AWS raises GPU prices 20%: second increase in six months

In July 2026, AWS raised prices for its EC2 Capacity Blocks instances — designed for artificial intelligence workloads — by approximately 20%. This is the second increase in less than six months, following a 15% hike announced on January 4, 2026. The cumulative increase reaches nearly 35% for the most in-demand GPU instances, such as the p5e.48xlarge (eight NVIDIA H200 GPUs), which went from $34.61/hour to $39.80/hour in January, and now faces another adjustment. This move breaks a historic trend of more than two decades: since the launch of Amazon EC2 in 2006, cloud computing prices had steadily declined, with over 100 price cuts by AWS. The last increase before 2026 dated back to 2014, when AWS raised prices for some storage services, but never for core compute instances.

Why does it matter?

Historically, cloud computing prices had followed a steady downward trend for over 20 years, driven by Moore's law and competition among providers. This double increase in such a short period breaks that golden rule and signals a structural market shift. The root cause is not speculative: HBM (High-Bandwidth Memory) — essential for NVIDIA and AMD GPUs in AI tasks — is in short supply because manufacturers like SK Hynix, Samsung, and Micron have redirected capacity toward this type of chip, neglecting standard server memory. As a result, DDR4 memory prices have risen 158% since September 2025 and DDR5 prices 307% over the same period, according to TrendForce. Additionally, energy costs have skyrocketed: in the PJM electricity market (which supplies much of the eastern U.S. data centers), the megawatt-day price went from $34 in 2023 to $329 in 2026, an 868% increase in three years. This energy surge is due to growing demand from AI data centers, which consume up to 10 times more power than traditional data centers, and regulatory pressure to reduce carbon emissions. Comparatively, during the semiconductor crisis of 2021-2022, GPU prices for gaming doubled, but cloud computing did not experience widespread increases. Now, the impact is directly on enterprise cloud.

Consequences for businesses and users

For startups and companies that rely on AWS to train and serve AI models, this increase is a direct blow to their margins. Many had already adjusted their budgets after the January hike, and now face another increase. This could accelerate migration to alternatives like Google Cloud, Azure, or specialized GPU providers (CoreWeave, Lambda Labs), or even drive adoption of on-premise hardware in data centers. For example, companies like OpenAI and Anthropic have already begun building their own GPU clusters to reduce dependence on hyperscalers. Large hyperscalers, like AWS, are also signing strategic contracts with governments — such as the Pentagon deal in May 2026 — suggesting they prioritize compute capacity for assured clients over commercial ones. This could lead to a war for access to GPUs, where prices continue to rise for smaller customers. In the stock market, NVIDIA shares rose 12% since the July announcement, while AWS (Amazon) shares fell 2%, reflecting investor concern about the impact on cloud customer profitability. Moreover, the increase could slow AI adoption among SMEs, which already account for 40% of cloud spending according to Gartner, and which may now delay generative AI projects due to budget constraints.

What should readers know?

• Not a temporary adjustment: The HBM shortage and energy surge are long-term trends. Prices are expected to remain high at least until 2027, when new memory factories come online. SK Hynix plans to open a new plant in 2028, but HBM demand has tripled in 2026.
• Viable alternatives: Google Cloud has frozen TPU prices until 2027, and Azure offers reservation discounts of up to 40% for three-year contracts. Smaller providers may offer more competitive prices, but with lower availability and without AWS's service ecosystem. CoreWeave, for example, offers H100 instances at $2.50/hour, compared to AWS's $4.50/hour after the hike.
• Impact on startups: Companies basing their business model on generative AI will need to review their cost structure or seek additional funding to cover the increase. According to a CB Insights report, 30% of U.S. AI startups have already reduced cloud spending in 2026, and 15% have shut down due to unsustainable costs.

“AWS's price hike is a symptom of an AI hardware market reaching its physical and geopolitical limits. The era of cheap, unlimited cloud is over.” — Analyst at TheVortiq

In summary, AWS's price increase is not an isolated event but a reflection of a confluence of structural factors: HBM memory shortage, skyrocketing energy costs, and AI demand outstripping GPU supply. Customers should diversify providers, consider long-term reservations, and evaluate on-premise options to mitigate the impact. The window of cheap cloud has closed, and the new normal will be higher prices and competition for scarce resources.