Report Highlights

Cloud HPC in the U.S.: Market Overview

The U.S. cloud high performance computing market reached an estimated USD 9.4 billion in 2024 and is forecast to reach USD 23.9 billion by 2032. Hyperscale providers dominate supply: Amazon Web Services, Microsoft Azure, Google Cloud and Oracle Cloud Infrastructure control most on-demand GPU and CPU cluster capacity, while specialist operators such as CoreWeave and Lambda have built dedicated AI-training clusters. Private enterprise has led commercial adoption, particularly in pharmaceuticals and financial services.

Government has shaped the market's form more than its volume. The Department of Energy built the world's leading public supercomputers through on-premises procurement rather than cloud, including Frontier at Oak Ridge National Laboratory, the first exascale system in 2022, and El Capitan at Lawrence Livermore National Laboratory, ranked first on the TOP500 list in November 2024. Federal policy has therefore produced two markets: publicly owned HPC procured by DOE and the National Nuclear Security Administration, and commercial cloud HPC purchased through authorised federal contracts.

Policy-Driven Growth in U.S. Cloud HPC

The CHIPS and Science Act of 2022 allocates USD 52.7 billion to semiconductor manufacturing and research, including USD 39 billion in manufacturing incentives administered by the Department of Commerce. By funding domestic fabrication of advanced accelerators, it eases the chip scarcity that gates cloud HPC capacity. The National Science Foundation's National Artificial Intelligence Research Resource pilot, launched in January 2024, extends compute allocations to vetted researchers, converting grant-funded teams into recurring buyers of commercial GPU hours.

The Joint Warfighting Cloud Capability contract, awarded in December 2022 with a USD 9 billion ceiling to Amazon Web Services, Google, Microsoft and Oracle, obliges the Department of Defense to procure cloud services at Impact Levels 2 through 6. This establishes a multi-year demand floor for government-grade HPC, because defence simulation, logistics and intelligence analytics migrate onto authorised platforms under contract vehicles. Providers therefore compete for workloads that would otherwise remain on government-owned clusters.

Regulatory Barriers and Compliance Costs

Cloud providers selling HPC to federal agencies must obtain FedRAMP authorisation, administered by the General Services Administration's FedRAMP Program Management Office under the FedRAMP Authorization Act, enacted within the FY2023 National Defense Authorization Act. A FedRAMP High authorisation requires more than 400 NIST SP 800-53 Revision 5 controls, typically takes 12 to 18 months, and commonly costs several million dollars in assessment and remediation. Defence workloads add Impact Level 5 authorisation under the DoD Cloud Computing Security Requirements Guide, issued by the Defense Information Systems Agency, which extends timelines further.

The Bureau of Industry and Security administers export controls under the Export Administration Regulations, classifying advanced accelerators under ECCNs 3A090 and 4A090 since its October 2022 interim final rule. Providers must verify foreign customers under know-your-customer obligations for infrastructure-as-a-service, and licence reviews for restricted destinations take months. Power is the slowest barrier: the Federal Energy Regulatory Commission governs wholesale transmission, while state commissions such as the Virginia State Corporation Commission set large-load tariffs, and Northern Virginia interconnection queues now run for several years.

Policy-Created Opportunities in U.S. Cloud HPC

The National Artificial Intelligence Research Resource is moving from pilot toward a sustained national service, creating a procurement channel for cloud providers that supply credits, GPU clusters and secure data enclaves to NSF-designated resource providers. NSF's ACCESS framework, which succeeded XSEDE, already allocates compute across university and commercial systems, giving qualified operators a standardised onboarding route to a publicly funded research customer base that is otherwise difficult to reach.

The Department of Defense's Cybersecurity Maturity Model Certification program, codified at 32 CFR Part 170 and effective from December 2024, requires defence contractors handling controlled unclassified information to obtain third-party certification. Contractors running engineering simulation on cloud HPC must move workloads into compliant enclaves, creating demand for managed, CMMC-attested HPC environments. Phased DFARS enforcement beginning November 2025 makes that demand contractually binding across the defence industrial base.

Cloud HPC at a Glance

Leading Cloud HPC Participants

• Amazon Web Services
• Microsoft Azure
• Google Cloud
• Oracle Cloud Infrastructure
• IBM Cloud
• CoreWeave
• Lambda
• Hewlett Packard Enterprise
• Penguin Solutions
• Rescale

Regulatory and Policy Environment

The primary legislation is the FedRAMP Authorization Act, enacted within the FY2023 National Defense Authorization Act, which codified FedRAMP as the government-wide cloud authorisation programme. It operates alongside the Federal Information Security Modernisation Act of 2014, which requires agencies to secure cloud systems against NIST SP 800-53 controls, and the DoD Cloud Computing Security Requirements Guide, administered by the Defense Information Systems Agency. Key requirements include continuous monitoring, U.S. data residency for high-impact workloads, and U.S. person restrictions on administrative access at Impact Levels 4 and 5.

Upcoming changes include the FedRAMP 20x initiative, launched in 2025 to automate authorisation evidence and shorten assessment timelines, and CMMC phased enforcement under the DFARS rule beginning November 2025. Export compliance will change with BIS revisions to advanced-computing controls and the proposed infrastructure-as-a-service know-your-customer rule. Compared with regional peers, the U.S. framework is security-led and market-driven, whereas the EU Data Act, applicable from 12 September 2025, and the EU Cybersecurity Certification Scheme for Cloud Services prioritise portability and sovereignty requirements.

Long-Term Policy Outlook for U.S. Cloud HPC

By 2032, the most consequential change will be federal AI infrastructure. The Department of Energy selected federal sites, including Oak Ridge, Idaho National Laboratory, Paducah and Savannah River, for AI data centre development through partnerships with private operators. This shifts part of public exascale capacity from direct procurement toward leased commercial models, expanding the addressable market for cloud HPC providers operating on federal land.

Tax and energy policy will reinforce that shift. The One Big Beautiful Bill Act, signed in July 2025, restored immediate domestic research expenditure deductions under Section 174, lowering the after-tax cost of in-house software development on HPC platforms. Energy policy will be the binding constraint, with FERC large-load interconnection rules and Texas Senate Bill 6 requiring large facilities to accept curtailment obligations, forcing providers to co-locate generation and reshape regional capacity by 2032.