Sharon AI Signs Five-Year Rafay Systems Agreement to Scale AI Infrastructure

Sharon AI has signed a five-year strategic agreement with Rafay Systems to standardize how the company provisions, governs, monitors and delivers access to computing capacity across its expanding AI Factory network.

Rafay’s software will serve as a centralized orchestration and operations layer above Sharon AI’s physical infrastructure. The arrangement is intended to replace a more fragmented, cluster-by-cluster operating model with a common framework spanning locations, customer tenants and different types of AI workloads.

The September 4 announcement says the architecture established through the agreement is designed to support orchestration of as many as 150,000 GPUs over the five-year term. That figure describes the capacity of the planned operating architecture, not a commitment by Sharon AI to purchase, deploy or place 150,000 GPUs into service.

Rafay becomes a common control layer across Sharon AI’s AI Factories

At the center of the agreement is an operational problem that becomes more difficult as GPU infrastructure spreads across sites and customers. Sharon AI said Rafay will provide infrastructure lifecycle management, workload orchestration, governance, monitoring and secure multi-tenancy through a unified platform that supports Kubernetes and virtual-machine environments.

In practical terms, that gives Sharon AI a software layer for controlling how compute resources are configured and made available without requiring each environment to be managed as an isolated deployment. Rafay describes its platform as a control plane that can orchestrate GPUs, bare-metal servers, virtual machines, Kubernetes and other infrastructure while applying common policies, usage visibility and automation.

The companies also expect the arrangement to help Sharon AI build internal expertise around Rafay and establish more consistent operating practices across future deployments. Sharon AI said the platform should improve automation, infrastructure utilization and reliability while giving customers a more standardized way to access compute resources. Those are company expectations rather than demonstrated results from the five-year agreement, which has only just been announced.

Financial terms were not disclosed. Sharon AI furnished the press release to the U.S. Securities and Exchange Commission in a Form 8-K dated September 4 under Item 7.01, Regulation FD Disclosure, along with Item 9.01 for the exhibit. The filing did not include an Item 1.01 disclosure for entry into a material definitive agreement. The company also stated that the release was furnished rather than deemed filed for purposes of Section 18 of the Securities Exchange Act.

The 150,000-GPU figure is a scaling ceiling, not a deployment target

The largest number in the announcement is also the one that requires the most careful reading. Sharon AI said the architecture is designed to support orchestration of up to 150,000 GPUs during the five-year term. The release does not say that Sharon AI has ordered that many chips, secured data-center power for that total, signed customer contracts requiring that level of compute, or set a timetable for reaching it.

Instead, the figure indicates the scale that the software and operating framework are being designed to accommodate if Sharon AI’s physical footprint continues to expand. That distinction matters because orchestrating capacity is different from financing, procuring and installing the underlying servers, networking, storage and electrical infrastructure needed to operate a GPU fleet.

Sharon AI’s currently disclosed buildout is smaller. In its second-quarter results released August 6, the company said it had secured 212 megawatts of AI Factory capacity and expected to have more than 64,000 NVIDIA GPUs deployed by mid-2027. The same update said its VAST Data deployment was sized to support the data needs of roughly 100,000 GPUs. Those figures show why the company is investing in a control layer capable of operating beyond its near-term deployment schedule, but none of them establishes that a 150,000-GPU fleet will be built.

The Rafay agreement therefore addresses a different part of the expansion challenge. Sharon AI still needs the physical compute, power, data-center space, networking and customer demand that ultimately determine deployed capacity. Rafay’s role is to help manage and expose those resources through a consistent operational system once they are available.

The software agreement sits inside a much larger infrastructure buildout

Sharon AI has been expanding rapidly through a series of infrastructure and customer agreements. Its August 6 second-quarter update put total contract value at about $8.8 billion as of that date and said revenue was expected to ramp materially from the third quarter of 2026 through 2027. The company also reported $1.9 billion of cash and cash equivalents at June 30, providing context for the capital-intensive buildout it is pursuing.

Recent commitments include a six-year compute collaboration with NVIDIA covering 72 megawatts of new data-center capacity in Australia and up to 40,000 Grace Blackwell GB300 GPUs. Sharon AI has also announced a five-year, $1.32 billion cloud agreement tied to a New Zealand deployment and a separate five-year, $373 million cloud agreement in Australia whose initial deployment is expected to use 2,048 NVIDIA Blackwell Ultra B300 GPUs.

Those agreements are primarily about acquiring, deploying or selling access to compute capacity. The Rafay arrangement is different because it concerns the operating layer that sits above that hardware. As the number of clusters, locations and customers increases, a common system for provisioning resources, enforcing policies and monitoring usage becomes more important to maintaining a consistent service.

Rafay’s own platform materials emphasize multi-tenancy, policy controls, self-service access and orchestration across several infrastructure types. For Sharon AI, that means the software can potentially serve customers running different workload models without forcing the company to build separate operational tooling for every deployment. The agreement also gives Sharon AI a framework that can extend across Australia, New Zealand and other markets as its footprint grows.

The announcement does not change Sharon AI’s previously disclosed near-term GPU deployment targets or provide a new revenue contribution tied specifically to Rafay. Its significance is operational: Sharon AI is putting a standardized management layer in place before its planned physical infrastructure reaches the scale envisioned in the five-year architecture.

Execution now depends on how effectively that platform is rolled out across existing and new AI Factory environments as Sharon AI brings contracted capacity online. The company’s August results said more than 64,000 NVIDIA GPUs were expected to be deployed by mid-2027 and that revenue should ramp through 2027, providing the next measurable milestones against which the broader infrastructure expansion can be assessed.