NVIDIA’s Indonesia “AI Compute Partnership” points to a new way to finance GPU buildouts

NVIDIA is pushing beyond hardware sales with a financing concept designed to help cloud builders deploy large GPU clusters while aligning NVIDIA’s revenue to the future usage of those chips. The approach, described in recent reporting and linked to a broader set of “campus” deals, is gaining attention as companies race to scale data centers for AI workloads in Southeast Asia and elsewhere.

In separate announcements and coverage, NVIDIA’s model has been tied to neocloud providers that plan to procure major volumes of NVIDIA accelerators for multi-year operations. One highlighted project is Firmus Technologies, an Australian AI infrastructure startup, which has been reported to be building a 360 megawatt (MW) AI data center in Batam, Indonesia, with a target scale of up to 170,000 NVIDIA GPUs in the 2027 to 2028 window.

Reporting on the structure says NVIDIA would act as a financial backstop for GPU deployments, with the program designed so NVIDIA can recapture economics from the cloud services generated by the hardware. The concept is being described as a “revenue-sharing and credit-support” model, meaning NVIDIA can potentially earn not only from selling GPUs but also from the downstream revenue generated when the installed capacity is actually used to deliver cloud compute.

The same reporting points to NVIDIA’s first two “adopters” of the program as Sharon AI, an Australian sovereign-cloud provider planning up to 40,000 Grace Blackwell GB300 GPUs, and Firmus Technologies in Indonesia. In the coverage, NVIDIA is said to have declined to provide the specific revenue-share percentage and other detailed terms beyond what it described publicly in connection with the program.

For the industry, the appeal is straightforward: mega-scale data center projects often hinge on financing risk, particularly in the early years when utilization, customer demand, and power costs may be uncertain. A financing arrangement that reduces the downside for providers, while sharing in upside when GPUs produce billable compute, can lower barriers to building at scale and accelerate timelines for AI capacity.

For NVIDIA, the strategic shift is also notable. If usage-linked economics become meaningful, the company’s financial exposure could move closer to a “platform” profile rather than staying limited to one-time equipment purchases. That is consistent with the broader market interest in how major chip suppliers will participate in the economics of AI infrastructure, not just the procurement cycle for accelerators.

Still, key details remain unclear from the publicly available reporting. The exact pricing mechanics, the duration of any “backstop” obligations, the treatment of idle or underutilized GPUs, and the precise calculation of usage-linked revenue are not fully disclosed in the accounts currently circulating. The reported Indonesia and Australia targets also leave room for changes as power availability, permitting, supply timing, and customer onboarding affect deployment milestones.

What to watch next is whether NVIDIA offers additional specifics about the financing program, including how widespread it becomes with other cloud builders, and whether Firmus’ Indonesian facility progresses on schedule toward the reported GPU capacity targets. Analysts and market participants will likely look for corroborating disclosures from the partners themselves, such as project timelines, procurement updates, and any public documentation of how the revenue-sharing terms work. Meanwhile, traders may also watch near-term sentiment around NVIDIA’s earnings model as the company continues to market new ways to tie its chips to AI compute monetization.