• AWS will raise reserved GPU compute prices by 15% on October 7, following a 20% increase in July, according to Wells Fargo (WFC).
  • It marks the fourth consecutive quarterly increase, with H100/H200 and B200/B300 pricing now up 40–75% year to date.
  • Wells Fargo sees the pricing power supporting AWS revenue acceleration into early 2027 and maintains Overweight on Amazon (AMZN) ($AMZN) with a $338 target.

Another Price Hike for AWS GPU Capacity

AWS is reportedly preparing another 15% increase in EC2 Capacity Block prices for reserved high-end GPU capacity on October 7, following a 20% July increase. If confirmed, it would underscore an unusually tight market for AI compute and strengthen the case that AWS can translate scarce Nvidia (NVDA) GPU capacity into faster cloud revenue growth—though it also raises customer-cost, competition, and infrastructure-policy risks.

The reported change concerns reserved GPU compute, apparently AWS EC2 Capacity Blocks—not necessarily every on-demand GPU instance price. Capacity Blocks let customers reserve defined GPU capacity for time-bound AI/ML workloads, so the impact will be concentrated among customers seeking guaranteed access to scarce H100/H200 and Blackwell-generation B200/B300 systems.

The headline cites Wells Fargo’s view that the October 7 step would be the fourth consecutive quarterly increase, putting affected high-end GPU reservation prices roughly 40%–75% higher year to date. Public reporting independently characterizes the planned 15% move as applying to AWS GPU Capacity Blocks, following prior increases in January and July; AWS’s official pricing documentation or announcement would be the definitive confirmation.

The economic logic is straightforward: AI labs, enterprises, and cloud customers are competing for limited accelerators, data-center power, networking, and deployable capacity. When customers value availability and speed more than unit cost, a cloud provider can charge more for guaranteed access.

AWS’s Revenue Momentum

Amazon is a global consumer-internet and technology company spanning e-commerce marketplaces, fulfillment and logistics, advertising, subscriptions/Prime, devices and media, and cloud computing. AWS is its cloud-infrastructure division, selling compute, storage, databases, networking, cybersecurity, analytics, AI platforms, foundation-model access through Bedrock, and custom chips such as Trainium and Graviton.

In its latest reported quarter, Amazon net sales hit $200.6 billion, up 20% year over year, while operating income rose 43% to $27.5 billion. AWS revenue climbed 37% to $42.2 billion, and AWS operating income more than doubled to $16.6 billion from $10.2 billion a year earlier. The cloud unit’s annualized revenue run rate stands at about $169 billion, with its AI business and chips business each exceeding $25 billion annualized.

AWS’s Q2 revenue growth was its fastest in 18 quarters. That means the pricing report arrives during a period when demand growth, not merely price, is already lifting AWS results. At the same time, the company’s capital intensity is increasing: purchases of property and equipment rose by $66.1 billion year over year on a trailing-12-month basis, and Amazon attributed the decline in free cash flow—negative $7.6 billion on a trailing-12-month basis—principally to AI investment.

Andy Jassy remains Amazon’s president and CEO, and Matt Garman leads AWS. The material AWS leadership restructuring was implemented in June 2024 following Garman’s succession to Adam Selipsky, including tighter integration of global sales and reorganization of partner, specialist, services, sovereign-cloud, and industry groups. There is no major new AWS executive transition evident in the sources reviewed.

A Broader AI Compute Crunch

This is part of a broader shift from cloud computing as a broadly commoditized utility toward capacity-constrained AI infrastructure. Demand exceeds immediately available supply. Training frontier models and serving AI agents require clusters of advanced GPUs, ultra-fast networking, memory, electricity, cooling, and data-center construction. The bottleneck is therefore wider than Nvidia chip supply alone.

The cloud market is segmenting. Customers can buy AWS hyperscale capacity and enterprise-grade integrations, turn to Azure or Google (GOOGL) Cloud, contract with GPU-specialist clouds such as CoreWeave (CRWV) or Nebius (NBIS), or use lower-cost marketplace providers. The tradeoff is typically price versus reliability, scale, geographic availability, security/compliance, and bundled cloud services.

Competitor evidence supports the scarcity narrative. Nebius reportedly raised AI-cloud GPU prices effective October 1, with H100 pricing up about 17%, H200 20%, B200 19%, and B300 21%, citing demand outpacing capacity.

Pricing is highly differentiated. Published comparisons suggest H100 and B200 rental rates can vary several-fold by provider, region, commitment term, instance configuration, and use of spot/preemptible capacity. That variation means the headline should not be interpreted as a uniform 15% rise across all AWS GPU products or all customers.

Amazon’s response is not solely price-driven. It is expanding supply while trying to improve economics through its own chips. Amazon says OpenAI and Anthropic have made multi-year, multi-gigawatt Trainium commitments, and it reports rapid growth in its AI and chips businesses.

For the U.S. and global economy, the investment cycle supports chip makers, networking suppliers, power producers, construction firms, and data-center operators. But it can also put upward pressure on electricity-system investment, scarce construction inputs, high-bandwidth memory, and enterprise AI budgets. Amazon itself lists energy prices, tariffs/trade policy, memory-chip volatility, inflation, interest rates, and global geopolitical conditions among factors that could affect results.

Policy and Geopolitical Dimensions

High-end AI GPUs are now strategically regulated technology, not ordinary commodity hardware. U.S. export controls continue to restrict advanced chips and systems containing them for China and other controlled destinations, as well as certain entities headquartered in restricted jurisdictions. These rules can alter where cloud capacity gets built, who can use it, and how providers screen customers.

A major unresolved issue is remote access: a customer may seek cloud compute in a third-country data center without physically importing chips. U.S. policymakers are considering how, and how far, export rules should cover this type of AI-cloud access. The proposed Remote Access Security Act passed the U.S. House in January 2026, but its final prospects remained uncertain as of late September.

The policy direction is mixed rather than simply tighter. The administration stopped enforcing the prior AI Diffusion Rule while retaining key restrictions, has applied targeted changes toward China, and moved the UAE into a more favorable export-control category for specified governmental and corporate recipients. That creates both opportunity and compliance complexity for hyperscalers operating global AI infrastructure.

Stakeholders and Public Debate

For AWS and its shareholders, higher reserved-GPU prices can lift revenue per unit of scarce capacity and potentially margins, supporting the bullish thesis in the Wells Fargo note. The key limitation is that revenue only rises if usage and commitments remain strong and customers do not materially shift workloads elsewhere.

AI labs and large enterprises face higher training and inference costs, but may still pay a premium because delay—missing a product cycle or model-training window—can cost more than higher compute rates. Startups, researchers, and smaller developers are most exposed. Higher reservation prices can widen the resource gap between well-capitalized AI firms and smaller builders, encouraging use of open models, smaller models, optimized inference, spot capacity, specialist clouds, and custom silicon.

Local communities and electricity customers see both benefits and burdens. Data centers can bring construction work, skilled jobs, tax revenue, and infrastructure spending. They can also increase demand for grid capacity, water, land, and backup generation. A recent report on rural U.S. development notes that more than 3,000 data centers were operating in the U.S. as of spring 2026, with more than 1,500 in development; it also describes concerns over grid and water pressure.

Environmental and civic groups focus on whether local economic benefits and new energy investment adequately offset energy use, emissions exposure, water consumption, tax concessions, and potential utility-rate effects. Data-center projects have faced increasingly organized local opposition; one tracker estimates $64 billion in U.S. projects have been blocked or delayed.

AWS says it is pursuing energy procurement, recycled/reclaimed water, grid-management participation, and water-efficiency improvements. Amazon reported its global data centers were more than seven times as water-efficient as the industry average and said it had reached 75% progress toward becoming water-positive across global data-center operations by 2030. Those claims are company-reported and do not eliminate local concerns around site-specific resource impacts.

History and Outlook

The current moment follows several linked developments. Generative-AI demand made accelerated computing a strategic resource rather than a niche cloud category. Nvidia H100 systems became a benchmark for frontier-model training, followed by H200 and Blackwell B200/B300 demand. AWS, Microsoft (MSFT), Google, and specialist GPU clouds accelerated multiyear data-center and accelerator investments. Supply constraints shifted negotiating power toward providers with installed, deployable capacity. AWS is now pairing Nvidia-based capacity with its own Trainium strategy, seeking both supply diversification and better long-run economics.

The closest precedent is the early public-cloud era, when compute and storage prices generally fell over time as scale improved. The present AI-infrastructure cycle differs because the immediate constraint is access to leading hardware plus data-center capacity, which has allowed price increases despite the industry’s historic deflationary tendency.

In the near term, if the reported increase takes effect and commitments remain firm, it should support AWS revenue growth into late 2026 and early 2027, consistent with Wells Fargo’s thesis. It could also improve returns on Amazon’s heavy AI infrastructure build-out. Customers are likely to respond by locking in capacity earlier, negotiating longer commitments, shifting noncritical workloads to spot/preemptible instances, using smaller or distilled models, optimizing inference, and evaluating multi-cloud placement. Sustained price inflation creates room for specialized GPU clouds and lower-cost providers. It also gives Azure and Google Cloud an opening to compete through price, capacity guarantees, chips, model ecosystems, and enterprise bundling. Pricing power helps, but Amazon still faces a major investment burden. The company’s Q2 figures already show a large free-cash-flow outflow tied primarily to increased AI property-and-equipment spending.

Longer term, prices may remain elevated while capacity is scarce, but they are unlikely to rise indefinitely. New GPU shipments, hyperscaler data-center build-outs, alternative accelerator supply, and increasingly capable custom chips should eventually ease the shortage. The durable issue is cost per useful AI task, not GPU-hour price alone. Newer chips may cost more per hour but finish workloads faster; similarly, optimized models and custom accelerators can reduce total cost even if headline hourly prices rise. Regulation may become more consequential. Export controls, data-center permitting, electricity interconnection, water requirements, and local tax arrangements could determine where capacity is built and which customers can access it.

The Wells Fargo $338 price target is an analyst opinion, not a forecast certainty. The strongest supporting evidence is AWS’s current 37% growth and high operating income; the main risks are customer price sensitivity, growing competition, capex overruns, chip/memory supply constraints, energy availability, and geopolitical/export-control shocks.

The central takeaway is that the reported price rise is less a standalone billing event than a signal of a wider AI-compute shortage: AWS appears able to monetize scarce reserved capacity today, while simultaneously spending aggressively to ensure that the constraint—and its associated pricing power—does not become a long-term brake on growth.