Research Finder
Find by Keyword
AWS Graviton5: Why Would Enterprises Leave the AWS Cloud for x86 Now?
Fifth-generation processor delivers 3nm Arm efficiency along with formal verification security; improved performance means AWS is potentially reshaping investment logic for infrastructure
12/06/2025
Key Highlights:
- At AWS re:Invent 2025 this week, the company introduced Graviton5, its flagship CPU delivering up to 25% higher compute performance than the previous generation, combining 192 cores in a more efficient design than previous, addressing simultaneous performance and sustainability requirements.
- The new M9g instances have a 5x larger L3 cache and reduce inter-core latency by up to 33%, thus mitigating the inter-die communication penalties and resulting constraints to workload consolidation.
- Graviton5 debuts the Nitro Isolation Engine, utilizing formal verification to provide mathematical proof of workload isolation, representing a potential paradigm shift in cloud security assurance.
- More than half of new AWS CPU capacity runs on Graviton, and has for the last 3 years, with 98% of the top 1,000 EC2 customers, including SAP, Siemens, and Airbnb, already on the platform.
The News
AWS has unveiled Graviton5, the company's most powerful custom processor to date, architected to deliver up to 25% higher compute performance than the previous generation while consolidating 192 cores into a single socket. The announcement, made at AWS re:Invent 2025, introduces EC2 M9g instances featuring a 5x larger L3 cache, 33% lower inter-core latency, and the new Nitro Isolation Engine designed to provide mathematically verified workload isolation. Enterprise customers including SAP, Siemens, and Airbnb have reported performance improvements ranging from 20% to 60% in early testing, with SAP observing a 35% to 60% increase in OLTP query performance on SAP HANA Cloud. Full announcement details available here.
Analyst Take
The timing of Graviton5's arrival carries strategic weight that extends well beyond processor benchmarks. I observe AWS launching this fifth-generation chip precisely as the cloud industry confronts its most significant narrative challenge: the persistent drumbeat of workload repatriation. With recent surveys indicating a continuing majority of CIOs considering moving at least some public cloud workloads back to private infrastructure, AWS appears to be architecting its response directly into silicon. This is not merely a performance upgrade. It reads more like an economic countermeasure.
The contrarian observation here is worth stating plainly: while the industry debates whether enterprises should leave public cloud, AWS seems more focused on ensuring they mathematically cannot justify doing so. Graviton5's value proposition is less about raw computational supremacy and more about systematically eliminating the cost thresholds that trigger repatriation conversations in the first place.
What Was Announced
Graviton5 represents a significant architectural departure from its predecessors - architected to address the inter-die latency challenges present in the previous Graviton4's multi-chip module (MCM) design, consolidating all 192 cores onto a single silicon package/die. This architectural refinement is designed to cut inter-core latency by approximately 33%, benefiting latency-sensitive applications including real-time gaming, high-performance databases, and electronic design automation workloads.
The cache architecture has been substantially expanded, with Graviton5 providing five times more L3 cache than previous generations and 2.6 times more cache per core. This aims to reduce data retrieval delays and improve application response consistency. Memory performance gains accompany faster DDR5 memory speeds, while network and storage bandwidth have increased by 15% and 20% respectively across instance sizes, with the largest configurations offering double the network bandwidth.
Built on TSMC's 3nm process technology, Graviton5 adopts a bare-die cooling approach that AWS claims improves energy efficiency while maintaining thermal performance. The processor also introduces the Nitro Isolation Engine, which employs formal verification techniques to provide what the company describes as mathematical proof that customer workloads remain isolated from one another and from AWS operators. This represents a notable evolution in cloud security assurance methodology.
Market Analysis
The competitive context for this launch has shifted considerably over the past year. AMD now commands 27%-30% of the server CPU market by unit share while capturing roughly 40% of revenue, according to recent data. This discrepancy underscores AMD's success in penetrating high-value, high-core-count segments. Intel's unit share remains dominant at approximately 72% of the x86 server market, though it is losing ground in high-margin segments. Arm-based servers are projected to capture 10-12% of global server CPU revenue within the next two to three years, driven primarily by hyperscalers leveraging custom silicon. Yet the real competitive dynamic is not about market share among processor vendors; it is about whether customers perceive sufficient value to remain in the public cloud at all.
Deloitte's research on hybrid cloud economics suggests that cost remains the primary motivator for enterprises considering workload repatriation once usage crosses certain thresholds. Graviton5 appears designed to perpetually extend that threshold. The combination of higher core density, improved cache efficiency, and single-socket architecture aims to deliver better price-performance at every utilization level, making the economic case for leaving cloud progressively more difficult to construct.
As generative AI inference is becoming a compelling use case for edge computing as enterprises seek to manage suppliers, protect data, and control total cost of ownership. Graviton's high-efficiency Arm-based architecture positions AWS to address this distributed computing paradigm, potentially capturing inference workloads that might otherwise migrate to on-premises or edge deployments.
The x86 response from Intel and AMD has accelerated, with both vendors now offering high-core-count options (up to 192 cores for AMD and 288 cores planned for Intel's Sierra Forest-AP) and emphasizing integrated acceleration units for AI and security workloads. Yet these responses largely compete on traditional performance metrics. Graviton's advantage may lie in forcing competitors to compete on TCO, a historically difficult proposition for the x86 ecosystem given its different power and thermal characteristics.
Looking Ahead
Based on what I am observing, the Graviton5 announcement signals AWS's strategic recognition that the cloud value proposition requires continuous economic reinforcement, not merely technical iteration. The 98% adoption rate among top EC2 customers demonstrates that enterprises have already internalized Graviton's price-performance thesis; the question now becomes whether Graviton5 can extend that thesis to workloads currently sitting on the repatriation consideration list.
HyperFRAME will be monitoring several indicators in the coming quarters. First, whether the promised 25% performance improvements translate into measurable shifts in customer migration patterns and repatriation timelines. Second, how the formal verification approach embodied in the Nitro Isolation Engine influences regulated industry adoption, particularly in financial services and healthcare where security assurance requirements have historically favored on-premises deployments. Third, the competitive response from Intel's Clearwater Forest and AMD's Turin Dense architectures, both expected to emphasize efficiency metrics that more directly challenge Arm's traditional advantages.
The subtext of this launch suggests AWS views infrastructure economics as a renewable competitive moat, one that must be deepened with each processor generation to maintain cloud's fundamental value proposition against the rising tide of repatriation sentiment.
Stephen Sopko | Analyst-in-Residence – Semiconductors & Deep Tech
Stephen Sopko is an Analyst-in-Residence specializing in semiconductors and the deep technologies powering today’s innovation ecosystem. With decades of executive experience spanning Fortune 100, government, and startups, he provides actionable insights by connecting market trends and cutting-edge technologies to business outcomes.
Stephen’s expertise in analyzing the entire buyer’s journey, from technology acquisition to implementation, was refined during his tenure as co-founder and COO of Palisade Compliance, where he helped Fortune 500 clients optimize technology investments. His ability to identify opportunities at the intersection of semiconductors, emerging technologies, and enterprise needs makes him a sought-after advisor to stakeholders navigating complex decisions.