What a GPU-hour costs

Lambda significantly increased on-demand pricing for flagship chips, with H100 rising up to 18 percent to $4.69 and B200 climbing 16 percent to $7.79 per hour. Verda also implemented a broad 3 percent price hike across its entire GPU inventory, including H200 and B200 units. AWS introduced new high-end capacity, listing B300 chips at $21.36 and H200 units at $7.91 per hour. These widespread list-price increases suggest that training and inference budgets should be adjusted upward for upcoming cycles. Leaders should prioritize securing reserved capacity to hedge against these rising on-demand costs.