV-level debt secured against the GPUs and the contracts they serve.

Let me be direct on one of the questions we get the most from investors: Sharon AI does not rely on the unsecured debt markets. The remaining build-out is being funded in the secured debt markets.

That is, we will raise SPV-level debt facilities against signed take-or-pay contracts and that are secured by the GPUs dedicated to them. This is aligned alongside the capital we've already raised and the customer prepayments we will receive.

The majority of our TCV comes from customers that hold strong investment-grade credit ratings. This is pretty important.

I'll explain why that matters in the next couple of slides. In October, we committed our inaugural GPU-backed facility supported by Goldman Sachs and two other large institutions in Asia-Pacific.

It's for $356 million and is priced at 4-year SOFR plus 518 basis points. It is fixed at 9.95%.

The structure is relatively straightforward: The GPUs dedicated to a customer contract sit in an SPV, and the security package comprises those GPUs and the take-or-pay cash flows from that contract. We also provide a temporary parent guarantee that applies until we achieve customer acceptance.

The facility is sized to the compute dedicated to signed contracts. This is why the secured market is the right one for us.

For a business with our contract profile, the secured market is deep and far more open than the unsecured market, and the terms of our first facility reflect that. Interestingly, customer prepayments or deposits are also a funding source under our contracting model.

We have seen the level of customer deposits increasing, with the market now experiencing customer deposits in the order of 30% of TCV. This reduces the amount of debt we need to raise ahead of each stage.

Moving forward, this debt facility is just the first of an expected series of future facilities. The template is now in place: a dedicated SPV for each deployment, debt sized to signed contracts, and security over the GPUs and their cash flows.

We intend to repeat this same model with our other contracts, including ESS, NVIDIA AICP, the global AI lab, the US-based inference provider, and our most recent win, the global services company contract. By value, we expect more than 75% of the debt requirement to relate to the NVIDIA AICP program, which is backed by NVIDIA's 6-year anchor commitment.

If we include the debt programs attached to other investment-grade customers, this greater than 75 percentage increases well into the 80s. The further secured debt financing work streams are well underway, and we are confident in the process and all the lenders we're engaged with.

We intend to update the market as facilities are committed. With that, I'll hand back to James to close out.

James Manning

Thanks, Anuj. Let me bring this together with five key points on our outlook.

First, demand, which continues to materially outweigh supply: Cumulative TCV is up around seven times since the first quarter to approximately $9.6 billion. We have 212 MW of capacity secured, and our contracting visibility extends throughout 2031.

Second, revenue is expected to ramp materially from the first quarter of 2027, with ARR rising eight to nine times from $50 million in September to $400 to $450 million in January, with further contracts commencing billing throughout 2027. Thirdly, financing for the build-out is well advanced, with $1.3 billion of cash at 30 September, our first GPU-backed debt facility in place, and further financing well advanced in the secured market.

Fourthly, megawatts: We have approximately 155 MW of secured data center and power capacity in Australia available for sale in 2027. With GPU capacity in high demand and short- and long-term pricing firming, we're very well placed to continue expanding into a strong market environment.

Fifth, GPUs: We expect more than 68,000 NVIDIA GPUs from our current contracts to be deployed in 2026 and 2027. We started this year with 54 MW, which has now grown 4x to 212 MW.

We start early next year billing customers at scale, and anticipating strong ARR growth throughout 2027. What matters now is execution, and there are three things we'll hold ourselves to over the coming quarters: billing from January at our anticipated ARR and that this ARR growing throughout 2027; the successful delivery of the NVIDIA AICP program, which is 40,000 GB300s online in Australia in 2027; and further secured financing facilities committed for the next stage of the build-out in Australia and New Zealand.

Those are our priorities for 2027, and I'm confident our team will deliver them. Operator, please open the line for Q&A.

Operator

Certainly. Everyone, at this time, we'll be conducting a question-and-answer session.

If you have any questions or comments, please press *1 on your phone at this time. We do ask that while posing your question, please pick up your handset if you're listening on speakerphone to provide optimum sound quality.

Once again, if you have any questions or comments, please press *1 on your phone. Please hold while we poll for questions.

Your first question is coming from Darren Aftahi from Lucid Capital. Your line is live.

Darren Aftahi

Hey, guys. Thanks for taking the questions, and congrats on the progress, and appreciate all the color on the deck.

On the AICP program, can you maybe just help us understand how the dominoes might fall? Do we expect the GPU financing first before any kind of release announcements?

Maybe kind of some color there, please.

James Manning

Thanks, Darren. Absolutely.

We are well advanced on our debt funding process with the AICP program, currently working alongside rating agencies for the underlying debt. I think you should expect to see us announce the debt before we announce any major customer wins on that compute program.

That said, we're materially drafted some contracts for customers on that program, and we'd expect to announce them shortly thereafter the debt being secured.

Darren Aftahi

Got it. Then could you enlighten us maybe what jurisdictions those customers might fall into?

Is it going to be mostly APAC?

James Manning

The AICP program will be only North America. They'll all be North American.

Yeah.

Darren Aftahi

Okay. Perfect.

Then just one one last one for me. This is a bit real-time in nature, apologies maybe for the curveball.

That the US Commerce Department put out that Bureau of Industry and Security kind of impact this morning in trying to stop circumventing of chips to certain places. Does that impact your business in any way with any of your signed capacity to date?

James Manning

Look, good question. Obviously, we're monitoring those new BIS rules really closely.

We note that, I guess, it's an interim rule and subject to comment and feedback. What I'd say is we anticipate that it's likely that we might have some licensing obligations as written in its current form.

We'd expect a far more favorable outcome than most operators. I think if I think about that, it really reflects that our compute is located in Australia, New Zealand.

We've got our GPUs in Australia, the strength of our control environment, including our robust know your client, KYC, procedures, beneficial ownership screening, continuous monitoring of how infrastructure is used. Some of our existing use cases are likely covered by commercial, reprimanded provisions, further mitigating any additional requirements.

Look, at the end of the day, Sharon AI is a US company. We're listed on the NASDAQ with Australian operations, and we'll comply with all the laws as they are ultimately implemented.

Darren Aftahi

Appreciate the color. Thanks, James.

Operator

Thank you. Your next question is coming from Michael Donovan from Compass Point.

Your line is live.

Michael Donovan

Thank you, operator. Hi, guys.

Thanks for taking my question. On vertical integration, where are you in evaluating your own data center build-outs?

Would you expect to pursue that on your own or with a partner?

James Manning

Great question. I think if you think back to how Sharon's approached our go-to-market strategy more broadly, we've always considered our data centers to be partners for us.

In that strategy, we've always looked to colocate inside of existing data center facilities. I don't think it's a surprise, but we did and I think the market's aware we did acquire a site in regional New South Wales, and we're looking at how we best put that project to the market and get it built out, such that it will be longer-term treated like our partnerships.

That's finding the appropriate partners to help us build and run that facility. I don't think you'll see Sharon being a truly vertically integrated operator in that sense.

We are definitely thinking about this as Sharon remaining a pure-play GPU provider in the market. Where we do identify sites, we'll work with partners to jointly develop those sites and lean on that expertise of those developers to run those assets.

Ultimately, we think that the AI factory with the data center is a separate asset to the GPU asset, and they have different return profiles and different funding profiles. We are currently taking the view that we won't fully vertically integrate.

Michael Donovan

That's helpful. Then you touched upon this a bit in your prepared remarks, but on the capacity you've secured that isn't yet contracted to customers, how are you thinking about your strategy for filling it?

Are you primarily looking for multiyear commitments, or do you want to reserve some capacity for potentially higher-priced, shorter-term contracts?

James Manning

Great question. I think that's one of the great advantages of the NVIDIA AICP program: that we have the ability to finance long term and sell short term.

If you think about that capacity that we have unsold, 72 MW of that is already on 72 MW of that is already on short-term sales program. What we're really trying to do there is achieve shorter-term, higher price per megawatt sales to customers in the market with that capacity.

Look, this is ultimately a balancing act between financeability, ensuring that we can finance these clusters and ensure that they're fully funded, which you need term contracts, and then the shorter-term sales where you can, at the higher prices, to maximize the revenue or dollars per megawatt, which we're very focused on. We're very comfortable with the current mix of GPUs in that 40,000 GPUs on that AICP that we can sell short term.

When we think about short term, that's contracts up to 12 months term, but ideally six between that 6- and 12-month period. It's 40,000 GPUs of our 60 64, 67,000 GPUs.

It's a material amount of that that can be sold short term. Then longer term, we're thinking about getting a more of a blended mix of 50-50, both short term on AICP and that's 50% on the long-term take-or-pay.

I would say to you that on the shorter-term stuff, we're thinking of pricing greater than $30 million a megawatt, and that's a lot more favorable than our term contracts. We're seeing increases from between $15 million a megawatt to up to $19 million a megawatt on the long-term contracts.

There's obviously a desire to sell as much short term as we can, and that's very much achievable for a company like Sharon, where we have that AICP program. I think we're in a good position on that.

Michael Donovan

Thank you for that color, and best of luck going forward.

James Manning

Thank you.

Operator

Thank you. Your next question is coming from Brett Knoblauch from Cantor Fitzgerald.

Your line is live.

Brett Knoblauch

Hey, guys. Thanks for taking my question.

I've got two. One more high level: I think the presentation does a good job outlining Neoclouds and the share of the AI market and AI capacity and where that share is going.

I think if we take a step back and look at maybe what has happened in the markets over the last month, it feels like that Neoclouds have come under a lot of pressure with concerns over funding fears and where they can get the capital to fund CapEx and inherently kind of questioning the viability of that model. Kind of wanted you to elaborate on what the market is effectively pricing in versus what you are seeing and why you are confident that Neoclouds, come 2031, will have the share that the presentation is talking about.

Then just as a follow-up, I believe you guys signed a $373 million contract with an AI platform that I don't see in the presentation. Is that lumped into another one, or did something happen with it?

If you could elaborate on that as well. Thank you.

Anuj Goel

Hi, Brett. Good question.

It's Anuj here. On your question regarding the funding program, I think the first point is that we're looking to fund off only secured debt packages, so as opposed to unsecured market.

What the market is really looking for in the secured space is who are the contracts with, what are the terms, and also the security over the assets. We think that that's pretty robust.

We're getting some very strong interests on all our different debt programs. In terms of the risk factors there, it's just in terms of delivery, which we have good track record inside operational to show them that.

It feels pretty good to us on all that and it's just a process to go through. On your second question, there is a customer that we sort of lifted the bar on and formed a view that we can do better than that, and so we have terminated that customer.

James Manning

From a net-net perspective, Brett, we're up on the quarter on the resales, and I think the focus there, Brett, is on revenue optimization. We're very much trying to ensure that we have customers with the right revenue mix, right security profile, and where we can maximize those returns on these scarce GPU assets.

That's what we're very much focused on. Fortunately, we like to think we've got good contract structures that in the right circumstances enable us to make those strategic moves as a business to ensure that we are focused on returns and delivery of services.

Brett Knoblauch

I'm sorry, I might have missed what you said. Why did you guys terminate the contract?

Was it something on the customer's end?

Anuj Goel

Yes.

James Manning

Yes. Coming to the view with respect to their credit risk profile.

Brett Knoblauch

Okay. I guess the contract you signed kind of gave you that out in it?

James Manning

Yes, we were able to reallocate that supply to a higher price per megawatt and with better security terms.

Brett Knoblauch

Awesome. Perfect.

Thank you, guys. Really appreciate it.

Operator

Thank you. Your next question is coming from Fedor Shabalin from B.

Riley. Your line is live.

Fedor Shabalin

Thank you very much, operator, and good afternoon, everyone. I have two questions.

One is a high level. It's about NVIDIA leadership change.

Have you spoken with NVIDIA since the CEO change? Is the AICP a binding take-or-pay with no change of control or termination for convenience clauses?

Because over three quarters of secured debt tied to the AICP. Have lenders asked for more protection or repriced?

Just curious about your thoughts on this. Thank you.

James Manning

Look, we've got a very strong relationship with NVIDIA at the end of the day. We talked to them almost daily within the business, and those conversations continue to be ongoing.

We are very focused on, from our perspective, ensuring that that relationship with NVIDIA continues to be very strong. We don't want to comment on the various changes within the business.

What I would say is that NVIDIA is very supportive of our business, continues to be supportive. It hasn't changed any of our GPU allocations, the AICP terms, or anything like that within our business.

We haven't seen any changes to the relationship from our perspective. We're very, very focused on ensuring that that relationship continues to be strong with NVIDIA.

They're obviously a critical part of the supply chain from our perspective. We're very comfortable from that perspective.

I think Anuj could probably talk to the financing perspective for that.

Anuj Goel

No, there's been no change or interest or engagement on that AICP debt program. The margins we expect are the same, and to know adverse impact at all.

Fedor Shabalin

Thanks for that. My follow-up is again on this program.

The partial AICP billings, if I understand correctly, starts in third quarter next year. How much of that 72 MW is billing by the end of 3Q '27?

When do you expect to reach full run rate, if you can comment on this one? Thank you.

Anuj Goel

Yeah. I think the simple answer is between July and November, the billing will go live across multiple clusters.

That's the timeframes of billing turning on. You can spread the clusters out over that period.

Fedor Shabalin

That's very helpful. Thank you very much, and continue best of luck.

Operator

Thank you. Your next question is coming from Nehal Chokshi from Northland Capital Markets.

Your line is live.

Nehal Chokshi

Thank you. Thank you for the question.

What's the PP&E you expect to have on your balance sheet or CapEx expense to support the $9.6 billion of total contract value signed so far?

Anuj Goel

That's not something we share directly, but I think, as a rough guide, $50 million per megawatt is pretty close. If you're talking to 129 MW worth of capacity signed up, that should give you a read.

Nehal Chokshi

Okay. Just to be clear, that $50 million per megawatt is just for the GPUs and networking that is not inclusive of the power infrastructure, right?

Because that's on our data center.

Anuj Goel

That includes what we incur, the $50 million per megawatt.

James Manning

That's an all-in number is the way we think about that. Because we're in colocation space, we aren't incurring any of the data center build-out cost.

We do have some non-NRCs, non-recurring costs, around the initial install, the physical install of the compute inside the data centers. We have that included in that number, that all-in number that we work from.

When we're guiding you to that, that's a fairly robust all-in cost per megawatt build-out cost.

Nehal Chokshi

Got it. Then when you're looking to do the GPU-backed financing, obviously you're only going to be getting it on the GPU portion, which will be maybe $35 million of that $50 million, right?

Anuj Goel

It's greater than that. The GPU component is more than three-quarters.

Nehal Chokshi

Got it. Okay.

Anuj, it sounds like you already have term sheets on your desk to cover rest of this CapEx needed to be funded. Is that correct?

Anuj Goel

The way to look at it is we have, for the AICP program, a very clear line of sight, as well as ESS. The other debt programs, we're talking to lenders.

We have initial terms, but not fully fledged term sheets.

James Manning

To give you some context, the AICP represents about 75% or 76% of our total debt requirements, and we're materially advanced on that. If you then add the ESS, which we're material term sheets on that as well, long form.

We're very very comfortable we're substantially down that process. Having recently completed that initial debt program that we announced last month, we're fairly comfortable where we are on the program from taking it from here to a closing in the near term.

Nehal Chokshi

What was the loan-to-value on that $356 million GPU debt facility that you obtained earlier in October?

Anuj Goel

The one that we announced in October was a tad over 75%.

Nehal Chokshi

74%?

Anuj Goel

Just over 75. 75.

Nehal Chokshi

75. Okay.

Thank you. Then finally, last question, sorry.

How much in absolute dollars has Sharon AI received or expect to receive in prepayments for the $9.6 billion TCV signed so far?

Anuj Goel

About $800 million.

James Manning

$800 million. About $800 million, and it's probably important to note that the AICP program didn't have any prepayments, if that gives you a bit of a guidance for the non-AICP component to prepayments.

Nehal Chokshi

Got it. Thank you very much.

James Manning

We are running out of time. I probably got time for one more question, if it's brief.

Operator

Certainly. Our next question is coming from Kevin Dede from JonesTrading.

Your line is live.

Kevin Dede

Thank you. Hey, James.

Thanks for fitting me in. Appreciate it.

Long queue. Appreciate you staying late.

James Manning

Anytime, Kev.

Kevin Dede

I was curious about a couple of things. I don't know if you have the time, but give it a shot.

I was wondering how many GPUs you have up and running now, what you expect to end the year with this year, and whether or not the 68,000 that you've talked to will be up and running by year-end '27 and fully financed.

James Manning

Absolutely. Maybe if we talk about in total megawatts, we will have, let's start with the easy one.

We'll be fully installed and online and financed by the end of '27. I think that's the easy one to knock off on your list.

We have installed the 8k

Anuj Goel

There's about 30 MW.

James Manning

30 MW, that's compute. That's going through client-wise testing at the moment and we'll be online next month.

Yeah. It's the timing.

We've sort of tried to explain that on that slide, Kev, about the sequence. A lot of the gear is on the ground, physically installed, and going through what we'd call L2 L12 testing, we're installing platform, completing NVIS testing and so forth on the equipment.

It is a definitional thing, but a lot of the equipment is on the ground and installed physically, and we're just in the commissioning or deployment phase of the equipment is how I'd describe a lot of it now. That's why we've got that high degree of confidence to give guidance on ARR for January, given that we now have clear line of sight, given that we're in the final testing stages on this compute, and we know we'll be doing customer acceptance and hand-over during December.

We think that that gives us a very strong piece of the puzzle, and obviously some of that was already financed within that puzzle. We're very comfortable about providing that guidance for January with the currently turned on.

Then, very comfortable to give you that guidance that we'll be fully online and fully financed across the entire fleet for 2027. Sorry, Kev, maybe one quick one because we do have to wrap after.

Kevin Dede

No, understand. Thanks, James.

Appreciate it. I was curious about the pipeline both on the customer side and the data center side.

I remember earlier this year you were super excited because you saw a lot of people knocking on the door, and I'm just kind of wondering how that depth looks now and whether or not you think you'll be able to find the data center capacity to support it, given increased interest and demand in the sector.

James Manning

I think we've got 212 MW. We have a lot of that available to be resold on the short-term market, which we think is incredibly powerful from both a capacity and a pricing perspective.

As to raw data center capacity, look, 2027 is quite tight. There are a few more opportunities for expansion, that capacity that we're aware of and that we're working through.

We think we're in a very strong position to take that capacity compared to all of our peer group, given the strength of, a, what we've executed to date and the strong funding pipeline for the compute that we've contracted to date. We're feeling fairly good about if we were to take any additional capacity in 2027, what that would look like.

As to additional growth beyond the 212 MW, I think the company's being very transparent that we are aware of a pipeline and we've been working with our partners on expanding that pipeline. With that, we are not naive or not aware of the broader market sentiments around capital and debt funding of new customer contracts in the current environment.

We're going to take a very measured approach to how we expand data center capacity commitments and customer contracts and ensure that any of those contracts are very financeable.

Kevin Dede

Thank you, James. Appreciate the extra time.

It's definitely a couple of slices above the Bitcoin mining projects.

James Manning

Absolutely. Thanks, Kev.

Operator

Thank you. That concludes our Q&A session.

I'll now hand the conference back to James Manning for closing remarks. Please go ahead.

James Manning

Thank you. Before we wrap up, I just want to recognize the people behind these results.

To the Sharon AI team: Thank you. This quarter was about delivering, and you guys delivered: bringing capacity online, standing up infrastructure for our customers, and executing against demanding timelines is hard work.

You've done it with a discipline and urgency that sets this company apart. What we reported today is really your guys' work.

To our shareholders: Thank you for your continued support and confidence in our strategy. We don't take it for granted.

We are building this business for the long term, and your backing lets us invest ahead of the demand that we see in front of us. To our customers and partners: Thank you for the trust you continue to place in us.

I'd just like to say, I think we're still very early in this opportunity. Our focus from here is simple: execute on what we've contracted, deliver for our customers, and keep building value for shareholders.

Thank you for joining us today. We look forward to updating you next quarter.

Operator, that concludes the call.

Operator

Thank you. Everyone, this concludes today's event.

You may disconnect at this time and have a wonderful day. Thank you for your participation.