Nvidia Unveils Revenue-Sharing Model to Accelerate AI Factory Deployment with Cloud Partners
Nvidia's new partnership strategy aims to rapidly scale AI compute capacity, while industry analysts highlight intensifying component shortages as the next major bottleneck.
The story
Nvidia has introduced a new collaboration model for AI infrastructure, designed to accelerate the deployment of large-scale AI factories through revenue-sharing and credit-support mechanisms with cloud service providers. This initiative aims to provide startups, model developers, enterprises, and research institutions with faster access to AI computing power.
Initial partners include Sharon AI, which plans to deploy up to 40,000 NVIDIA Grace Blackwell GB300 GPUs, and Firmus, which will build a DSX AI factory park on Batam Island, Indonesia. Firmus's project is slated to expand to a power scale of 360 megawatts and deploy up to 170,000 Nvidia GPUs, signaling a significant push to expand global AI compute capacity.
Silicon
Ascend 950DT / 950PR
Maker: Huawei
What: AI accelerators for training (950DT) and inference (950PR), positioned as a lower-cost alternative to Nvidia.
For Whom: South Korean market, including enterprises like DeepSeek V4 deployments.
Specialized inference silicon
Maker: Etched
What: AI chips with low-voltage inference for high-throughput workloads and cluster-scale memory for low latency.
For Whom: Hyperscalers and AI infrastructure builders seeking cost-effective inference.
Licensable GPU architecture
Maker: Oxmiq Labs
What: Modular chiplet and software stack offering custom AI compute without building a full GPU program.
For Whom: Semiconductor companies and AI infrastructure builders seeking custom AI compute solutions.
The build-out
| Project | Who | Scale | Where |
|---|---|---|---|
| Greenlight Electricity Center | Pembina | C$4.6 billion total project expenditure for dedicated power. | Undisclosed location for a hyperscale data center. |
| Jakarta Data Center Campus Expansion | STT Global Data Centers | Adding new facilities totaling 360 MW of AI-ready IT capacity. | Jakarta, Indonesia. |
| New Data Center Campus | Amazon | $10 billion investment. | Montgomery County, Missouri, US. |
| New Data Center Campus | $15 billion investment. | Montgomery County, Missouri, US. |
Supply & policy signals
A Nomura Report indicates shortages are shifting beyond advanced chips to critical components like PCB/CCL, IC substrate, higher-end capacitors, PMIC, and optical components.
Implication: Supply constraints across the hardware ecosystem are likely to intensify through 2027, despite strong AI investment.
TSMC CEO CC Wei states global demand for AI chips will outpace production for years.
Implication: Pressure on worldwide supply chains will continue, and fulfilling all orders will take considerable time despite capacity expansions.
Semiconductor industry group SEMI warned the US government against intervening in memory chip prices or production capacity.
Implication: Such interventions risk prolonging the demand downturn and worsening the existing supply squeeze driven by AI.
Virginia lawmakers approved a new consumption tax of $0.011 per kilowatt-hour on all electricity consumed by data centers, effective July 1, 2026.
Implication: This tax is estimated to generate $600 million annually for Virginia's general fund and could increase operational costs for data center operators in the state.
What we'll be watching
- Huawei's planned Q4 2026 launch of Ascend AI accelerators and Atlas 950 SuperPod in South Korea.
- Nvidia's upcoming Blackwell platform.
- Nvidia's Rubin architecture, scheduled for late 2026, detailing performance metrics and production timelines.
- AMD's Instinct MI400-class silicon and Helios rack systems gaining further traction among hyperscale customers.
- Digital Realty's progress towards 2 GW capacity at its Kansas City hyperscale development by early 2028.
- Hscale's MXP2 campus in Milan delivering 120 MW by the end of 2028.
Reporting + analyst voices: grounded via Google Search at publish time.