> ## Content Index
> Fetch the complete content index at: https://www.sotkn.com/llms.txt
> Use this file to discover other available public pages before exploring further.

# What Does Sharon AI Do? The Accept-Then-Bill Neocloud Model
- URL: https://www.sotkn.com/stack/neoclouds/what-does-sharon-ai-do/
- Published: 2026-11-21T04:43:04.000Z
- Updated: 2026-11-21T04:43:04.000Z
- Description: Sharon AI rents Nvidia GPU capacity from halls in Australia and New Zealand. It has about US$9.6 billion of contracts but booked US$9.9 million of revenue last quarter. The power is on. Billing starts only when customers sign off each cluster, and most of that is set for December and January.
- Author: Sean
- Tags: #layer-neoclouds

STACK · Neoclouds · Status: Holding · Last reviewed: 21 Nov 2026

Sharon AI has signed about US$9.6 billion of contracts to rent out GPUs. Last quarter it booked US$9.9 million of revenue. Both numbers are true. The space between them is one step: a customer has to test each cluster and say yes before the meter starts.

**In short.** Sharon AI is an Australian neocloud listed on Nasdaq. It buys Nvidia GPU systems, puts them in partner data centers in Australia and New Zealand, and rents the computing to AI labs and tech firms, mostly on five-year take-or-pay contracts. It says it has about US$9.6 billion of contracts. Last quarter it booked US$9.9 million of revenue. The gap is not demand and, on its account, not power: power is on at all 12 sites with contracted customers. Our read: Nvidia, the landlords and the lenders get paid first, and Sharon AI earns last. The stuck step is customer sign-off. Each cluster starts billing only when the customer accepts it, and the big ones are due to pass that test in December 2026 and January 2027.

Why an investor cares: Sharon AI shows the last gate on the neocloud layer very clearly. Chips are bought, halls are powered, contracts are signed, and still the money waits for a sign-off. If you are new to the group, start with [who the neoclouds are](https://www.sotkn.com/explained/who-are-the-neoclouds/). This page is about how Sharon AI makes money, who gets paid before it does, and where the chain is stuck today.

## Numbers that matter

| Number                   | What it is                                                                              | Why it matters                                                              |
| ------------------------ | --------------------------------------------------------------------------------------- | --------------------------------------------------------------------------- |
| **US$9.9m**              | Revenue, quarter to 30 September 2026, with a net loss of US$43.8m                      | Today's business is small next to its contracts                             |
| **US$9.6bn**             | Total contract value at November 2026, over about 5.5 years                             | US$4.88bn of it is Nvidia's AI Compute Program, counted at the floor price  |
| **30 of 129 MW**         | Contracted capacity expected to be billing from January 2027                            | About 23% of what is signed, per the company                                |
| **US$50m → US$400–450m** | Annual run-rate revenue, September 2026 to January 2027 target                          | The jump depends on two clusters passing acceptance in December and January |
| **US$1.39bn**            | Construction in progress at 30 September, against US$89m booked as technology equipment | Most of the hardware money is not earning yet                               |
| **\~US$0.8bn**           | Customer deposits, typically held in escrow until each cluster is accepted              | Cash that is promised but locked until sign-off                             |

*Figures are listed at the end of the page. Sums and ratios are our arithmetic from those filings.*

## What does Sharon AI actually do?

It rents GPU computing. Sharon AI buys Nvidia systems, installs them in data halls, and sells the capacity to customers who want their own block of chips for training and running AI models. This is the [GPU-as-a-service](https://www.sotkn.com/explained/gpu-as-a-service-explained/) model. Most of it is sold on long contracts for reserved capacity, not by the hour.

The company calls itself a leading Australian neocloud. It listed on Nasdaq in February 2026, raising US$125 million, and trades as SHAZ. Its head office address in filings is in the US, but its chips and customers' data sit in Australia and New Zealand. It pitches that as sovereign hosting: data that stays onshore, in tier III and IV halls, with a SOC 2 Type II security report finished in the third quarter.

It does not build its own buildings. Its results presentation says it deploys into grid-connected partner data centers under multi-year leases, which avoids build costs and long waits for substations and power lines. One example is a 15 MW hall in Western Sydney that GreenSquare rebuilt from a 1995 data center; our [Catalyst note on that hall](https://www.sotkn.com/catalyst/greensquare-sharon-ai-15-mw-reused-1995-sydney-hall/) covers it. This is the [colocation model](https://www.sotkn.com/explained/hyperscale-vs-colocation-data-center/). Rent is cheaper than building, but it is due whether or not the chips are earning.

The scale has jumped this year. Sharon AI says it has secured 212 MW of capacity for 2027 across Australia and New Zealand, about four times what it had at the start of 2026\. Of that, 129 MW is contracted to customers. It plans to deploy more than 68,000 Nvidia GPUs across 2026 and 2027, mainly B300 and GB300 systems. It is a Nvidia Preferred Partner, which it says gives it priority access to those chips.

## How does Sharon AI make money?

Mainly through take-or-pay contracts. The customer reserves a set amount of capacity for several years and pays for it whether or not it uses every hour. Our explainer on [take-or-pay GPU contracts](https://www.sotkn.com/explained/take-or-pay-gpu-contracts-explained/) covers how that protects revenue once a cluster is live. The key word is live. Payment starts at acceptance, not at signing.

The customer list is short and mostly unnamed. The results presentation lists seven deals. Canva and GMI together are worth US$0.04 billion. ESDS, an Indian cloud company, signed US$1.25 billion over five years. A global technology company signed US$0.95 billion over five years. A global AI lab signed US$1.32 billion, a global services company US$1.00 billion, and a US-based AI inference provider US$0.17 billion over four years.

The biggest single item is Nvidia. Under its AI Compute Program, Nvidia has signed a six-year take-or-pay agreement covering 72 MW and about 40,000 GB300 GPUs, which Sharon AI counts at US$4.88 billion. That figure is the floor price. Sharon AI plans to resell the capacity to end customers on shorter deals and keep any upside above the floor, shared with Nvidia. Our explainer on the [Nvidia cloud partner](https://www.sotkn.com/explained/nvidia-cloud-partner-explained/) programme shows how Nvidia backs the clouds that buy its chips.

Prices have risen. The chief executive, James Manning, said long-term contract prices rose this year from about US$15 million to more than US$19 million of annual revenue per MW, with new customers prepaying up to 30% of contract value or 60% of the hardware cost. He expects shorter resale deals under the Nvidia program at over US$30 million per MW. Our explainer on [GPU rental prices](https://www.sotkn.com/explained/gpu-rental-price-explained/) shows why short deals pay more per hour than long ones.

Add it up and you get the headline: about US$9.6 billion of total contract value, over a weighted average of 5.5 years. Our explainer on [total contract value](https://www.sotkn.com/explained/total-contract-value-explained/) explains why that number can run years ahead of revenue. Sharon AI's own footnote says it is an operating metric, not revenue, and can change if contracts are changed or ended.

Then there is the margin. The company says a project earns more than 80% contribution margin at full deployment. That is before head office costs, interest, tax and depreciation. GPUs lose value fast, so depreciation takes a large share of that margin over five years. Our explainers on [GPU depreciation](https://www.sotkn.com/explained/gpu-depreciation-explained/) and [capex versus depreciation](https://www.sotkn.com/explained/capex-vs-depreciation/) walk through why.

## Why is revenue so small next to the contracts?

![What does Sharon AI do: 212 MW of capacity secured for 2027, 129 MW contracted to customers, and about 30 MW expected to be billing from January 2027](https://storage.ghost.io/c/cd/10/cd10e222-131f-4ba4-b4a9-ca4214e8c12d/content/images/2026/11/sharon-ai-capacity-vs-billing.png)

Sharon AI capacity, contracts and expected billing. Second Order chart from the company's 3Q 2026 results.

Because almost none of it is live yet. In the quarter to 30 September, revenue was US$9.9 million, up from US$0.5 million a year earlier. Analysts had expected about US$26.9 million, Benzinga reported. Net loss was US$43.8 million. Selling, general and administrative costs alone were US$26.3 million, more than two and a half times revenue on our arithmetic.

The company's own run-rate figure tells the same story. It puts annual run-rate revenue at about US$50 million for September 2026\. It expects US$400 million to US$450 million for January 2027, an 8 to 9 times jump, once two big clusters are accepted. Our explainer on [annual run-rate revenue](https://www.sotkn.com/explained/arr-explained/) and the companion on [ARR versus revenue](https://www.sotkn.com/explained/arr-vs-revenue/) show why a run-rate is a snapshot, not a forecast.

Even that jump covers only part of what is signed. The company says about 30 MW of its 129 MW of contracted capacity is expected to be billing from January 2027\. That is about 23%. The rest follows through 2027, with the Nvidia program last: its data center comes online in the first quarter of 2027, testing runs in the second quarter, and billing starts in the third quarter.

The balance sheet shows the same gap in hardware. At 30 September, Sharon AI held US$1.39 billion of construction in progress and US$221 million of deposits on equipment, against US$89 million booked as technology equipment. On our arithmetic, construction in progress was more than fifteen times the technology equipment line. Most of the hardware money is not earning yet.

## Where did the money come from?

Mostly from investors, and fast. Sharon AI says it has secured about US$2.6 billion since December 2025\. The presentation lists the pieces: US$104 million of convertible notes in December 2025, US$74 million from selling its stake in a Texas data center venture in January, US$125 million from the Nasdaq listing in February, US$350 million of convertible notes in April, a US$1.6 billion strategic financing in June, and a US$356 million GPU-backed loan in October.

Customers are paying in too. The company counts about US$0.8 billion of customer deposits as it deploys. Its footnote says these deposits are typically held in escrow until the customer accepts the cluster. So that money is promised and visible, but locked behind the same gate as revenue.

The October loan shows the template for what comes next. It is up to US$356 million, signed on 28 September, with Goldman Sachs and private credit funds as lenders. It has two parts: Facility A of up to US$150 million at 7.25% rising to as much as 9.95%, and Facility B of up to US$206 million at 9.95%. Both are repaid in one lump sum after 42 months. The GPUs and their contract cash flows are the security, held in a separate company. Our explainer on [GPU financing](https://www.sotkn.com/explained/gpu-financing-explained/) covers how these loans work.

One detail matters most. Sharon AI itself guarantees the loan only until certain events occur, and the presentation describes it as a temporary parent guarantee before customer acceptance. In plain words: lenders want the parent on the hook until the customer has signed off. After that, the cluster's own cash flows carry the debt. Our Analysis on [chip financing vehicles](https://www.sotkn.com/analysis/when-spv-chip-leasebacks-open-where-cash-sticks/) shows why this structure is spreading.

More loans are planned. The chief financial officer, Anuj Goel, said further GPU-backed facilities are well advanced and that he expects more than 75% of the secured debt to come from the Nvidia program. The 10-Q is blunter: if that debt is not raised on acceptable terms, the company would need other funding to buy the GPUs for its ESDS and AI lab contracts.

## Who gets paid first?

- **Nvidia and the server makers.** The chip and server chain is paid before installation. Sharon AI paid US$566 million in advances for property and equipment in the first nine months of 2026, US$293 million of it in the third quarter. Nvidia designs the GPUs and is also a customer, through its AI Compute Program, which makes it both supplier and anchor buyer.
- **Landlords and builders.** The data halls belong to partners. Rent and service fees start when a hall is handed over ready for service, before the GPUs inside earn. Sharon AI paid about US$60 million of upfront fees on leased space in the first nine months of 2026\. Our Markets piece on [Sydney data center power and leases](https://www.sotkn.com/markets/australia/australia-sydney-data-centre-power-lease-october-2026/) shows the landlord's side.
- **Lenders and note holders.** Interest on the October loan is paid monthly in cash. Sharon AI also carried about US$1.0 billion of convertible notes at 30 September. Interest expense and finance costs were about US$7.9 million in the third quarter.
- **Suppliers waiting for payment.** Trade and other payables were US$1.24 billion at 30 September, up from US$3.4 million at the end of 2025\. That is the build bill still owed.
- **Sharon AI, last.** It is paid when each cluster is accepted and billing starts. Only after chips, rent, power, interest and staff does it keep a margin.

This is the usual order on the neocloud layer, and it is why the timing of sign-off matters so much. Our explainer on [who finances GPU clusters](https://www.sotkn.com/explained/who-finances-gpu-clusters/) lays out the lenders, vendors and equity holders who fill the gap.

## What is the stuck step?

Customer acceptance. Not demand: the contracts are signed. Not power, on the company's account: power is onsite at all 12 sites with contracted customers, with no need for new grid connections. Not chips: about 8,000 B300 and 4,000 GB300 GPUs have been energised. What is missing is the customer's sign-off that each cluster meets the contract. Our explainer on [GPU cluster acceptance](https://www.sotkn.com/explained/gpu-cluster-acceptance-explained/) walks through what that test involves.

The presentation sets out five steps for each deal: the data hall is ready, the GPUs are installed, testing runs, the customer accepts, and billing begins. For the global technology customer, the first tranche, about 20% of its GPUs, was accepted in August 2026\. The second tranche, about 80%, is in testing, with acceptance due in December and billing in January 2027\. ESDS is on the same timetable. Testing includes Nvidia's own verification of the cluster, which the company calls NVIS.

The 10-Q explains why this step carries teeth. Each day of delay past the handover date can cut the paid service period, which directly lowers the lifetime revenue of that contract. If acceptance testing is not done by set deadlines, the customer may be able to end the contract. The filing calls execution risk extremely high in 2026 and 2027.

Our read on the stuck step: the money, the halls and the chips are mostly in place, and the gate is now a dated test. We mark the status Holding. The two big clusters have a clear December to January timetable, and Sharon AI holds US$1.34 billion of cash, so the wait is affordable for now. It would turn to Tightening if those dates slip. Our Analysis on [cash owed before a cluster is accepted](https://www.sotkn.com/analysis/if-276-billion-owed-before-cluster-accepted-where-cash-sticks/) shows the same gate at a larger scale.

## Who else does this?

The closest models are other neoclouds that rent Nvidia capacity on long contracts. [CoreWeave](https://www.sotkn.com/stack/neoclouds/what-does-coreweave-do/) is the largest and listed; [Nebius](https://www.sotkn.com/stack/neoclouds/what-does-nebius-do/) builds its own sites; [Lambda](https://www.sotkn.com/stack/neoclouds/what-does-lambda-do/) sells to AI labs and has borrowed against its GPUs. [Fluidstack](https://www.sotkn.com/stack/neoclouds/what-does-fluidstack-do/) leans on partners' halls, like Sharon AI. Our [Wire on Lambda's GPU loan](https://www.sotkn.com/wire/lambda-closes-1b-investment-grade-gpu-loan-wire/) is a useful comparison for the debt side.

The nearest neighbour is [Firmus](https://www.sotkn.com/stack/neoclouds/what-does-firmus-do/), another Australian neocloud. Firmus's problem is different: it pulled its float and must find cash for chips. Sharon AI raised its money and now waits on sign-off. The Australian landlord side is covered on our [NEXTDC page](https://www.sotkn.com/stack/ai-data-centers/what-does-nextdc-do/) and in our Markets piece on [Australian AI data center funding](https://www.sotkn.com/markets/australia/australia-ai-data-centre-funding-october-2026/). Our explainer on [neoclouds versus hyperscalers](https://www.sotkn.com/explained/neocloud-vs-hyperscaler/) explains why AI labs rent from firms like Sharon AI at all.

## What would change this view?

- **Easing:** Sharon AI confirms acceptance and billing for the ESDS and global technology clusters by January 2027, and signs the next GPU-backed loans on terms close to the October facility.
- **Tightening:** acceptance slips past January, a customer cuts its service period or uses a termination right, or the next loans do not close and equity has to fill the gap.
- **A model shift:** Nvidia's program becomes most of the business. That would make Nvidia supplier, anchor customer and the base for most of the debt at once, which concentrates the risk in one partner.

## What should you watch next?

Four dates. First, December 2026: acceptance of the second global technology tranche and of ESDS. Second, January 2027: whether the run-rate actually reaches US$400 million to US$450 million. Third, the next GPU-backed loan, its size and its rate against the October 9.95%. Fourth, the Nvidia program: data center online in the first quarter of 2027, testing in the second, billing from the third. Our explainer on [revenue backlog versus RPO](https://www.sotkn.com/explained/revenue-backlog-vs-rpo/) is a guide to reading the contract numbers as they turn into revenue.

## Practical takeaways

Read Sharon AI as a renter of Nvidia chips with signed customers, powered halls and money raised, waiting on the last test. The question is no longer whether it can get the inputs. It is whether customers sign off on time. Our Analysis on [signed but not billing](https://www.sotkn.com/analysis/nextdc-signed-not-billing-stuck-step-what-price-assumes/) shows the same gap from an Australian landlord's side.

## FAQ

**What does Sharon AI do?**  
Sharon AI is an Australian neocloud listed on Nasdaq under the ticker SHAZ. It buys Nvidia GPU systems, installs them in partner data centers in Australia and New Zealand, and rents the computing to AI labs, tech firms and cloud companies, mostly on five-year take-or-pay contracts.

**Who are Sharon AI's customers?**  
Named customers include Nvidia, Canva, GMI and the Indian cloud firm ESDS. The rest are unnamed in its filings: a global technology company, a global AI lab, a global services company and a US-based AI inference provider.

**How much revenue does Sharon AI make?**  
Revenue was US$9.9 million in the quarter to 30 September 2026, with a net loss of US$43.8 million. The company put its annual run-rate revenue at about US$50 million for September 2026 and expects US$400 million to US$450 million for January 2027 if its next clusters are accepted on time.

**What is Sharon AI's total contract value?**  
About US$9.6 billion as of November 2026, over a weighted average of 5.5 years. Close to half, US$4.88 billion, is a six-year arrangement with Nvidia's AI Compute Program, counted at its floor price. Total contract value is a company metric, not revenue.

**Why is customer acceptance so important for Sharon AI?**  
Its contracts start paying when a customer confirms that the installed cluster meets the contract specification. Customer deposits sit in escrow until then, and the 10-Q says late acceptance can shorten the paid service period or, in severe cases, let a customer end the contract.

**Is this a recommendation to invest in Sharon AI?**  
No. This page explains how Sharon AI makes money and where it can get stuck. It gives no buy or sell calls. Not investment advice. Do your own research.

## Keep reading on the map

Layer: [Neoclouds](https://www.sotkn.com/stack/neoclouds/) on [The Stack](https://www.sotkn.com/stack/). Also on the layer: [Corvex](https://www.sotkn.com/stack/neoclouds/what-does-corvex-do/) and [Firmus](https://www.sotkn.com/stack/neoclouds/what-does-firmus-do/). Explained: [what a neocloud is](https://www.sotkn.com/explained/neocloud-explained/) and [sovereign AI](https://www.sotkn.com/explained/sovereign-ai-explained/). Wire: [Nvidia's neocloud commitments](https://www.sotkn.com/wire/nvidia-56b-neocloud-commitments-shakeout-wire/).

[Get the Wire](https://www.sotkn.com/#/portal/signup) — short daily notes when compute, chips and GPU financing move.

Not investment advice. Do your own research.

**How we check this**

This page rests on Sharon AI's own documents, all read on 21 November 2026: its third-quarter results release and presentation of 16 November, its Form 10-Q for the quarter to 30 September, and its Form 8-K of 1 October on the GPU-backed loan. The analyst revenue estimate comes from Benzinga. Total contract value and annual run-rate revenue are company metrics, not revenue under US accounting rules, and the company says so. Delivery dates, the January 2027 run-rate and the 2027 capacity are company targets. Most customers are not named. Sums and ratios are our arithmetic and are labelled as ours. Who gets paid and the stuck step are our read. The featured image is a graded photograph of a test bench at a US supercomputer build, not Sharon AI hardware. This page explains a business. It gives no buy or sell calls. Not investment advice. Do your own research.

**Last reviewed:** 21 Nov 2026 · **Next review:** after Sharon AI reports customer acceptance of the ESDS and global technology clusters (due December 2026 to January 2027), and no later than its full-year 2026 results

Spot an error? [Tell us](mailto:sean@thekopinotes.com) and we will correct it and note the change here.

[Our editorial standards →](https://www.sotkn.com/editorial-standards/)

## References

1. [Sharon AI — Sharon AI Reports Third Quarter 2026 Results (16 Nov 2026)](https://sharonai.com/press-releases/sharon-ai-reports-third-quarter-2026-results/?ref=sotkn.com) — Q3 revenue, loss and balance sheet; US$9.6bn TCV; ARR US$50m to US$400–450m; 212 MW; deposits in escrow until acceptance; pricing per MW
2. [Sharon AI — 3Q 2026 Results Presentation (16 Nov 2026)](https://sharonai.com/media/2026/11/Sharon-AI-3Q26-Earnings-Call-Presentation-FINAL.pdf?ref=sotkn.com) — contract list and price per MW; delivery steps and dates; NVIDIA AI Compute Program terms; capital raised since December 2025; capacity by site status
3. [SharonAI Holdings — Form 10-Q for the quarter to 30 September 2026 (SEC, filed 16 Nov 2026)](https://www.sec.gov/Archives/edgar/data/2068385/000149315226052382/form10-q.htm?ref=sotkn.com) — property and equipment split; risk factors on delivery delays, acceptance deadlines and termination; customer concentration; financing risk
4. [SharonAI Holdings — Form 8-K on the US$356m secured facility (SEC, filed 1 Oct 2026)](https://www.sec.gov/Archives/edgar/data/2068385/000149315226045368/form8-k.htm?ref=sotkn.com) — Facility A and B sizes and rates, 42-month bullet, parent guarantee with automatic release, security package
5. [Benzinga — SharonAI posts Q3 double miss despite strong customer momentum (Nov 2026)](https://www.benzinga.com/markets/earnings/26/11/63324567/sharonai-posts-q3-double-miss-despite-strong-customer-momentum?ref=sotkn.com) — Q3 revenue of US$9.93m against a US$26.94m analyst estimate

S

**[Sean](https://www.sotkn.com/author/sean/)**

Writes Second Order by TKN — a plain-English map of AI infrastructure and semiconductors for non-specialist investors. Focus: who gets paid, where money sticks, and which step is stuck.

Covers AI infrastructure, semiconductors, and second-order stack reads.

[Author page →](https://www.sotkn.com/author/sean/) · [Editorial standards →](https://www.sotkn.com/editorial-standards/)