Cloud computing prices are supposed to go down over time. That was the unwritten rule for the better part of a decade. AWS just broke it twice in six months.

Starting July 1, Amazon Web Services will raise prices on its Nvidia-powered GPU cloud instances reserved for machine learning workloads by 20%. The flagship P6-B300 instances will cost $14.04 per hour, while P6-B200 instances come in at $12.355 per hour and P5 instances in US regions will run $5.191 per hour.

The second hike in 2026

This isn’t a one-off adjustment. AWS already implemented a 15% price increase on the same category of instances back in January 2026. Two hikes in a single year, targeting the same GPU capacity blocks, represents a sharp departure from the long-running trend of cloud price deflation that defined the industry through at least mid-2025.

The increases specifically target EC2 Capacity Blocks, which are reserved instances designed for machine learning applications. These blocks guarantee customers dedicated access to accelerator capacity, something you don’t get with on-demand or spot pricing models.

Most other EC2 pricing structures remain unchanged. If you’re running a standard web application or database on AWS, your bill probably looks the same. But if you’re training large language models or running GPU-intensive AI workloads, you’re now paying meaningfully more for the privilege.

AWS has pointed to supply and demand dynamics as the driving force behind these adjustments. Advanced Nvidia GPUs remain in constrained supply, and the appetite for AI training infrastructure shows no signs of cooling.

What the numbers actually mean

Here’s the thing about $14.04 per hour. That sounds manageable until you do the math on a serious AI training run. A single P6-B300 instance running continuously for a month would cost roughly $10,250.

The cumulative impact of both 2026 increases is substantial. A customer who was paying a baseline rate before January has now absorbed a 15% hike followed by a 20% hike. The effective increase from the original pricing is closer to 38%.

And alternatives do exist. Decentralized compute platforms have been marketing themselves aggressively against AWS’s updated pricing, with some advertising GPU rates 60-90% lower than what AWS now charges. Platforms like Akash have positioned themselves as cost-effective options for teams that don’t need the full enterprise wrapper that comes with an AWS relationship.

What this means for investors

For companies heavily reliant on AWS for AI infrastructure, rising GPU costs directly compress margins. Any business whose unit economics assumed stable or declining cloud compute costs now needs to revisit those assumptions.

This is particularly relevant for the wave of AI startups that have raised capital based on financial projections built during the era of cloud price deflation. If your model assumed compute costs would stay flat or decrease, a 38% effective increase in GPU pricing reshapes your burn rate and runway calculations.

The decentralized compute narrative gets a boost from every AWS price hike. Projects offering distributed GPU access at a fraction of centralized cloud costs become more compelling as the price gap widens.

Investors watching the cloud computing space should pay attention to whether this becomes a broader industry pattern. If Google Cloud and Microsoft Azure follow with similar GPU pricing increases, it would confirm that the era of cheap cloud compute for AI workloads is genuinely over. If they hold prices steady, AWS risks losing market share to rivals willing to absorb thinner margins for customer acquisition.