As AI adoption accelerates, GPU pricing is evolving into a dynamic reflection of global compute supply. Our latest analysis tracks median GPU prices for NVIDIA’s A100 and H100 across U.S. regions — comparing traditional hyperscalers and emerging marketplace providers.

This regional lens reveals how the economics of AI infrastructure are changing — with marketplaces driving cost efficiency and regional imbalances shaping compute accessibility.

A100 Pricing: Marketplaces Lead in Cost Efficiency

Across all major U.S. regions, marketplace A100s are consistently cheaper than hyperscaler offerings.

• US East and Central regions show steady discounts of 30–50% compared to hyperscaler prices.

• The US South demonstrates the largest gap, suggesting deeper liquidity or underutilized capacity in secondary data centers.

• In the US West, the gap narrows slightly but remains meaningful — even amid higher regional demand near AI hubs like the Bay Area and Seattle.

Key takeaway: The A100 market has stabilized, and marketplaces now provide predictable, lower-cost options for buyers prioritizing flexibility over long-term reservations.

H100 Pricing: Hyperscalers Command Premiums, But Marketplaces Are Closing the Gap

The H100 market — representing the newer generation of NVIDIA’s compute hardware — shows clear price stratification between sources.

• Hyperscalers currently price H100s at 2–3× the marketplace rate, reflecting limited allocation, reservation-based models, and enterprise SLAs.

• Marketplaces, while more fragmented, are becoming a serious alternative: their median prices across all regions indicate growing supply and stabilization.

• The South and Central regions show the most aggressive marketplace pricing, hinting at emerging compute liquidity beyond traditional cloud zones.

Key takeaway: For H100s, hyperscalers still dominate capacity, but marketplace pricing is becoming too efficient to ignore — especially for workloads that can tolerate more flexible provisioning.