Diamonds aren’t just for jewelry anymore — they’re cooling NVIDIA H200 GPUs in Bengaluru. On February 23, 2026, NxtGen AI deployed the world’s first Diamond-Cooled NVIDIA H200 GPU servers using Akash Systems’ patented Diamond Cooling® technology, delivering 15% more usable compute, 5°C lower GPU hotspot temperatures, and throttle-free operation at ambient temperatures up to 50°C (122°F). This isn’t theoretical — the servers are live in production, powering India’s sovereign AI infrastructure.
For context: AI data centers typically run at 24–29°C (75–85°F). India’s data centers often hit 35–40°C. Diamond Cooling® eliminates thermal throttling entirely, meaning GPUs stay at peak performance even in extreme heat — a structural advantage for countries where cooling costs and ambient temperatures are AI infrastructure bottlenecks.
Table of Contents
The Physics: Why Diamond Beats Copper by 5x
Here’s the simple physics: diamond is the most thermally conductive material known to humanity, with a thermal conductivity of 2200 W/mK — 5x better than copper (400 W/mK), the current industry standard.
Akash Systems’ breakthrough isn’t just using diamond — it’s coupling synthetic diamond materials directly to semiconductor chips to accelerate heat extraction from GPUs running at 90–100°C. The technology was originally developed in partnership with NASA for satellite systems and is currently operating in active orbit before being adapted for terrestrial AI infrastructure.
| Cooling Material | Thermal Conductivity (W/mK) | Heat Removal vs Copper |
|---|---|---|
| Diamond | 2200 | 5x faster |
| Copper | 400 | Baseline (1x) |
| Aluminum | 205 | 0.5x |
| Thermal Paste | 3–12 | 0.01x |
The Diamond Cooling® layer is an additive innovation — it complements existing air and liquid cooling architectures rather than replacing them. You can stack it on top of liquid immersion cooling, direct-to-chip liquid cooling, or traditional air cooling, and get incremental performance gains.
Performance Results: 15% More Compute Without Buying More GPUs
NxtGen’s deployment delivers measurable improvements:
| Metric | Improvement | Real-World Impact |
|---|---|---|
| GPU Hotspot Temperature | ~5°C lower | Sustained peak silicon performance, improved thermal stability |
| FLOPs per Watt | ~15% improvement | Higher compute efficiency per unit of power consumed |
| Usable Compute Output | ~15% increase | Eliminates thermal throttling = higher effective GPU productivity |
| Operating Ambient Temp | Up to 50°C (122°F) | Throttle-free in extreme heat (critical for India, Middle East, Africa) |
| GPU Fan Speed | 50% reduction | 90% reduction in GPU fan energy consumption |
| Total Cost of Ownership | Lower TCO | Enhanced capital productivity per server |
The 15% improvement in usable compute is the key number. In AI infrastructure economics, even a 1–2% compute gain is meaningful. 15% is transformative. It’s the equivalent of getting 1.15 H200 GPUs for the price of 1.
Dr. Felix Ejeckam, Co-Founder & CEO of Akash Systems, framed it perfectly: “In a world where a one to two percent increase in compute is meaningful, a 15 percent increase in GPU compute is truly game-changing.”
Why India? The Sovereign AI Infrastructure Play
NxtGen AI isn’t just another cloud provider — it’s India’s largest sovereign cloud provider, built to ensure data, operational, and technological sovereignty. This matters enormously as India races to build domestic AI capacity under the India AI Mission and Digital India RISC-V (DIR-V) initiatives.
A.S. Rajgopal, MD & CEO of NxtGen, positioned it clearly: “At NxtGen, we are architecting India’s most advanced sovereign AI infrastructure. The deployment of Diamond-Cooled NVIDIA H200 servers strengthens our AI backbone by increasing usable compute per watt and per rack.”
NxtGen operates data centers across Ahmedabad, Bengaluru, Faridabad, and Mumbai and offers GPU clusters unified across NVIDIA, AMD, and Intel architectures, giving enterprises flexibility while maintaining sovereign data residency.
For Indian enterprises, government agencies, and AI startups, NxtGen’s Diamond-Cooled infrastructure means:
- 15% more AI training throughput without buying additional GPUs
- Lower power bills (improved FLOPs per watt reduces electricity costs)
- Guaranteed performance in summer months when traditional data centers throttle
- Sovereign data residency (BFSI, healthcare, defense workloads stay in India)
The $27 Million Deal: Khosla Ventures & Founders Fund Backing
This isn’t a pilot. In December 2024, Akash Systems and NxtGen signed a $27 million contract (₹225 crore) to deploy Diamond-Cooled servers across NxtGen’s data center footprint. The deal positions NxtGen as the world’s most energy-efficient AI compute provider by combining:
- Akash’s Diamond Cooling®
- Waterless liquid cooling
- NVIDIA and AMD GPUs
The result? Double the performance per watt, enabling NxtGen to reduce AI compute costs by 50% vs prevailing market rates while maintaining higher margins.
Akash Systems is backed by Vinod Khosla (Khosla Ventures) and Peter Thiel (Founders Fund) — two of Silicon Valley’s most legendary AI infrastructure investors. Vinod Khosla said of the partnership: “This partnership with NxtGen highlights the scalability of Akash’s technology and showcases its ability to accelerate AI workloads by reducing thermal throttling and maximizing compute cycles.”
Akash also secured $68 million in CHIPS Act funding from the U.S. Department of Commerce ($18.2M direct funding + $50M in federal and California state tax credits) to scale production.
The Competitive Landscape: Who Else Is Solving GPU Heat?
GPU thermal management is becoming the defining constraint in AI infrastructure. Everyone’s racing to solve it:
| Company | Approach | Backing | Deployment Status |
|---|---|---|---|
| Akash Systems | Diamond Cooling (5x copper) | Khosla Ventures, Founders Fund, $68M CHIPS Act | Production (NxtGen, Feb 2026) |
| Liquidstack | Liquid immersion cooling | Series C, $100M+ raised | Production (multiple data centers) |
| Submer | Liquid immersion (Spain-based) | Series B, €60M raised | Production (Europe, US) |
| CoolIT Systems | Direct-to-chip liquid cooling | Acquired by Daikin (2023) | Production (enterprise servers) |
| ZutaCore | Waterless direct-on-chip cooling | Series B, $32M raised | Pilot deployments |
Akash’s edge: Diamond Cooling is additive, not replacement. You can layer it on top of liquid cooling for compounded benefits. Most competitors force you to choose one thermal architecture; Akash enhances whatever you already have.
What’s Next: AMD GPUs, Blackwell, and Beyond
Akash isn’t limiting Diamond Cooling® to NVIDIA. The company offers:
- NVIDIA H200 (deployed now with NxtGen)
- AMD GPU variants (available)
- NVIDIA Blackwell series (aftermarket solutions planned for 2026)
This multi-vendor strategy matters. As enterprises diversify away from NVIDIA-only stacks (due to cost, availability, or optimization needs), Akash’s ability to cool AMD MI300X, Intel Gaudi 3, and future architectures makes it infrastructure-agnostic.
The India Data Center Opportunity
India’s AI infrastructure market is exploding:
- Yotta Data Services raised $340M, building 100MW+ AI-ready capacity
- CtrlS Datacenters expanding to 1GW capacity by 2028
- Airtel Nxtra investing ₹5,000 crore in hyperscale data centers
- AWS, Google Cloud, Microsoft Azure all expanding India regions
Every one of these data centers faces the same thermal constraints. Ambient temperatures in Mumbai, Hyderabad, and Chennai regularly exceed 35°C, forcing expensive cooling overhead. If Diamond Cooling delivers 15% compute gains consistently, every hyperscaler will want it.
For Indian data center operators and cloud providers, Akash’s technology could be the difference between profitable AI infrastructure and thermally-constrained underperformance.
The Verdict: From Satellites to Sovereign AI
Diamond Cooling® started in orbit, validating its durability and thermal performance in space-grade satellite systems where failure isn’t an option. Now it’s powering India’s sovereign AI infrastructure, delivering measurable compute gains in production workloads.
The 15% improvement in usable compute isn’t marketing fluff — it’s the delta between GPUs that thermal-throttle under load and GPUs that sustain peak performance continuously. In AI training and inference, that 15% compounds across every epoch, every batch, every inference call.
NxtGen just deployed the world’s first Diamond-Cooled GPU servers. If the economics hold, they won’t be the last.
Follow the latest AI infrastructure developments, data center innovation, and Indian sovereign AI initiatives at Technosports.





