Blog

Insights on GPU infrastructure, AI workloads, and bare metal computing.

Pinned
All Posts

India's $200 Billion AI Infrastructure Moment

India is targeting $200B+ in AI infrastructure investment by 2028. Between government incentives, hyperscaler commitments, and domestic conglomerates, the country's compute capacity is about to increase 8x.

NovaCore TeamFebruary 25, 2026

DeepSeek and the Open-Source Inference Revolution: Why Hardware Matters More Than Ever

DeepSeek V3 trained for $5.6M on 2,048 GPUs. As open-source models match frontier performance at a fraction of the cost, the bottleneck shifts from model access to inference infrastructure.

NovaCore TeamFebruary 21, 2026

Adani's $100 Billion Bet on Sovereign AI Infrastructure

The Adani Group committed $100 billion to build green-energy-powered, hyperscale AI data centers by 2035. What this means for India's position in the global AI compute landscape.

NovaCore TeamFebruary 18, 2026

India's Zero-Tax AI Policy Through 2047: What It Means for GPU Infrastructure

India's Union Budget 2026-27 introduced a tax holiday through 2047 for foreign cloud providers operating AI data centers in India. Here's why this changes the math for AI infrastructure decisions.

NovaCore TeamFebruary 3, 2026

NVIDIA Blackwell Ultra: What the GB300 NVL72 Means for AI Infrastructure

NVIDIA's Blackwell Ultra architecture delivers 1.5x the performance of B200 with 288GB HBM3e per GPU. Here's what the specs mean in practice for training and inference workloads.

NovaCore TeamJanuary 20, 2026

Why Bare Metal Matters for AI Training

Virtualization overhead costs you 10-15% of your GPU compute. At supercluster scale, that's millions in wasted spend.

NovaCore TeamNovember 15, 2025

What to Look for in a GPU Cloud Provider

Not all GPU clouds are equal. Beyond price-per-hour, here are the factors that actually determine whether a provider will work for serious AI workloads.

NovaCore TeamJuly 22, 2025

Training and Inference Need Different Infrastructure

Teams that optimize their GPU clusters for training often find them poorly suited for inference — and vice versa. Here's why the hardware requirements diverge and what to do about it.

NovaCore TeamMay 8, 2025

The GPU Shortage: What's Real, What's Not, and What Comes Next

H100 lead times have dropped from 52 weeks to under 8. The GPU shortage narrative is outdated — but the market dynamics that replace it are more nuanced than 'supply is fixed.'

NovaCore TeamMarch 12, 2025