AI infrastructure projects continued to expand in June, with GPU clusters, sovereign cloud, and data center power emerging as core themes

Overview

This report is the June 2026 AI Project Pulse published by Goldman Sachs’ Americas technology hardware team, focusing on the latest project announcements in neocloud, sovereign cloud, and enterprise AI infrastructure. The report emphasizes that key developments in June included CoreWeave validating the NVIDIA Vera Rubin NVL72 rack, SMCI raising financing to support AI server orders, Vultr selecting HPE and NVIDIA to build next-generation AI infrastructure, and progress in sovereign cloud projects in Turkey and India.

Core views

The core view is that AI infrastructure buildout remains in a phase of intense expansion, with demand for GPUs, AI servers, liquid cooling, networking, data center power, and sovereign cloud jointly driving industry strength. The report does not equate these announcements with the full market opportunity, but through project value, GPU counts, power scale, and deployment timing across multiple projects, it demonstrates order visibility and geographic expansion for AI compute infrastructure in 2026 and beyond.

Analysis framework

The report uses a project announcement tracking approach, organizing AI infrastructure events that occurred in June in reverse chronological order and classifying them into neocloud, data center operators, sovereign cloud, and enterprise projects, extracting key partners, GPU or server configurations, contract size, location, and deployment cadence for each project.

Methodology notes

• Project trackingAI Project Pulse announcement tracking

  Selective project announcement summary

  The report explicitly states that the listed projects are selective announcements and do not represent the full AI infrastructure opportunity set, making the report more suitable as a source of clues on orders, capacity, and regional expansion rather than a complete measure of market size.

• Factor frameworkGS Factor Profile

  Comparison across four attributes: Growth, Financial Returns, Multiple, and Integrated

  The appendix states that Goldman Sachs uses growth, financial returns, valuation multiples, and integrated percentile rankings to compare stocks with the market and industry peers, but the main body of this report does not present those factor results company by company.

• M&A frameworkM&A Rank

  Tier 1 to Tier 3 probability scoring for M&A

  The appendix states that Goldman Sachs uses a 1-to-3 scale to assess the probability of a company becoming an acquisition target and includes an M&A component in some target prices; the main body of this report does not provide specific company M&A Ranks.

Asset mapping & comparison

Structured mapping from thesis to named assets (strengths, weaknesses, peers, risks).

• NVIDIA CORP（US.NVDA）

  Core provider of GPUs, GB300/B200/B300, Vera Rubin NVL72, Spectrum-X, and NVLink platforms.

  Strengths

  Multiple projects directly use NVIDIA GPUs and system architectures, demonstrating its ecosystem advantages in AI training, inference, and rack-scale systems.

  Weaknesses

  Project execution depends on customer capex, data center power, supply chain conditions, and the production ramp of new platforms.

  Comparison

  Compared with server integrators and data center operators, NVIDIA is positioned further upstream in the compute platform and networking ecosystem.

  Risks

  If the 2H26 production ramp, customer deployments, or AI demand fall short of expectations, the pace of related order conversion could slow.

• HEWLETT PACKARD ENTERPRISE CO（US.HPE）

  Vultr selected HPE and NVIDIA to deploy next-generation large-scale AI data center infrastructure.

  Strengths

  HPE has system integration opportunities in enterprise infrastructure, private cloud, and AI cluster deployment.

  Weaknesses

  The report does not disclose project revenue scale or margins for HPE, so the investment impact still requires validation from subsequent orders and deliveries.

  Comparison

  HPE appears alongside NVIDIA in the Vultr project, with a role more focused on systems and infrastructure solutions.

  Risks

  Competition, delivery complexity, and implementation risks related to liquid cooling and high-density GPU clusters may affect project returns.

• COREWEAVE INC（US.CRWV）

  The first AI cloud provider to complete full operation and validation of the NVIDIA Vera Rubin NVL72.

  Strengths

  Being first to validate a new-generation rack-scale architecture helps strengthen its technology positioning in neocloud and high-performance AI cloud.

  Weaknesses

  It is highly capital intensive and sensitive to GPU supply, power, liquid cooling, customer demand, and financing conditions.

  Comparison

  Compared with traditional cloud vendors, CoreWeave is positioned in the report as a frontier neocloud focused on AI cloud and new architecture validation.

  Risks

  Large-scale deployment of the new platform, customer concentration, and capex payback cycles are sources of uncertainty.

• GORILLA TECHNOLOGY GROUP INC（US.GRRR）

  Participates in an approximately $2 billion AI infrastructure agreement related to Yotta sovereign cloud.

  Strengths

  The project is large in scale and involves B300, B200, networking equipment, and the GPUaaS platform, reflecting sovereign cloud opportunities.

  Weaknesses

  Project execution depends on Super Micro Computer, Yotta Data Services, and the realization of subsequent regional expansion.

  Comparison

  Compared with NVIDIA and server vendors, Gorilla is more exposed to the project and platform implementation layer.

  Risks

  Sovereign cloud projects may be affected by financing, regulation, regional expansion, and deployment progress.

• Super Micro Computer（SMCI）

  The report mentions that SMCI plans to raise up to $7 billion and fulfill about $39 billion in AI server orders, while also participating in multiple sovereign cloud and data center projects.

  Strengths

  Its order scale and project visibility are notable, making it a direct beneficiary of expanding AI server demand.

  Weaknesses

  Large component purchases and future-quarter deliveries place high demands on supply chain, working capital, and execution capability.

  Comparison

  Compared with GPU suppliers, SMCI is more directly exposed to server integration, delivery, and hardware gross margin volatility.

  Risks

  Financing terms, customer order conversion, component supply, and delivery delays may affect earnings recognition.

• Genesis Digital Assets（GDA）

  AiOnX acquired its 77% majority stake and plans to convert its crypto mining data centers toward AI and HPC services.

  Strengths

  It owns 15 data centers in the United States and Sweden with 1.3GW of available power, providing a resource base for AI data center transformation.

  Weaknesses

  The transition from crypto mining to AI/HPC requires capital investment, customer acquisition, and infrastructure upgrades.

  Comparison

  This asset is more focused on power and data center site resources rather than being a core supplier of GPUs or servers.

  Risks

  Transformation outcomes may be affected by retrofit progress, AI customer demand, power access, and the regulatory environment.

Key data

• Argentum AI cloud contract$4.1 billion; 27,000 NVIDIA GB300 GPUs; 300MW+; Bielsko-Biala, Poland; phased deployment in 2026This project is a long-term AI cloud infrastructure agreement, indicating demand for large-scale GPU data center construction in Europe.
• Vultr next-generation AI infrastructureNVIDIA GB300 NVL72 systems; Spectrum-X Ethernet 400GbE/800GbE; partners HPE and NVIDIAAimed at private cloud, model training, and inference workloads, reinforcing the roles of HPE and NVIDIA in enterprise AI infrastructure.
• AiOnX acquisition of majority stake in Genesis Digital Assets$500 million; 77% stake; 15 data centers; 1.3GW available power; United States and SwedenThe transaction aims to redirect the power and site resources of crypto mining data centers toward AI and HPC services.
• Datasection Thailand GPU deployment4,696 NVIDIA B200 GPUs; 587 servers; about $257.4 million; 10MW; BangkokThe project is expected to begin phased operations in July and August 2026, serving an unnamed U.S. technology company.
• CoreWeave Vera Rubin NVL72 validation72 Rubin GPUs; 36 Vera CPUs; 260 TB/s NVLink; Dell PowerEdge XE9812; Micron 7600 SSDsCoreWeave became the first AI cloud provider to complete full operation and validation of the NVIDIA Vera Rubin NVL72 rack-scale architecture.
• Turkey Odine sovereign cloudNVIDIA-validated GPU systems; AI factory solutions; location Türkiye Istanbul; partners Odine and SupermicroThe project shows continued expansion in demand for sovereign AI infrastructure and multi-cloud management.
• Gorilla Technology and Yotta sovereign cloud projectabout $2 billion; 20,736 B300 cards; 5,120 B200 cards; India with planned expansion across multiple Asian locationsThe project is associated with Super Micro Computer and Yotta Data Services, focusing on GPUaaS and sovereign cloud infrastructure.
• SMCI financing and ordersup to $7 billion financing; about $39 billion in AI server orders; more than 20 customersThe financing will be used to procure components to fulfill AI server orders in coming quarters.
• Global server TAMGoldman Sachs raised global server TAMThe increase was driven by AI server rack ramp-up and higher memory costs boosting CSP spending.

Impact & implications

For investors, the report reinforces multiple key themes across the AI infrastructure value chain: GPU platform upgrades benefit NVIDIA and related networking, liquid cooling, and server ecosystems; AI server orders and component procurement support the server supply chain; and data center power, geographic footprint, and sovereign cloud demand are becoming critical constraints for project execution. Because the report does not provide stock ratings or target prices, its investment implications are more oriented toward industry momentum and order clues rather than direct trading recommendations.

Risks

• The projects listed in the report are selective announcements and do not represent the full AI infrastructure opportunity set, so they cannot be directly extrapolated into a complete TAM.
• AI data center projects are highly dependent on power, liquid cooling, networking, GPU supply, and construction progress; any bottleneck could delay revenue recognition.
• Large AI server orders and GPU procurement carry risks related to financing, customer concentration, component supply, and future-quarter deliveries.
• Sovereign cloud projects may be affected by policy, data sovereignty, cross-border expansion, and the execution capabilities of local partners.
• If the production ramp of next-generation GPUs and rack-scale architectures falls short of expectations, it could affect demand timing across the value chain.

What to watch

• Whether the 2H26 production ramp of NVIDIA Vera Rubin NVL72 progresses according to plan.
• Delivery, revenue recognition, and financing progress for SMCI’s approximately $39 billion in AI server orders in coming quarters.
• GPU deployment and data center commissioning timelines for projects such as Vultr, CoreWeave, Argentum AI, and Datasection.
• Whether sovereign cloud projects in India, Turkey, and other Asian markets continue to expand in capacity.
• Whether the higher AI server TAM can translate into sustained orders for servers, memory, networking, liquid cooling, and power infrastructure.