Sharon AI signs NVIDIA deal worth up to US$4.88 billion for 72MW in Australia

At a glance

• Sharon AI (NASDAQ:SHAZ) announced a six-year AI infrastructure compute collaboration with NVIDIA (NASDAQ:NVDA) on 12 June 2026, enabling 72MW of new data centre capacity in Australia.

• Sharon AI’s 8-K puts the contract value at up to US$4.88 billion, under a Master Cloud Services Agreement and Order Form 1 dated 8 June 2026.

• The companies will deploy NVIDIA’s DSX AI factory design, scaling up to 40,000 Grace Blackwell GB300 GPUs for AI startups, enterprises and university researchers.

• The agreement uses a revenue-sharing and credit-support model: Sharon AI sells NVIDIA-powered cloud services, and NVIDIA earns standard product revenue plus a share of cloud revenue on the supported capacity.

• Sharon AI’s total AI factory capacity rises to 132MW, of which 102MW is contracted to end customers. The company expects more than 55,000 total NVIDIA GPUs deployed by mid-2027.

• The 132MW total exceeds the 100MW-by-early-2027 target Sharon AI set in its Q1 results on 15 May 2026, four weeks ago.

• SHAZ shares rose about 25 per cent following the announcement.

What Sharon AI announced

Sharon AI signed a six-year AI infrastructure compute collaboration with NVIDIA, announced on 12 June 2026 in New York. Under the agreement, the two companies will enable 72MW of new data centre capacity in Australia, deploying NVIDIA’s DSX AI factory design and scaling up to 40,000 Grace Blackwell GB300 GPUs to serve AI startups, enterprises and university researchers.

The press release carries no dollar figure, but the accompanying 8-K filed with the SEC does: the collaboration sits under a Master Cloud Services Agreement and Order Form 1 dated 8 June 2026, with a contract value of up to US$4.88 billion. For comparison, Sharon AI’s entire contracted book stood at US$2.2 billion at its Q1 results four weeks ago.

James Manning, Sharon AI’s co-founder and chief executive, called the agreement “a pivotal moment in Sharon AI’s mission to deliver sovereign, large-scale AI compute infrastructure”, adding that the 72MW provides “access to accelerated compute to enterprise, startup and AI native customers who otherwise may not have been able to access it.”

The release frames the collaboration as expanding Sharon AI’s established position as a certified NVIDIA Cloud Partner and its existing footprint inside Australian data centres. That footprint is anchored by Sharon AI’s sovereign GPU cluster at NEXTDC M3 in Melbourne, where the company’s 50MW lease represents roughly 22 per cent of Australia’s largest Tier IV facility.

How the revenue-sharing and credit-support model works

Sharon AI commits to large-scale NVIDIA infrastructure, and NVIDIA supports that commitment with credit while taking two income streams: standard product revenue on the hardware, plus a share of the cloud revenue generated on the supported capacity.

The release describes the outcome for each side plainly. Sharon AI gets a capital-efficient path to scale. NVIDIA gets a recurring, usage-linked earnings stream on top of the hardware sale. Customers who historically lacked access to capital-intensive AI infrastructure get a route to GB300-class compute.

For an operator that closed a US$350m convertible note offering on 20 May to fund GPU and network procurement, following the Oaktree-led raise we explained in April, vendor credit support changes the funding equation. Less of the GPU bill needs to be pre-funded from Sharon AI’s own balance sheet, and the vendor’s return is tied to the same utilisation outcomes the operator is chasing.

Chips ship in months, data centres take years

The structure also reflects a timing problem the chip maker cannot solve alone. NVIDIA now turns out a new chip generation roughly every year, and a GPU order ships in months. A data centre is a three-to-five-year object: site, planning approval, grid connection, shell, fit-out. In Australia, grid connection alone carries an approximate two-year target from application to energisation, on AEMO’s own connection pipeline numbers. The bottleneck has moved from silicon to power and floor space, and NVIDIA’s growth is now gated by other people’s construction schedules. A chip that sits in a warehouse waiting for a hall earns nobody anything, and it delays the customer’s next order.

The capacity: from 60MW to 132MW in one agreement

Sharon AI’s disclosed AI factory capacity now totals 132MW. The 72MW addition implies a prior base of 60MW, and the new total passes the 100MW-by-early-2027 target the company set in its Q1 results just four weeks ago, a target we examined in our 100MW multi-site read-through. Of the 132MW, 102MW is contracted to end customers.

The announcement caps a five-month run of disclosures:

Source: Certified Strategic Editorial, company disclosures, June 2026.

On GPU count, the company expects more than 55,000 total NVIDIA GPUs deployed by mid-2027. For scale, the Vera Rubin launch cohort counted Sharon AI among the Australian-founded operators scaling Blackwell-class capacity ahead of NVIDIA’s next platform generation.

Where the 72MW could land

The release does not name sites. It does specify Australia, which matters: Sharon AI’s existing deployments sit inside NEXTDC facilities, and NEXTDC’s network spans Australian sites and the recently opened KL1 in Kuala Lumpur. The Australia framing points the new capacity at domestic facilities.

Our May read-through of Sharon AI’s deployment options identified M3 expansion, NEXTDC S3 in Sydney and the planned M4 Melbourne campus as the candidates with available power and Tier IV credentials. The 12 June release gives no allocation, and Sharon AI’s next quarterly results and ASX listing documents are the likely venues for site-level detail.

What the investment panels at DCD Connect APAC were asking

Certified Strategic attended DCD Connect APAC in Bali this week, our first time at the region’s flagship data centre gathering, from 9 to 11 June. A full recap is coming. The mood on neoclouds in the investment sessions was bullish. Capital from private credit through to private equity is looking for a way into the sector, panellists described demand that has outrun their own year-ago expectations, and successful exits across Asia-Pacific platforms were tipped on stage inside the next 12 to 18 months. The open questions were structural: how to price contracts signed with counterparties that did not exist three years ago, who carries the technology risk on a GPU cluster’s revenue in its fifth year, and how platforms convert contracted capacity into an exit. A full recap of the three days is coming.