Global AI Infrastructure Dynamics in May: New Clouds Enhance Software Services; Enterprise Hybrid Deployment Demand Expands

Overview

This report is the monthly update of Goldman Sachs' Americas Technology Hardware team's 'AI Project Pulse' (May 2026), systematically compiling recent announcements and developments in global AI infrastructure across new cloud providers, sovereign projects, and enterprise clients. The core conclusion is that the industry is undergoing a critical transition: on the supply side, providers are extending beyond pure GPU compute to integrate software orchestration and inference optimization; on the demand side, enterprise deployment scenarios are rapidly broadening, with hybrid and localized solutions becoming the dominant trend.

Core views

New cloud strategic focus has clearly shifted upstream. IREN acquired Mirantis (a Kubernetes orchestration provider) for approximately $625 million, and Nebius acquired Eigen AI (an inference optimization company) for $643 million, aiming to deeply embed orchestration and model optimization capabilities into GPU infrastructure to extend services from model training research environments to enterprise-grade production inference. The report notes that such vertical integrations mark a strategic pivot by new clouds from 'compute suppliers' to 'seamless production-grade infrastructure and managed inference service providers'. Enterprise deployment demand has significantly broadened and deepened. At Dell Technologies World 2026, Dell highlighted cross-industry use cases: Samsung and Mazda for semiconductor design and manufacturing automation; Eli Lilly deploying a 1,016-GPU Blackwell Ultra Lilypad supercomputer, extending into digital twins and simulation for manufacturing; Hudson River Trading building a dedicated AI research data center in Norway; and Mazda planning a unified CAD and model development storage architecture. These cases confirm that AI workloads are permeating beyond internet companies into traditional industries, with hybrid and on-premise deployment models prevailing.

Analysis framework

The report employs a structured monthly tracking framework, categorizing global key project announcements in reverse chronological order under two dimensions: 'New Clouds & Data Center Operators' and 'Sovereign & Enterprise'. Analysis focuses on quantifiable elements: contract values, compute scale (GPU count, MW capacity), technical architecture (Blackwell/TPU/edge layer), deployment scope, and partners, extracting common trends (e.g., software-layer integration, rise of hybrid deployment). All conclusions are derived solely from publicly disclosed project details without external forecasts or subjective extrapolations, offering an objective snapshot of industry dynamics.

Methodology notes

• Industry/Industrial Analysis FrameworkUpstream-downstream transmission

  Layered tracking of interactions and evolution between supply-side (new clouds, data center operators) and demand-side (enterprise clients) in AI infrastructure

  This framework examines how technology deployment propagates from underlying compute to end-user applications. The report classifies projects into 'New Cloud/Data Center' and 'Enterprise/Sovereign' layers, clearly illustrating how supply-side upgrades (e.g., software integration) respond to and drive demand-side scenario expansion (e.g., hybrid deployment), helping readers understand the协同演进 logic of the industry ecosystem.

• Event-Based博弈与行为金融Event-driven analysis

  Systematically tracking key industry events (M&A, large contracts, product launches) to identify trend signals

  Institutions regularly aggregate representative public events to avoid isolated interpretations of isolated points, instead extracting directional patterns (e.g., concentrated new cloud acquisitions of software firms). This report selects flagship projects on a monthly basis, providing readers with a de-noised view of industry dynamics suitable for capturing industrial evolution pace rather than short-term trading signals.

Key data

• IREN-Dell Procurement Agreement$1.6 billionBlackwell system, estimated to include ~26K B200 GPUs (45MW)
• Akamai-Anthropic Cloud Agreement$1.8 billionSeven-year term, covering 130+ countries, leveraging 4,200+ edge nodes
• Blackstone/Google TPU Cloud JV$5 billionInitial equity commitment, 500MW capacity, scheduled to launch in 2027
• NVIDIA-IREN Collaboration5GWDSX architecture deployment scale; NVIDIA receives $2.1 billion in stock purchase rights
• Nebius Acquisition of Eigen AI$643 millionStrengthens inference optimization and model fine-tuning capabilities
• Sharon AI Cloud Deal$950 millionAustralia, five-year term, 1,000 B200 GPU supercluster

Impact & implications

The report suggests that these dynamics collectively indicate accelerating maturity in the AI infrastructure industry: supply-side players are enhancing unit economic value through software and service integration, while demand-side adoption is expanding from internet companies into traditional sectors such as manufacturing, finance, and healthcare, driving deployment models toward hybrid and localized configurations. Continued capital investment and deepening ecosystem partnerships (e.g., Dell collaborating with multiple AI software firms) reflect industry confidence; however, the report does not extend to market impact, competitive landscape, or investment opportunities.