H100 vs MI300X — which GPU should I choose?

H100 if you need the broadest software ecosystem (CUDA, TensorRT, vLLM). MI300X if you need maximum VRAM (192GB vs 80GB) for large model inference. MI300X offers better $/TFLOP but NVIDIA's software stack is more mature.

How much VRAM do I need for LLM inference?

A model needs ~2x its parameter count in GB for FP16 inference (70B model = ~140GB VRAM). With INT8 quantization ~70GB, INT4 ~35GB. A single H100 (80GB) runs 70B at INT8; MI300X (192GB) runs it at full FP16.

Should we buy GPUs or use cloud GPU instances?

If running GPUs 60%+ of the time, on-premise ownership wins on 3-year TCO. Below 40% utilization, cloud is more cost-effective. Many enterprises use hybrid: owned hardware for baseline, cloud for peak demand.

What is the cheapest way to rent an H100 GPU?

As of 2026, H100 cloud pricing ranges from $2.23/hr (Lambda, RunPod spot) to $4+/hr (AWS, Azure on-demand). Reserved instances and spot pricing offer 30-60% savings. CoreWeave and Lambda typically offer the lowest rates.

What GPU is best for LLM training in 2026?

NVIDIA H200 SXM (141GB HBM3e) for proven clusters, B200 for next-gen 4-5x speedup over H100, or AMD MI300X (192GB) for budget-conscious teams. For JAX workloads, Google TPU v5p pods offer unmatched scale.

GPUADVISOR

Advisory Enterprise ReportsAbout

GPUADVISOR

Updated May 2026

Cloud GPU Pricing — Cheapest H100 & B200 Rates

Compare real-time GPU rental prices across AWS, Azure, GCP, Lambda Labs, CoreWeave, RunPod, and Vast.ai. Find the cheapest H100, A100, and L40S instances — including on-demand, reserved, and spot pricing — to optimize your AI infrastructure spend.

✓ 6 cloud providers✓ H100, A100, L40S, MI300X✓ On-demand & reserved rates✓ Spot & interruptible pricing

Cloud GPU Pricing — May 2026

Compare Cloud GPU Costs

Verified pricing across AWS, GCP, Azure, Lambda, and CoreWeave — updated monthly. Find the best $/hr for your workload.

Prices verified May 2026 · updated weekly·Some "Rent Now" links are affiliate links — we earn a small commission at no cost to you. Price data and rankings are never influenced by commercial relationships.

$4.76

Cheapest H100/hr

CoreWeave per GPU

$5.50

Best Spot Deal

AWS L40S 48GB

Providers Tracked

Major cloud platforms

10+

GPU Types

B200, H200, H100, MI300X, TPU...

Provider:

GPU:

Cloud GPU instance pricing across AWS, GCP, Azure, Lambda, and CoreWeave — on-demand, spot, and reserved rates
Provider	GPU	Instance	VRAM	On-Demand $/hr ↑	Spot $/hr	1Y Reserved $/hr	Best For	Start
CoreWeave	L40S 48GB	l40s-48gb	48 GB	$1.50	N/A	N/A	Per-GPU L40S pricing	Deploy →
CoreWeave	A100 80GB	a100-sxm-80gb	80 GB	$2.21	N/A	N/A	Per-GPU A100 pricing	Deploy →
CoreWeave	MI300X 192GB	mi300x-sxm-192gb	192 GB	$4.10	N/A	N/A	Per-GPU MI300X pricing	Deploy →
CoreWeave	H100 80GB	h100-sxm-80gb	80 GB	$4.76	N/A	N/A	Per-GPU H100 pricing	Deploy →
GCP	TPU v5e	ct5e-minitpu-8t	128 GB HBM	$4.80	N/A	$3.02	Cost-effective JAX Inference	Deploy →
CoreWeave	H200 141GB	h200-sxm-141gb	141 GB	$5.20	N/A	N/A	Per-GPU H200 pricing	Deploy →
CoreWeave	B200 192GB	b200-sxm-192gb	192 GB	$6.50	N/A	N/A	Per-GPU B200 pricing	Deploy →
CoreWeave	B300 288GB	b300-sxm-288gb	288 GB	$8.50	N/A	N/A	Per-GPU B300 pricing	Deploy →
Lambda	L40S 48GB	gpu_8x_l40s	384 GB (8×48)	$12.00	N/A	N/A	Cost-Effective Inference & Rendering	Deploy →
GCP	TPU v6e	ct6e-standard-8t	256 GB HBM	$12.50	N/A	$7.88	Cost-Efficient JAX Training & Inference	Deploy →
GCP	TPU v4	ct4p-lowtpu-4t	128 GB HBM	$12.80	N/A	$7.68	JAX Training	Deploy →
RunPod	A100 80GB	SECURE-A100-SXM	640 GB (8×80)	$13.12	$6.5050% off	N/A	Budget large-model training	Deploy →
Lambda	A100 80GB	gpu_8x_a100_80gb_sxm4	640 GB (8×80)	$14.32	N/A	N/A	Training (best value A100)	Deploy →
AWS	L40S 48GB	g7.48xlarge	384 GB (8×48)	$16.00	$5.5066% off	$10.00	Enterprise Inference & Video	Deploy →
GCP	TPU v5p	ct5p-hightpu-4t	384 GB HBM	$21.10	N/A	$13.29	JAX/TPU-native training	Deploy →
Lambda	MI300X 192GB	gpu_8x_mi300x	1536 GB (8×192)	$24.50	N/A	N/A	High-VRAM Training	Deploy →
Lambda	H100 80GB	gpu_8x_h100_sxm5	640 GB (8×80)	$27.60	N/A	N/A	Training (best value H100)	Deploy →
RunPod	H100 80GB	SECURE-H100-SXM	640 GB (8×80)	$27.92	$14.0050% off	N/A	Lowest H100 price (Secure Cloud)	Deploy →
GCP	TPU v7	ct7p-hightpu-4t	384 GB HBM3e	$28.50	N/A	$18.50	Next-Gen JAX/TPU Frontier Training	Deploy →
Lambda	H200 141GB	gpu_8x_h200_sxm5	1128 GB (8×141)	$30.00	N/A	N/A	LLM Inference (best value)	Deploy →
Azure	A100 80GB	ND96amsr_A100_v4	640 GB (8×80)	$32.77	$9.8370% off	$20.43	Training & fine-tuning	Deploy →
Azure	MI250X 128GB	NDm_MI250X_v4	512 GB (4×128)	$36.00	$10.8070% off	$22.00	HPC & Scientific Computing	Deploy →
GCP	A100 80GB	a2-ultragpu-8g	640 GB (8×80)	$40.22	$12.0770% off	$25.34	Training & fine-tuning	Deploy →
RunPod	H200 141GB	SECURE-H200-SXM	1128 GB (8×141)	$44.00	$22.0050% off	N/A	Best-value H200	Deploy →
Azure	MI300A 128GB	ND_MI300A_v5	512 GB (4×128)	$45.00	$13.5070% off	$28.00	Unified Memory HPC	Deploy →
Lambda	B300 288GB	gpu_8x_b300_sxm	2304 GB (8×288)	$52.00	N/A	N/A	Cheapest B300 (per-node)	Deploy →
Azure	MI300X 192GB	ND_MI300X_v5	1536 GB (8×192)	$92.50	$27.7570% off	$58.10	High-memory LLM training	Deploy →
Azure	MI325X 288GB	ND_MI325X_v5	2304 GB (8×288)	$98.00	$29.4070% off	$62.00	Extreme VRAM LLM Training	Deploy →
AWS	H100 80GB	p5.48xlarge	640 GB (8×80)	$98.32	$35.5064% off	$62.12	Large-scale training	Deploy →
Azure	H100 80GB	ND96isr_H100_v5	640 GB (8×80)	$98.32	$29.5070% off	$60.96	Large-scale training	Deploy →
GCP	H100 80GB	a3-highgpu-8g	640 GB (8×80)	$98.35	$29.5170% off	$61.64	Large-scale training	Deploy →
AWS	H200 141GB	p5e.48xlarge	1128 GB (8×141)	$104.00	$38.0063% off	$68.00	Optimized LLM Inference	Deploy →
GCP	H200 141GB	a3-megagpu-8g	1128 GB (8×141)	$105.00	$31.0070% off	$68.00	Optimized LLM Inference	Deploy →
Azure	MI355X 288GB	ND_MI355X_v6	2304 GB (8×288)	$108.00	$32.4070% off	$70.00	CDNA 4 Frontier Training	Deploy →
GCP	B200 192GB	a4-highgpu-8g	1536 GB (8×192)	$110.00	$33.0070% off	$72.00	Next-Gen Frontier Training	Deploy →
Azure	B200 192GB	ND_B200_v6	1536 GB (8×192)	$112.00	$33.6070% off	$73.00	Next-Gen Frontier Training	Deploy →
AWS	B200 192GB	p6.48xlarge	1536 GB (8×192)	$115.00	$42.0063% off	$75.00	Next-Gen Frontier Training	Deploy →
GCP	B300 288GB	a5-ultragpu-8g	2304 GB (8×288)	$142.00	$42.6070% off	$92.00	Frontier Model Training (Blackwell Ultra)	Deploy →
Azure	B300 288GB	ND_B300_v7	2304 GB (8×288)	$145.00	$43.5070% off	$94.00	Frontier Model Training (Blackwell Ultra)	Deploy →
AWS	B300 288GB	p7.48xlarge	2304 GB (8×288)	$148.00	$52.0065% off	$96.00	Frontier Model Training (Blackwell Ultra)	Deploy →

Use Spot for Training

Save 60-70% on training runs with checkpointing. GCP Spot offers up to 70% discount on A100/H100 instances.

Best for: Fault-tolerant training with checkpoints

Reserved for Inference

1-year commitments save 35-40% for always-on inference endpoints. Azure and AWS offer the deepest reserved discounts.

Best for: Production inference workloads

Lambda/CoreWeave for Value

GPU cloud specialists offer 2-4x lower per-GPU pricing than hyperscalers, ideal for teams that don't need full cloud ecosystems.

Best for: Pure GPU compute without cloud services

Prices are approximate and vary by region and availability. Pricing reflects May 2026 estimates — always verify with provider pricing pages before procurement.

Ask AI Advisor