AM Intelligence (AMI), the artificial intelligence compute and infrastructure venture established by the founders of renewable energy giant Greenko Group, has placed two firm, binding orders for 20,000 NVIDIA Rubin GPUs. Announced on October 5, 2026, the hardware procurement will be deployed as liquid-cooled NVIDIA Vera Rubin NVL72 rack-scale systems across high-density AI data centers in India and Malaysia, targeting delivery in the second quarter of 2027.
The transaction follows AMI’s initial procurement of 9,000 Rubin GPUs on August 25, 2026. Together, the three commitments elevate the firm’s contracted fleet to approximately 29,000 NVIDIA Rubin GPUs, representing nearly 100 megawatts (MW) of operational power draw.
The expansion marks a major step forward within AMI’s 400 MW development pipeline of Compute-as-a-Service (CaaS) capacity spanning India, Malaysia, the United States, and Europe, establishing one of the earliest hyperscale enterprise deployments of NVIDIA’s next-generation Rubin computing architecture outside North America.
Key Takeaways
- Firm Order for 20,000 GPUs: AM Intelligence has placed two binding orders for 20,000 NVIDIA Rubin GPUs, scheduled for commercial deployment in Q2 2027.
- NVL72 Rack-Scale Architecture: The compute blocks will be assembled as liquid-cooled NVIDIA Vera Rubin NVL72 systems, integrating high-performance NVLink interconnects and advanced networking.
- Approaching 100 MW in Committed Compute: Combined with an August order for 9,000 GPUs, AMI’s committed capacity reaches approximately 29,000 Rubin GPUs, drawing nearly 100 MW of critical IT power.
- Geographic Footprint: Hardware will expand AMI’s flagship operational AI factory in Hyderabad, India, while anchoring a parallel greenfield compute hub in Malaysia.
- Strategic Pipeline Context: The 100 MW committed tranche forms part of AMI’s ~400 MW development pipeline, advancing toward an interim target of 1 GW in Compute-as-a-Service and a long-term vision of 5 GW of powered AI campuses.
- Energy-Compute Convergence: Founded by Anil Kumar Chalamalasetty and Mahesh Kolli (promoters of Greenko and AM Green), the venture couples captive, round-the-clock (RTC) clean energy with high-density server halls to optimize “electron-to-token” processing costs.
AM INTELLIGENCE: COMPUTE CAPACITY EXPANSION ROADMAP
AUGUST 2026 COMMITMENT:
┌─────────────────────────────────┐
│ • 9,000 NVIDIA Rubin GPUs │ ──► ~30 MW Dedicated Capacity Tranche
│ • Hyderabad Phase-1 AI Factory │
└─────────────────────────────────┘
│
▼
OCTOBER 5, 2026 BINDING EXPANSION:
┌─────────────────────────────────────────────────────────────┐
│ • 20,000 Additional NVIDIA Rubin GPUs │
│ • Deployment: Vera Rubin NVL72 Liquid-Cooled Systems │ ──► +70 MW Power Envelope
│ • Dual-Hub Allocation: Hyderabad (India) & Malaysia Campus │ Delivery Target: Q2 2027
└─────────────────────────────────────────────────────────────┘
│
▼
CUMULATIVE COMMITTED FOOTPRINT:
┌────────────────────────────────────────────────────────────────────────┐
│ ~29,000 NVIDIA RUBIN GPUs │ ~100 MW ACTIVE CRITICAL COMPUTE CAPACITY │
└───────────────────────────────────┬────────────────────────────────────┘
│
▼
GLOBAL DEVELOPMENT PIPELINE (~400 MW across India, Malaysia, US & Europe)
│
▼
LONG-TERM TARGET: 1 GW Compute-as-a-Service │ 5 GW Powered AI Data Centers
The Infrastructure Strategy: Mastering “Electron-to-Token” Economics
As artificial intelligence workloads transition from baseline model training to real-time agentic reasoning, inference clusters require high-density power envelopes. Standard enterprise server racks historically required 8 to 15 kilowatts (kW) of electrical capacity; in contrast, liquid-cooled NVIDIA NVL72 architectures consume between 100 kW and 130 kW per rack.
The core operational thesis behind AM Intelligence is bridging the gap between volatile utility power grids and massive computational demand. Leveraging the industrial infrastructure expertise of the Greenko leadership group—which built India’s largest pumped-storage hydropower projects and gigawatt-scale wind-solar hybrid assets—AMI co-locates computing hardware directly alongside dedicated clean power corridors:
THE INTEGRATED ENERGY-TO-COMPUTE ARCHITECTURE
CAPTIVE ENERGY ASSETS (Greenko / AM Green Infrastructure)
• Round-the-Clock (RTC) Renewable Hybrid Power
• Integrated Pumped Hydro Energy Storage (PPHES)
• Substation Interconnects at Dedicated High Voltages
│
▼
HIGH-DENSITY DATA CENTER FACILITY
• Advanced Direct-to-Chip Liquid Cooling Loops
• Zero-Water Evaporative Chiller Plants
• Extreme Power Usage Effectiveness (PUE < 1.15)
│
▼
ACCELERATED SILICON LAYER
• NVIDIA Vera Rubin NVL72 Rack-Scale Infrastructure
• 5th-Gen NVLink Multi-Terabit Optical Fabric
• Quantum-X InfiniBand & Spectrum-X Ethernet Switching
│
▼
COMPUTE-AS-A-SERVICE (CaaS) PLATFORM
• Hyperscalers & Neo-Cloud Providers
• Sovereign AI Frameworks & Frontier Research Labs
• Enterprise Fine-Tuning & Real-Time Agentic Workloads
By owning and structuring both the electrical generation assets and the mechanical data center envelope, AMI minimizes transmission wheeling losses, grid curtailment, and diesel-generator reliance. This integrated model provides what the company terms “best-in-class electron-to-token economics”—reducing the levelized cost of energy required to generate every individual token of AI output.
Next-Gen Hardware: Why the NVIDIA Rubin Architecture Matters
NVIDIA formally outlined the production roadmap for its Vera Rubin computing platform on May 31, 2026, positioning the silicon as the operational architecture designed to train multi-modal, agentic reasoning models at scale:
| Technical Parameter | NVIDIA Blackwell (Preceding Gen) | NVIDIA Vera Rubin (AMI Deployment) |
| Rack-Scale Unit | NVL72 (Blackwell Ultra) | Vera Rubin NVL72 |
| Memory Architecture | High-Bandwidth Memory (HBM3E) | Next-Gen HBM4 Memory Stacks |
| Compute Packaging | Dual-die CoWoS-L packaging | Advanced multi-chiplet unified CPU+GPU module |
| Interconnect Bandwidth | 1.8 TB/s bidirectional NVLink 5 | Upgraded high-bandwidth optical NVLink interconnects |
| Target Workload Profile | Mixed precision training & inference | Autonomous Agentic AI factories & multi-step reasoning |
| Thermal Management | Hybrid air/liquid cooling support | Mandatory direct-to-chip 100% liquid cooling |
Deploying the hardware in an NVL72 form factor unifies 72 individual Rubin GPUs into a single, cohesive liquid-cooled rack that functions logically as a massive, unified graphics processing unit.
“AM Intelligence is bringing the next generation of AI factories to India and Malaysia with NVIDIA Vera Rubin,” stated Vishal Dhupar, Managing Director, Asia South, at NVIDIA. “By combining accelerated computing with high-performance networking and liquid-cooled infrastructure, AMI is creating the foundation for developers and enterprises to build more ambitious AI.”
Dual-Hub Deployment: Hyderabad and Malaysia
The geographical split of the 20,000-unit procurement between Hyderabad and Malaysia carries distinct strategic advantages:
┌────────────────────────────────────────────────────────────────────────┐
│ AMI REGIONAL COMPUTE HUBS │
├─────────────────────┬──────────────────┬───────────────────────────────┤
│ LOCATION │ PRIMARY ROLE │ STRATEGIC ADVANTAGE │
├─────────────────────┼──────────────────┼───────────────────────────────┤
│ Hyderabad, India │ Domestic AI Hub │ • Proximity to Indian tech │
│ (Flagship AI │ & Sovereign Base │ talent and enterprise SaaS │
│ Factory) │ │ • Direct access to Greenko's │
│ │ │ Telangana clean power grid │
│ │ │ • Supports IndiaAI Mission │
├─────────────────────┼──────────────────┼───────────────────────────────┤
│ Malaysia Campus │ Southeast Asian │ • Regional hub for ASEAN data │
│ (Johor / Cyberjaya │ Gateway & Cloud │ cross-connects │
│ Corridor) │ Transit Node │ • Low latency to Singapore │
│ │ │ • Favorable industrial power │
│ │ │ tariffs and land access │
└─────────────────────┴──────────────────┴───────────────────────────────┘
1. Anchoring India’s Sovereign AI Ecosystem in Hyderabad
Expanding the Hyderabad campus supports India’s national mandate under the IndiaAI Mission to deploy subsidized, indigenous compute capacity. Domestic software development platforms, regional foundation model builders (such as Sarvam AI and Krutrim), and public-sector entities can access cutting-edge Rubin hardware within sovereign borders, addressing data localization directives under the Digital Personal Data Protection (DPDP) Act.
2. The Malaysian Gateway to Southeast Asia
Malaysia—particularly the southern industrial corridor of Johor and the technology clusters in Cyberjaya—has established itself as Southeast Asia’s fastest-growing computing corridor. With neighboring Singapore imposing tight constraints on new data center power allocations due to resource limits, AMI’s Malaysian facility positions Rubin clusters right across the Johor Strait, providing sub-millisecond round-trip latency to Singapore’s international subsea cable hubs.
Competitive Dynamics: The Global AI Compute Race
AM Intelligence’s rapid orders place it in a specialized class of global infrastructure builders known as “neo-cloud” and GPU-native infrastructure operators:
┌────────────────────────────────────────────────────────────────────────┐
│ GLOBAL AI COMPUTE PLAYERS: POSITIONING │
├──────────────────┬──────────────────┬──────────────────────────────────┤
│ OPERATOR │ GEOGRAPHIC FOCUS │ PRIMARY MOAT │
├──────────────────┼──────────────────┼──────────────────────────────────┤
│ AM Intelligence │ India, Malaysia, │ Direct ownership of captive │
│ (AMI) │ US, Europe │ renewable power & pumped hydro │
├──────────────────┼──────────────────┼──────────────────────────────────┤
│ CoreWeave │ US & Western │ First-mover specialized GPU cloud│
│ │ Europe │ backed by deep NVIDIA equity │
├──────────────────┼──────────────────┼──────────────────────────────────┤
│ Yotta Data │ India (Pan-India │ Large enterprise co-location │
│ Services │ campuses) │ footprint with H100/Blackwell │
├──────────────────┼──────────────────┼──────────────────────────────────┤
│ Lambda Labs │ US & International│ Developer-centric software stack │
│ │ Cloud │ and on-demand GPU clusters │
└──────────────────┴──────────────────┴──────────────────────────────────┘
While Western neo-clouds like CoreWeave and Lambda have secured massive venture capital and private equity backing, their primary operational bottleneck remains procuring physical power interconnects from stressed North American and European electric utilities.
By having founders who control massive renewable generation pipelines, AMI bypasses years-long utility interconnection queues in India and emerging markets, allowing server racks to be energized as soon as silicon arrives from Taiwan Semiconductor Manufacturing Company (TSMC) and NVIDIA packaging plants.
What Remains Uncertain
While the binding orders represent a milestone in regional computing infrastructure, several execution variables will be monitored:
- Foundry Packaging and Delivery Timelines: The NVIDIA Vera Rubin platform depends on advanced chip packaging and cutting-edge HBM4 memory stacks. Any manufacturing bottlenecks at TSMC or memory suppliers could alter the guided Q2 2027 delivery schedule.
- Liquid-Cooling Facility Commissioning: Operating 100 MW of high-density direct-to-chip liquid cooling requires precision mechanical engineering. Facility construction teams in Hyderabad and Malaysia must hit testing benchmarks across pumping loops and secondary fluid networks prior to rack delivery.
- Capacity Off-Take Contracts: While AMI is targeting hyperscalers, sovereign entities, and frontier research labs, the commercial market will monitor whether capacity is committed via multi-year take-or-pay anchor tenant contracts or exposed to spot-market pricing volatility.
What Happens Next
Over the coming quarters leading into mid-2027:
- Phase-1 Civil Readiness: Engineering teams will complete the physical shell and high-voltage substation connections for the expanded Hyderabad AI factory and the Malaysian facility.
- Cooling Infrastructure Integration: Deployment of closed-loop liquid-cooling manifolds and dry coolers configured to support the thermal dissipation of NVL72 racks.
- Follow-On Pipeline Tranches: AMI will advance commercial negotiations for the remaining 300 MW balance of its 400 MW pipeline, evaluating prospective campus sites across the United States and Europe for subsequent deliveries scheduled through late 2027 and early 2028.
Frequently Asked Questions
Who is behind AM Intelligence?
AM Intelligence (AMI) is an AI infrastructure and Compute-as-a-Service company founded by the promoters of Greenko Group and AM Green (Anil Kumar Chalamalasetty and Mahesh Kolli), known for their large-scale renewable energy, pumped-storage hydropower, and green hydrogen ventures.
How many NVIDIA GPUs has AM Intelligence ordered in total?
Following its initial order of 9,000 NVIDIA Rubin GPUs announced on August 25, 2026, the company placed two additional binding orders for 20,000 units on October 5, 2026. This brings its total committed capacity to approximately 29,000 NVIDIA Rubin GPUs.
When will the 20,000 GPUs be delivered and where will they be installed?
The 20,000 GPUs are scheduled for delivery in the second quarter of 2027 (Q2 2027). They will be installed across AMI’s AI factory in Hyderabad, India, and an upcoming facility in Malaysia.
What does “NVL72” mean in the context of NVIDIA Rubin?
The Vera Rubin NVL72 is a rack-scale system that integrates 72 GPUs into a single liquid-cooled cabinet connected by high-speed NVLink switches. The architecture enables all 72 chips to function as one unified accelerator, designed for large-scale AI training and real-time inference workloads.
Get the day’s top stories in your inbox
One concise email. No spam, unsubscribe anytime.
