Amazon AWS
🇺🇸 US Provider🇮🇳 2 India Regions👑 #1 Cloud GloballyThe world's largest cloud - 38 global regions, 200+ services, Mumbai and Hyderabad data centres, H100 GPU in India, Graviton ARM processors, and the deepest managed service ecosystem on the planet. India is AWS's second-largest market globally by web traffic.
About Amazon AWS
Amazon Web Services (AWS), launched in 2006 and headquartered in Seattle, is the world's leading cloud platform by every measurable dimension - market share (31%), revenue ($116B+ annualised in 2025), number of regions (38 globally), services offered (200+), and enterprise penetration (85%). What began as internal Amazon infrastructure became the blueprint for the entire cloud computing industry.
For Indian customers, AWS is not a foreign cloud at arm's length - it has two full-scale Indian regions: Asia Pacific (Mumbai), launched in 2016, and Asia Pacific (Hyderabad), launched in 2022. Both offer the complete suite of compute, storage, database, AI/ML, networking, and security services. Crucially, Mumbai is one of the few regions worldwide with H100 GPU availability (p5 instances), meaning Indian teams can train frontier AI models on local infrastructure without routing traffic to the US or Europe.
India is AWS's second-largest market by web traffic globally (10.83%), and AWS has pledged $13.2 billion in India infrastructure investment through 2030. The company operates 53 renewable energy projects in India. AWS Bedrock, SageMaker, Trainium, and Inferentia bring a managed AI stack that no Indian provider can yet match in depth.
AWS's key advantages over alternatives are: depth of managed services (no need to self-manage databases, queues, ML pipelines), global reach for multi-region architectures, the Graviton ARM processor (up to 40% better price-performance than x86), Spot Instances (up to 90% discount for fault-tolerant workloads), and unmatched enterprise certifications - including ISO, SOC, PCI DSS, HIPAA, FedRAMP, and MeitY-aligned compliance frameworks.
What's good about AWS
- Two Indian regions - Mumbai (ap-south-1) and Hyderabad (ap-south-2) - with full service availability and 3 Availability Zones each
- H100 GPU (p5 instances) available in Mumbai - one of very few regions globally; Indian teams can train frontier AI models locally
- Graviton4 ARM processors (m8g, c8g, r8g) - up to 40% better price-performance than x86, ~10-20% cheaper list price
- Spot Instances - up to 90% discount for fault-tolerant and interruptible workloads; ideal for ML training, batch jobs, CI/CD
- Savings Plans and Reserved Instances - 37-72% savings with 1-3 year commitments; EC2 Savings Plans are the most flexible
- 200+ fully managed services - no other provider comes close in breadth: databases, queues, ML, analytics, IoT, security, edge, and more
- Amazon Bedrock - managed LLM platform supporting Claude, Llama, Mistral, Titan; pay-per-token, no GPU management needed
- AWS SageMaker - end-to-end ML platform for training, fine-tuning, and deployment; deeply integrated with all AWS services
- Trainium2 and Inferentia2 chips - AWS's own AI silicon delivering 60-70% better price-performance than H100 for many training and inference workloads
- 12-month Free Tier for new accounts - 750 hrs/month t3.micro, 30 GB EBS, 5 GB S3, and more - enough to run a full dev environment
- $13.2 billion committed India investment through 2030 - expanding capacity, localisation, and compliance coverage
- Most comprehensive enterprise compliance - ISO 27001, SOC 1/2/3, PCI DSS, HIPAA, GDPR, and dozens more certifications globally
- AWS Marketplace - 42,000+ third-party software products deployable in one click on any AWS region
Watch-outs for Indian teams
- USD billing only - no INR pricing, no UPI/NEFT, no GST-ready invoices; currency risk is real for INR-revenue businesses
- Complex pricing - 200+ services each with their own pricing models; EBS, data transfer, NAT gateway, and CloudWatch costs add up fast and surprise teams
- Egress charges - $0.09/GB outbound data transfer after 100 GB/month free; large-scale data pipelines face significant surprise bills
- Not MeitY empanelled - cannot be used for Indian government and regulated PSU procurement without separate assessment
- Significantly pricier than Indian providers for GPU - AWS H100 (p5) ~₹2,300–2,700/hr in Mumbai vs. ₹219–249/hr on E2E/Cyfuture
- Mumbai region adds 20-30% premium over us-east-1 pricing - regional cost variance is substantial for budget-sensitive teams
- Complexity overhead - AWS expertise requires months of learning; onboarding cost and accidental misconfiguration bills are real risks for small teams
- Stopped instances still bill for EBS volumes - $8-50/month per volume; orphaned Elastic IPs bill $3.60/month each
EC2 Pricing - Mumbai Region (ap-south-1)
All prices in USD per month (On-Demand, Linux). Mumbai adds ~20-30% premium over us-east-1. Per-second billing with 60-second minimum. EBS storage billed separately (~$0.10/GB-month for gp3).
Full price, no commitment. Per-second billing. Best for unpredictable or short-term workloads. Use as a baseline to compare other models.
1-3 year commitment. 37-72% savings. EC2 Savings Plans are flexible across instance families. Best for steady, predictable workloads.
Up to 90% off On-Demand. Interruptible with 2-min notice. Best for ML training, batch jobs, CI/CD, and stateless workloads.
General Purpose - t3 (Burstable) · Mumbai On-Demand
| Instance | vCPU | RAM | On-Demand / hr | ~Monthly | 1-yr Savings Plan |
|---|---|---|---|---|---|
| t3.micro | 2 | 1 GB | $0.0116 | ~$8.40 | ~$5.20 |
| t3.small | 2 | 2 GB | $0.0232 | ~$17 | ~$10 |
| t3.medium | 2 | 4 GB | $0.0464 | ~$33 | ~$21 |
| t3.large | 2 | 8 GB | $0.0928 | ~$67 | ~$41 |
General Purpose - m7g (Graviton3 ARM, best price-performance) · Mumbai On-Demand
| Instance | vCPU | RAM | On-Demand / hr | ~Monthly | vs x86 saving |
|---|---|---|---|---|---|
| m7g.medium | 1 | 4 GB | $0.0413 | ~$30 | ~20% cheaper |
| m7g.large | 2 | 8 GB | $0.0826 | ~$60 | ~20% cheaper |
| m7g.xlarge | 4 | 16 GB | $0.1651 | ~$119 | ~20% cheaper |
GPU / Accelerated Instances - On-Demand (Mumbai & Global)
| Instance | GPU / Chip | GPUs | VRAM | Region | On-Demand / hr |
|---|---|---|---|---|---|
| p5.48xlarge | ⚡ NVIDIA H100 | 8× | 640 GB | 🇮🇳 Mumbai | $98.32 |
| p5en.48xlarge | 🆕 NVIDIA H200 | 8× | 1.1 TB | 🇮🇳 Mumbai | $41.61 |
| g6.xlarge | ⚡ NVIDIA L4 | 1× | 24 GB | US / EU | $0.80 |
| trn2.48xlarge | 🔧 AWS Trainium2 | 64× | 1.5 TB | US East | ~$32 |
| inf2.xlarge | 🔧 AWS Inferentia2 | 1× | 32 GB | Multiple | $0.76 |
AWS Mumbai GPU is ~4× more expensive than Indian providers for raw compute. The premium buys deep managed services, SageMaker integration, Bedrock, and enterprise SLAs - worth it for production AI, less so for pure training runs.
Key Managed Services - Starting Prices
| Service | Starting price | Notes |
|---|---|---|
| Amazon S3 | $0.023/GB-mo | First 50TB in ap-south-1. Egress $0.09/GB after 100GB free. |
| Amazon RDS (MySQL/Postgres) | ~$25/mo | db.t4g.micro, 20 GB gp3. Multi-AZ doubles the cost. |
| Amazon EKS (Kubernetes) | $0.10/hr/cluster | $73/mo for control plane + worker node Droplet costs. |
| AWS Lambda | Free tier | 1M requests + 400K GB-sec free forever. Beyond: $0.20/1M requests. |
| Amazon Bedrock (Claude) | Pay per token | No GPU management. Claude Sonnet: ~$3/M input, $15/M output tokens. |
| Amazon SageMaker | ~$0.046/hr | ml.t3.medium notebook instance. Training + hosting billed separately. |
| Amazon CloudFront (CDN) | Free tier | 1 TB/mo + 10M requests free forever. Beyond: $0.0085/GB (India). |
| AWS Free Tier (12 months) | Free | 750 hrs t3.micro, 30 GB EBS, 5 GB S3, 15 GB data transfer, + more. |
Key Services (200+ total)
AWS offers 200+ services. Below are the most relevant categories for Indian developers and enterprises.
India & APAC Regions
| Region | City | Code | AZs | GPU Available |
|---|---|---|---|---|
| 🇮🇳 Asia Pacific | Mumbai | ap-south-1 | 3 | ✓ H100 (p5), H200 (p5en), L4 |
| 🇮🇳 Asia Pacific | Hyderabad | ap-south-2 | 3 | CPU / select instances |
| 🇸🇬 Asia Pacific | Singapore | ap-southeast-1 | 3 | ✓ GPU available |
| 🇯🇵 Asia Pacific | Tokyo | ap-northeast-1 | 4 | ✓ H100 (p5), H200 |
| + 34 more regions across USA, Europe, Middle East, Africa, LATAM | ||||
AWS operates 38 regions globally with 121 Availability Zones (as of 2025). Full list at aws.amazon.com/about-aws/global-infrastructure.
Alternatives to Amazon AWS
AWS is the right choice when you need its depth. But for many Indian workloads, the alternatives below offer dramatically better value:
India's NSE-listed cloud with H200/H100 GPU from ₹49/hr spot - ~4× cheaper than AWS Mumbai for raw GPU. INR billing, Indian data sovereignty. Ideal when you need GPU training without AWS's managed service overhead.
MeitY-empanelled Indian cloud with 20+ years of enterprise track record. H100 from ₹219/hr - ~11× cheaper than AWS on-demand GPU. Better for Indian government and regulated sector workloads.
Blackstone-backed India neocloud with managed AI stack (MLOps, Inference, RAG, Voicebot). Combines Indian sovereignty with SageMaker-like tooling - at a fraction of AWS pricing.
India's GPU-first cloud with H100, A100 at INR pricing. Self-serve, simple, up to 60% cheaper than AWS. Good for AI/ML teams that want Indian data sovereignty without the complexity of AWS's 200+ services.
India's most affordable VPS cloud - NVMe Droplets from ₹399/mo with INR billing and 7 DC locations. Best AWS alternative for budget-conscious teams that don't need the AWS managed service ecosystem.
The developer cloud - Droplets from $4/mo, Bangalore DC, best-in-class DX and docs. Great AWS alternative for startups and developers who want simplicity, predictable pricing, and no AWS complexity tax.
750 hrs/month EC2 t3.micro, 30 GB EBS, 5 GB S3, 15 GB transfer, Lambda free tier, and more. No credit card surprises for 12 months.
Start Free Tier →AWS bills in USD and is not MeitY-empanelled. GPU in Mumbai costs ~4–11× more than Indian providers like E2E or Cyfuture. Use AWS when you need Bedrock, SageMaker, or its managed service ecosystem - use Indian providers for raw compute cost savings.
Common questions
AWS GPU FAQs
What GPUs does AWS offer?
AWS EC2 accelerated-computing families include NVIDIA B300, B200, H200, H100, A100, RTX PRO 6000, RTX PRO 4500, L40S, L4, A10G, T4, V100 and K80, plus AMD Radeon Pro V520 graphics instances.
What is the cheapest AWS GPU instance in this comparison?
The example AMD Radeon Pro V520 g4ad.xlarge configuration is the lowest at about $0.38 per hour. The NVIDIA T4 g4dn.xlarge example is about $0.53 per hour.
Does AWS offer Spot GPU instances?
Yes. EC2 Spot Instances use spare capacity at variable discounts, but availability differs by instance type, region and Availability Zone, and instances can be interrupted.
Can AWS GPU instances use Savings Plans or reservations?
Many EC2 usage patterns can benefit from Compute Savings Plans, EC2 Instance Savings Plans or Reserved Instances. Eligibility and coverage vary by family, region and arrangement.
What are EC2 Capacity Blocks for ML?
Capacity Blocks let customers reserve eligible GPU capacity for a future time window for planned training, experiments, fine-tuning and temporary demand spikes.
Which AWS instance is best for LLM training?
P6-B300, P6-B200, P5e H200 and P5 H100 systems target demanding model training. P4d A100 remains useful where its memory, performance and software compatibility are sufficient.
Does AWS offer GPU instances in India?
AWS operates regions in Mumbai and Hyderabad. GPU family availability is region-specific, so confirm the exact EC2 SKU and quota before deployment.
Are the prices on this page guaranteed?
No. The amounts are representative examples and can vary by region, operating system, tenancy and agreement. Confirm the final estimate with AWS Pricing Calculator.