The evolution of Amazon Web Services from a specialised cloud provider into a foundational layer of the global digital economy invites careful analysis of the competitive forces that now bear upon it. Two distinct but interacting pressures demand attention: the intensifying regulatory scrutiny, most notably under the European Union's Digital Markets Act (DMA), and the dynamics of pricing power in an environment constrained by GPU supply. A Marshallian perspective compels us to move beyond surface-level descriptions and instead examine the time horizons, substitution possibilities, and adjustment mechanisms that will shape outcomes. We must distinguish between temporary frictions and structural shifts, between apparent concentration and durable market power. Only by tracing the organic adaptation of AWS to these pressures can we form a balanced assessment.
The DMA and the Logic of Regulatory Adaptation
The European Commission's preliminary designation of AWS as a gatekeeper under the DMA 5,6,7,9,25,29,32,33 represents a deliberate attempt to alter the architecture of competition in cloud services. The obligations it imposes—interoperability mandates, data portability requirements, anti-steering rules, and restrictions on bundling 7,15,24—are not merely compliance burdens; they are institutional changes that reshape the field of possible business models. In the short run, AWS must absorb the costs of re-engineering its platforms and processes. Yet the longer-run significance lies in the potential unbundling of cloud infrastructure from the higher-level services and productivity tools that have deepened customer embeddedness. Non-compliance carries penalties severe enough to concentrate managerial attention, but the deeper question is whether these rules genuinely open pathways for rival ecosystems or merely impose a uniform cost on all large operators.
AWS's response through the European Sovereign Cloud—with EU-resident personnel, euro-denominated billing, and data residency features 34—illustrates a classic Marshallian adjustment. By adapting its organisational structure to the specific demands of a regional regulatory environment, AWS seeks to preserve its customer relationships while partially insulating the broader global infrastructure. Critics, however, note that legal dependencies on the home jurisdiction persist 20, suggesting that true technological sovereignty may require more fundamental architectural changes 1. The broader trend toward sovereign cloud requirements, across multiple jurisdictions, raises the prospect of a fragmented equilibrium in which cloud services become increasingly localised 1. For a firm whose efficiencies derive from global scale, such fragmentation would represent a secular headwind. Domestically, antitrust discussions around potential structural separation of AWS from Amazon's retail operations 30 remind us that regulatory risk is not confined to a single continent.
GPU Supply Constraints and the Economics of Pricing Power
The observed 20% increase in EC2 Capacity Block reservation prices for GPU instances 10,11,12,13,28—with combined six-month fee increases reaching 35% 12—is a direct consequence of demand outstripping near-term supply. Tight GPU markets and strong enterprise appetite for AI workloads 10 have granted AWS a degree of pricing power that warrants careful scrutiny. But we must distinguish between a temporary quasi-rent earned during a supply squeeze and a structural ability to price above long-run competitive levels. The latter would depend on barriers to entry, customer switching costs, and the elasticity of substitution across cloud providers.
Here the strategic dimension becomes apparent. Trainium-based instances are explicitly exempt from these price hikes 12. This differential pricing signals not only a preference for steering customers toward AWS's proprietary silicon but also a calculated investment in building an installed base that, over time, reduces reliance on NVIDIA GPUs and strengthens the internal chip ecosystem. It is a quiet form of vertical integration—using pricing incentives to accelerate the adoption of a substitute input whose supply AWS partially controls. The capacity constraints at TSMC, which fabricates Trainium chips 8,41, remind us that even this strategy is subject to upstream bottlenecks. The long-run promise of custom silicon is tempered by the real-world frictions of advanced semiconductor manufacturing, where lead times are measured in years.
Custom Silicon as an Evolutionary Pathway
The rapid sell-out of Trainium chips 8 and the complete exhaustion of third-generation inventory 37 confirm strong demand, even as the next major iteration (Trainium4) remains more than a year away 8. AWS’s custom-silicon business has reached a $20 billion annual run rate 19,31, signalling a credible shift in the industry’s supply structure. Historically, AWS kept these chips internal to drive cloud adoption 8, but the exploration of external sales 17,18,37 indicates a strategic evolution. Selling Trainium to third parties could expand the ecosystem and potentially create a new revenue stream, but it also introduces complexities in managing supply allocation and competitive dynamics. Graviton, by contrast, remains exclusive to AWS for now 17—a choice that preserves a tight coupling between chip and platform and thus a stronger form of customer lock-in.
From a Marshallian standpoint, the interesting feature of this chip strategy is not its immediate financial contribution but its effect on the long-run elasticity of substitution for AWS's compute services. By developing credible in-house alternatives to dominant third-party GPUs, AWS gradually shifts the demand curve it faces, reducing the quasi-rents that external suppliers can extract and increasing the range of competitive responses it can deploy. The partnership with TSMC, while critical, introduces a dependency that is common to the industry; the challenge lies in managing the time profile of capacity expansions so that internal chip adoption does not outstrip the ability to manufacture.
Cost Governance and the Frictions of Trust
AWS’s introduction of the FinOps Agent 23,35,36, expansion of Compute Optimizer idle detection 35, and integration of natural-language querying into Cost Explorer 35 are incremental improvements in a domain where customer frustration has real strategic consequences. Persistent gaps—such as the absence of a universal hard spending cap 35, billing feed delays that slow anomaly detection 35, and automated alerts for credit application failures 39—represent frictions that, in the long run, can erode trust. A customer confronting unexpected charges from reserved IPs 40 or finding support staff unable to access billing details 40 may reassess the value of staying within a single-provider ecosystem. These are not merely operational shortcomings; they are points at which the customer relationship, which AWS has so carefully cultivated, becomes strained. In a world where regulatory changes may increase switching options, such frictions could lower the effective barriers to migration.
Strategic Implications and Time Horizons
The interaction of regulatory pressures and pricing dynamics must be assessed across different temporal frames. In the short run, AWS enjoys considerable latitude: GPU scarcity supports pricing power, and the full operational impact of DMA rules is only beginning to be felt. But the medium term tells a more complex story. As custom silicon matures, it alters the competitive balance with NVIDIA and potentially opens new revenue channels; as sovereign cloud investments accumulate, they may partially neutralise regulatory risk; as cost transparency improves (or fails to improve), they affect customer loyalty. The long-run picture depends on the cumulative weight of these adjustments and on whether the regulatory push toward interoperability and unbundling succeeds in creating genuinely contestable markets.
We must be cautious about drawing definitive conclusions. The evidence suggests that AWS’s competitive position remains formidable, anchored by 28-31% market share 38, a 38% operating margin 4,26, and a 28% year-over-year revenue growth that revived after a period of slower expansion 2,3,14,16,19,27. Yet the same evidence points to forces that, over time, could reshape the landscape. The self-reinforcing flywheel that funds continuous reinvestment 21,22 is not immune to external friction. Whether regulatory intervention, chip supply bottlenecks, and customer trust deficits will significantly erode AWS’s advantages depends on the speed and completeness with which the firm adapts. As Marshall would remind us, nature does not leap; but it does, eventually, move.