• Nvidia provides hardware upfront — GPUs, networking, and the full Nvidia DSX (Data Center AI Factory) reference design, including liquid cooling, software stack, and architectural blueprints .
• Cloud partners build, operate, and sell AI compute services to their customers using that infrastructure .
• Nvidia earns two revenue streams: standard product revenue at the time of deployment plus a recurring share of the cloud revenue generated by customer usage over the multi-year partnership term .

In effect, Nvidia is moving from a one-time GPU sale to a "GPU-as-a-Service" or co-investment model, where its returns are directly tied to how much AI compute its partners successfully sell.

Sharon AI: $4.88 Billion for 72 MW of AI Capacity in Australia

The most detailed public example is Sharon AI (NASDAQ: SHAZ), which announced a six-year strategic compute collaboration with Nvidia on June 12, 2026 . The key terms, confirmed by an SEC 8-K filing, are:

• Contract value: Up to $4.88 billion under a Master Cloud Services Agreement with Order Form 1 dated June 8, 2026 .
• Location & scale: 72 megawatts (MW) of new data center capacity in Australia, bringing Sharon AI's total planned AI factory capacity to 132 MW .
• GPUs: Up to 40,000 Nvidia Grace Blackwell GB300 GPUs .
• Financial structure: Revenue-sharing and credit-support model. Sharon AI does not buy the chips outright; Nvidia and lenders fund much of the hardware in return for a share of the cloud revenue .
• Already contracted: 102 MW of capacity is already committed to end customers .

Sharon AI sells Nvidia-powered cloud services, and Nvidia earns both standard product revenue and a portion of the cloud revenue on that supported capacity . The filing notes significant execution risks, including tight delivery timelines and financing needs .

Firmus Technologies: $30 Billion, 170,000 GPUs in Indonesia

Just over two weeks later, on June 28, 2026, Firmus Technologies announced a strategic partnership with Nvidia running through 2034 — an 8+ year commitment . The headline numbers are striking:

• Campus scale: A dedicated 360 MW Nvidia DSX AI Factory campus in Batam, Indonesia, just off the coast of Singapore .
• GPUs: Up to 170,000 Nvidia GPUs .
• Expected off-take: Up to $30 billion in committed off-take agreements in the first six years .
• Financial structure: Firmus procures Nvidia hardware under the same revenue-sharing and credit-support DSX framework, targeting "global AI-natives" with cost-effective high-performance compute .

Firmus and Singapore-based DayOne will develop the Batam campus. The project is part of a broader push to provide compute capacity outside the dominant hyperscale clouds .

Other Known Partners

Nvidia has also announced similar arrangements with:

• Naver (South Korea): Building gigawatt-scale DSX-based AI infrastructure, starting at 55 MW and scaling to 200 MW by 2028, with a long-term target of 1 GW .
• SK Telecom (South Korea): Entering the DSX-based AI cloud business, expanding from Korea across Asia .
• IREN: Partnered with Nvidia to operationalize the DSX architecture at gigawatt scale, with IREN owning the physical, data, and AI layers . IREN signed a five-year, $3.4 billion AI cloud contract with Nvidia for Blackwell GPU deployments across 60 MW of data center capacity .

What This Means for Nvidia's Strategy

These partnerships reveal a deliberate shift in how Nvidia approaches the AI infrastructure market. The company is moving from selling individual chips to financializing its hardware advantage across several layers:

• From one-time sales to recurring revenue. Instead of selling a GPU once, Nvidia collects a portion of every AI token sold on that hardware for 6 to 10 years . This smoothes revenue cycles and ties growth to actual AI consumption, not just data center construction.

• From component supplier to platform architect. Nvidia provides the full DSX AI factory blueprint — liquid cooling designs, networking (NVLink/NvSwitch/InfiniBand), CUDA and AI Enterprise software, and reference architectures . Partners effectively operate Nvidia-branded factories, giving Nvidia control over the reference architecture even when it does not own the building.

• Layered financial exposure. Nvidia now has:

  1. Upfront product revenue from selling GPUs into these factories.
  2. Recurring cloud revenue share from usage over multi-year terms.
  3. Credit-support and financing exposure — by co-investing, Nvidia takes partial balance-sheet risk in its partners' success .
  4. Ecosystem lock-in — deeper financial entanglement makes it harder for competitors like AMD or Intel to break into those data centers.

• Scaling beyond the hyperscalers. By financing smaller AI cloud providers, Nvidia expands the addressable market beyond AWS, Azure, and GCP. These partners target AI startups and enterprises that may lack access to hyperscale cloud GPU capacity .

In short, Nvidia is betting that its balance sheet and product leverage can capture value at every layer of the AI compute stack — not just at the point of chip sale, but across years of usage, cloud service margins, and ecosystem adoption.

• Nvidia provides hardware upfront — GPUs, networking, and the full Nvidia DSX (Data Center AI Factory) reference design, including liquid cooling, software stack, and architectural blueprints .
• Cloud partners build, operate, and sell AI compute services to their customers using that infrastructure .
• Nvidia earns two revenue streams: standard product revenue at the time of deployment plus a recurring share of the cloud revenue generated by customer usage over the multi-year partnership term .

In effect, Nvidia is moving from a one-time GPU sale to a "GPU-as-a-Service" or co-investment model, where its returns are directly tied to how much AI compute its partners successfully sell.

Sharon AI: $4.88 Billion for 72 MW of AI Capacity in Australia

The most detailed public example is Sharon AI (NASDAQ: SHAZ), which announced a six-year strategic compute collaboration with Nvidia on June 12, 2026 . The key terms, confirmed by an SEC 8-K filing, are:

• Contract value: Up to $4.88 billion under a Master Cloud Services Agreement with Order Form 1 dated June 8, 2026 .
• Location & scale: 72 megawatts (MW) of new data center capacity in Australia, bringing Sharon AI's total planned AI factory capacity to 132 MW .
• GPUs: Up to 40,000 Nvidia Grace Blackwell GB300 GPUs .
• Financial structure: Revenue-sharing and credit-support model. Sharon AI does not buy the chips outright; Nvidia and lenders fund much of the hardware in return for a share of the cloud revenue .
• Already contracted: 102 MW of capacity is already committed to end customers .

Sharon AI sells Nvidia-powered cloud services, and Nvidia earns both standard product revenue and a portion of the cloud revenue on that supported capacity . The filing notes significant execution risks, including tight delivery timelines and financing needs .

Firmus Technologies: $30 Billion, 170,000 GPUs in Indonesia

Just over two weeks later, on June 28, 2026, Firmus Technologies announced a strategic partnership with Nvidia running through 2034 — an 8+ year commitment . The headline numbers are striking:

• Campus scale: A dedicated 360 MW Nvidia DSX AI Factory campus in Batam, Indonesia, just off the coast of Singapore .
• GPUs: Up to 170,000 Nvidia GPUs .
• Expected off-take: Up to $30 billion in committed off-take agreements in the first six years .
• Financial structure: Firmus procures Nvidia hardware under the same revenue-sharing and credit-support DSX framework, targeting "global AI-natives" with cost-effective high-performance compute .

Firmus and Singapore-based DayOne will develop the Batam campus. The project is part of a broader push to provide compute capacity outside the dominant hyperscale clouds .

Other Known Partners

Nvidia has also announced similar arrangements with:

• Naver (South Korea): Building gigawatt-scale DSX-based AI infrastructure, starting at 55 MW and scaling to 200 MW by 2028, with a long-term target of 1 GW .
• SK Telecom (South Korea): Entering the DSX-based AI cloud business, expanding from Korea across Asia .
• IREN: Partnered with Nvidia to operationalize the DSX architecture at gigawatt scale, with IREN owning the physical, data, and AI layers . IREN signed a five-year, $3.4 billion AI cloud contract with Nvidia for Blackwell GPU deployments across 60 MW of data center capacity .

What This Means for Nvidia's Strategy

These partnerships reveal a deliberate shift in how Nvidia approaches the AI infrastructure market. The company is moving from selling individual chips to financializing its hardware advantage across several layers:

• From one-time sales to recurring revenue. Instead of selling a GPU once, Nvidia collects a portion of every AI token sold on that hardware for 6 to 10 years . This smoothes revenue cycles and ties growth to actual AI consumption, not just data center construction.

• From component supplier to platform architect. Nvidia provides the full DSX AI factory blueprint — liquid cooling designs, networking (NVLink/NvSwitch/InfiniBand), CUDA and AI Enterprise software, and reference architectures . Partners effectively operate Nvidia-branded factories, giving Nvidia control over the reference architecture even when it does not own the building.

• Layered financial exposure. Nvidia now has:

  1. Upfront product revenue from selling GPUs into these factories.
  2. Recurring cloud revenue share from usage over multi-year terms.
  3. Credit-support and financing exposure — by co-investing, Nvidia takes partial balance-sheet risk in its partners' success .
  4. Ecosystem lock-in — deeper financial entanglement makes it harder for competitors like AMD or Intel to break into those data centers.

• Scaling beyond the hyperscalers. By financing smaller AI cloud providers, Nvidia expands the addressable market beyond AWS, Azure, and GCP. These partners target AI startups and enterprises that may lack access to hyperscale cloud GPU capacity .

In short, Nvidia is betting that its balance sheet and product leverage can capture value at every layer of the AI compute stack — not just at the point of chip sale, but across years of usage, cloud service margins, and ecosystem adoption.