AI Project Pulse, July 2026: neoclouds, sovereign AI, and enterprise use cases continue to expand

Overview

This report is the July 2026 monthly update from the Goldman Sachs Americas Technology Hardware team’s “AI Project Pulse,” covering selected project announcements in neoclouds, sovereign AI infrastructure, and enterprise AI applications. The report emphasizes that these announcements do not represent all AI infrastructure opportunities, but provide important signals for monitoring demand for AI computing, data centers, GPUs, and enterprise AI.

Core views

The core views are: first, chip manufacturers such as NVIDIA are helping neocloud customers sustain GPU capacity expansion through minimum revenue guarantees, revenue sharing, and credit support, thereby expanding the computing capacity available to end customers; second, the continued increase in 1GW-scale data centers and AI Factory projects indicates that investment in sovereign AI and hyperscale AI infrastructure remains underway; third, enterprise AI is moving from pilots into more specific internal tools, risk management, marketing, and operational use cases, which could drive demand for enterprise hardware OEMs, IT distributors, and AI cloud services.

Analysis framework

The report uses an event-tracking approach, summarizing publicly announced projects in July in reverse chronological order. It focuses on changes in project scale, location, timeline, partners, capacity, and business models, and maps them to potentially benefiting hardware, cloud services, and distribution ecosystems.

Methodology notes

• Industry event trackingAI Project Pulse

  Monthly monitoring of project announcements

  Tracks announcements for neocloud, sovereign AI, and enterprise AI projects to observe changes in AI infrastructure demand, capacity construction, and business models.

• Equity research disclosure frameworkGS Factor Profile

  Growth, Financial Returns, Multiple, and Integrated factor profiles

  The disclosure appendix states that Goldman Sachs calculates stock factor percentiles using metrics including forward sales, EBITDA, EPS, ROE, ROCE, and valuation multiples, but the main report does not provide this quantitative profile for any individual stock.

• Equity research disclosure frameworkM&A Rank

  Potential acquisition target scoring

  The disclosure appendix states that Goldman Sachs evaluates the probability of covered companies becoming acquisition targets on a scale of 1 to 3; the main report does not provide an M&A rank for any specific company.

Asset mapping & comparison

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

• NVIDIA

  A core GPU, AI Factory standards, and neocloud financing/revenue-support ecosystem participant

  Strengths

  Ties AI infrastructure construction to platforms including GPU supply, NeMo, Quantum InfiniBand, and Grace Blackwell, while expanding customer capacity through revenue sharing and credit support.

  Weaknesses

  The business model is highly dependent on neocloud customer utilization, end demand, and project execution.

  Comparison

  Compared with traditional hardware supply, NVIDIA plays a deeper role in project financing support, system standards, and ecosystem partnerships.

  Risks

  Excess GPU capacity, weaker-than-expected customer leasing, and data center power or construction delays could affect demand realization.

• DELL、SMCI

  Potential beneficiaries of AI server demand

  Strengths

  Goldman Sachs views sustained strong neocloud demand as positive for leading AI server manufacturers DELL and SMCI.

  Weaknesses

  Affected by GPU supply, customer capital expenditure timing, and server margin volatility.

  Comparison

  More directly benefit from the AI server procurement cycle than companies relying primarily on software or cloud services revenue.

  Risks

  Order concentration, intensifying competition, supply chain bottlenecks, and bargaining power from large customers could pressure profitability.

• DELL、HPE、SNX、INGM

  Potential beneficiaries in the enterprise AI adoption chain

  Strengths

  The expansion of enterprise AI use cases on premises and in the cloud could drive demand for enterprise hardware OEMs and IT distributors.

  Weaknesses

  Enterprise AI budget realization may lag project announcements, and some demand may shift to cloud services rather than on-premises purchases.

  Comparison

  Compared with the neocloud chain, the enterprise AI chain is more dependent on the deployment of internal customer tools and the reallocation of IT budgets.

  Risks

  Enterprise software spending cuts may coincide with broader budget constraints, creating uncertainty around the timing of hardware purchases.

• CoreWeave

  AI cloud platform and high-intensity training workload provider

  Strengths

  Received foundation model training workloads from Flow Traders and is taking on 133MW of critical IT load in Phase I of Galaxy Helios.

  Weaknesses

  Business growth depends on large data center leases and highly capital-intensive infrastructure.

  Comparison

  Within the neocloud sector, it is more focused on AI training cloud platforms and benefits from the outsourcing of computing by enterprises and trading institutions.

  Risks

  Long-term leases, financing costs, customer concentration, and computing price volatility are key risks.

• Nebius

  Participant in AI Cloud and asset-light expansion models

  Strengths

  Introduced an asset-light model in which partners finance and own the infrastructure while Nebius provides system architecture, the software stack, and sales organization.

  Weaknesses

  Highly dependent on partner execution, regional data center quality, and sales conversion capabilities.

  Comparison

  Compared with building data centers in-house, the asset-light model can reduce incremental capital requirements.

  Risks

  Partner financing, data residency regulations, and the pace of capacity commercialization could affect expansion outcomes.

• Microsoft、IBM、Oracle

  Third-party software vendors that Starbucks plans to partially replace

  Strengths

  Retain established enterprise software platforms and existing customer relationships.

  Weaknesses

  Large enterprises’ internally developed AI-assisted tools could replace some traditional software spending.

  Comparison

  The Starbucks case in the report shows that enterprise AI may redirect part of the budget from external software to internal tools.

  Risks

  If more enterprises follow this path, traditional software renewals and professional services spending could come under pressure.

Key data

• Meta/BlackRock El Paso AI data center1GW; $14bn; BlackRock-managed funds 80% ownership, Meta 20% ownership; expected to begin coming online in 2028The project is located in El Paso, Texas, with a lease term of up to 20 years.
• IREN new customer contracts$2.8bn in new customer contracts; 2026 annual recurring revenue target raised to more than $4bnCustomers include Microsoft, NVIDIA, Perplexity, Figure AI, Together AI, and Fluidstack.
• Sharon AI cloud computing agreement5-year, $1.32bn agreement; 132MW total capacity; more than 62,000 NVIDIA GPUs expected to be deployed by mid-2027The project is located in New Zealand, with revenue expected to begin in the first half of 2027.
• Galaxy Helios data centerPhase I to deliver 133MW of critical IT load to CoreWeave; total approved power of 1.63GW, expandable to 3.6GWThe West Texas campus; Phase II’s 260MW is expected to be delivered in the first half of 2027.
• Pantheon Atlas Croatia project€50bn total investment; 1GW total capacity; 800MW available IT load; planned operations in the first quarter of 2029Located in Topusko, Croatia, and designed according to NVIDIA GW-Scale AI Factory standards.
• Firmus/NVIDIA AI Factory360MW; 170,000 NVIDIA accelerators; 2027-2028Located in Batam, Indonesia, using a revenue-sharing and credit-support model.
• Starbucks enterprise AI tools$400mn in annual software spending; target technology budget reduction of approximately $30mn, including approximately $10mn in software spendingPlans to develop AI-assisted internal tools for inventory tracking, maintenance management, and other functions, replacing some third-party software.
• Revolut PRAGMA model26 million user records, 111 countries, 207 billion tokens; model development cycle accelerated by 3-5xTrained using NVIDIA H100 GPUs and Nebius AI Cloud for fraud, credit, marketing, and product recommendations.

Impact & implications

The report’s investment implication is that the AI infrastructure chain remains resilient: chip manufacturers’ support for neocloud business models should benefit GPU and AI server shipments; the expansion of sovereign AI and hyperscale campuses should create sustained demand for power, data centers, networking, storage, and integration services; and the deployment of enterprise AI tools could drive demand for enterprise hardware, cloud services, and IT distribution while creating substitution pressure for traditional software vendors.

Risks

• The report covers selected project announcements and does not represent all AI infrastructure opportunities or the eventual realization of capital expenditures.
• 1GW-scale data center projects face risks related to power access, construction schedules, financing costs, regulatory approvals, and supply chain execution.
• Neocloud revenue guarantees and credit-support models depend on end-customer leasing rates; insufficient demand could amplify risks for chip manufacturers or cloud service providers.
• Enterprise AI applications may aim to reduce traditional software and services budgets, creating substitution pressure for some incumbent software vendors.
• The report does not provide individual stock ratings or target prices; investment conclusions should be considered alongside subsequent company fundamentals and valuation analysis.

What to watch

• Whether NVIDIA’s revenue-sharing, credit-support, and minimum-revenue-guarantee models with neocloud customers expand to more customers.
• Power, construction, and delivery milestones for large data center projects including Meta/BlackRock, Pantheon Atlas, Firmus, and Galaxy.
• Contract conversion, ARR targets, and GPU utilization at neoclouds including IREN, Sharon AI, and Nebius.
• Whether enterprise AI cases spread from Starbucks, Revolut, and Flow Traders to more industries and generate quantifiable hardware or cloud spending.
• Whether subsequent orders, revenue, and margins at companies including DELL, SMCI, HPE, SNX, and INGM reflect AI infrastructure demand.