AI project activity across neoclouds, sovereigns and enterprise infrastructure Report Interpretation

Overview

Goldman Sachs’ August 2026 AI Project Pulse compiles selected announcements across neoclouds, sovereign AI and enterprise deployments. The report’s central message is that large AI infrastructure commitments remain active across GPUs, servers, networking, data-center capacity and associated power systems, with regulatory constraints emerging as a potential timing issue for less well-capitalized projects.

Core views

Goldman Sachs summarizes selected August announcements rather than the total AI-infrastructure opportunity. Its overview identifies continued project activity across neoclouds, sovereign deployments and enterprises, while flagging a new development constraint in the United States: Texas and Pennsylvania governors issued directives or executive orders on August 3 and August 18, respectively, tightening regulatory, energy and environmental guardrails for new data centers. The report notes that these measures could create uncertainty and slow project development, but Goldman Sachs Utilities analyst Carly Davenport expects minimal effect on high-quality, large-scale projects; speculative and under-capitalized developments are expected to bear the greater burden. Within Goldman Sachs’ US IT hardware coverage, DELL and SMCI are identified as having the most exposure to neoclouds building greenfield AI data centers. SMCI’s August 11 F4Q26 results provided a hardware-demand read-through. The company introduced FY27 revenue guidance of $65-$72 billion for the year ending June 2027, representing 75% year-over-year growth. Management cited more than $60 billion of orders received in F4Q, with roughly 70% tied to pure-AI deployments and GPU servers and roughly 30% tied to CPU-based AI and CPU demand. Goldman Sachs interprets this as support for continued GPU-server demand benefiting DELL as well as SMCI, while strong CPU-based server demand should also support DELL and HPE. The project announcements illustrate the breadth and scale of demand. IBM and Together AI signed a multi-year $240 million agreement to deploy a large-scale inference cluster on IBM Cloud using NVIDIA HGX B300 systems and Spectrum-X Ethernet networking, with availability expected in Q1 2027. Fermi’s binding lease agreement with TensorWave covers a 222-MW turnkey data-center solution in its first phase, approximately $6.5 billion of contracted revenue over a 15-year term beginning in the second half of 2027, and expansion rights up to 650 MW. The facility in Carson County, Texas is designed to support tens of thousands of next-generation AMD Instinct GPUs through a behind-the-meter power mix including combined-cycle natural gas, nuclear power, grid supply, solar and battery storage; first power is targeted for 2026. Sovereign AI projects are also prominent. HUMAIN, AMD and Cisco brought a Saudi Arabian AI platform live using AMD Instinct MI355X GPUs, AMD EPYC CPUs and Cisco Silicon One-based AI networking. The next phase is planned at up to 250 MW beginning in 2027, with capacity expected in H2 2027, and the related joint venture targets up to 1 GW by 2030. In India, IntelliDB Enterprise and Yotta Data Services announced sovereign, AI-ready database infrastructure; Yotta’s previously announced deployment includes 20,736 liquid-cooled NVIDIA Blackwell Ultra GPUs, of which 10,000 are allocated to the India AI Mission. Mistral and HUMAIN also announced a collaboration valued in the hundreds of millions of euros to develop sovereign infrastructure, localized cybersecurity and voice models, and Arabic-language frontier models for regulated industries. Additional deployments reinforce demand for integrated hardware systems. Firebird’s Armenia project has brought phase one online with a $500 million first-phase investment and plans to scale beyond 400 MW using Dell PowerEdge servers with NVIDIA Blackwell GPUs. In the UK, Era4 is deploying sovereign, lower-carbon modular data centers using HPE AI Mod PODs with NVIDIA Blackwell B300 GPUs and direct liquid cooling; its GPU inventory is planned to grow from 1,500 to more than 10,000 units within 12 months across access to more than 40 UK sites. Cisco and Supermicro expanded their Secure AI Factory with NVIDIA, offering rack-scale liquid- and air-cooled infrastructure for trillion-parameter training and high-throughput inference, with Supermicro systems available through Cisco channel partners from October 2026. The report also records CoreWeave’s multibillion-dollar agreement with Hudson River Trading for NVIDIA Vera Rubin NVL72 and HGX B200 systems, and Dell’s deployment at ZyBiSys supporting 99.99% availability across six Indian data centers and more than 50 stockbrokers.

Analysis framework

The report tracks selected monthly AI-project announcements across neocloud, sovereign and enterprise customers. It organizes each announcement around partners, deployed hardware, capacity or investment size, location, timing and implications for covered hardware suppliers, while distinguishing the selected examples from the total addressable AI-infrastructure opportunity.

Methodology notes

• Other

  Monthly AI project announcement tracking

  Goldman Sachs compiles selected public project announcements and compares their hardware, capacity, timing and customer type to identify demand read-throughs for AI infrastructure suppliers.

Asset mapping & comparison

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

• SMCI

  Covered hardware supplier with exposure to neocloud greenfield AI data centers and reported strong AI-server order demand.

  Strengths

  FY27 revenue guidance of $65-$72 billion and more than $60 billion of F4Q orders, with about 70% tied to pure-AI/GPU-server deployments.

  Comparison

  Goldman Sachs sees DELL as another beneficiary of GPU-server demand and DELL and HPE as beneficiaries of CPU-based server demand.

  Risks

  New data-center guardrails could slow speculative or under-capitalized developments.

• DELL

  Covered hardware supplier identified as exposed to neocloud greenfield AI data-center buildouts.

  Strengths

  The report expects continued GPU-server demand across neoclouds and enterprises to benefit DELL; it also identifies strong CPU-based server demand as a positive read-through.

  Comparison

  Shares GPU-server demand exposure with SMCI and CPU-server demand exposure with HPE.

  Risks

  Potentially slower project development amid tighter data-center regulatory, energy and environmental requirements.

• HPE

  Covered hardware supplier positioned to benefit from CPU-based server demand and sovereign AI deployments.

  Strengths

  Goldman Sachs expects uplift from strong CPU-based server demand; HPE AI Mod PODs are used in Era4’s UK deployment.

  Comparison

  The report groups HPE with DELL for CPU-based server demand read-throughs.

• AMD

  AI accelerator and CPU supplier in sovereign and data-center projects.

  Strengths

  AMD Instinct MI355X GPUs and EPYC CPUs power HUMAIN’s Saudi AI platform; Fermi plans tens of thousands of next-generation AMD Instinct GPUs.

  Comparison

  AMD systems feature in the Saudi project, while NVIDIA systems feature in several other tracked deployments.

• NVIDIA

  GPU and networking technology supplier across tracked AI infrastructure deployments.

  Strengths

  The report cites HGX B300, Blackwell, Vera Rubin, HGX B200 and Spectrum-X deployments across IBM Cloud, Yotta, Firebird, Era4, Cisco/Supermicro and CoreWeave projects.

  Comparison

  NVIDIA hardware is deployed across multiple projects alongside AMD-based infrastructure in Saudi Arabia and Texas.

Key data

• SMCI FY27 revenue guidance$65-$72 billionYear ending June 2027; implies 75% year-over-year growth.
• SMCI F4Q orders receivedMore than $60 billionApproximately 70% tied to pure-AI deployments/GPU servers and approximately 30% to CPU-based AI/CPU demand.
• IBM and Together AI agreement$240 millionMulti-year IBM Cloud inference-cluster agreement; expected availability in Q1 2027.
• Fermi-TensorWave first phase222 MW and approximately $6.5 billionTurnkey data-center lease; contracted revenue over a 15-year term beginning in H2 2027.
• HUMAIN AI infrastructure planUp to 250 MW, targeting up to 1 GW by 2030Next-phase deployment planned from 2027, with capacity expected in H2 2027.
• Yotta NVIDIA deployment20,736 liquid-cooled NVIDIA Blackwell Ultra GPUsPreviously announced February 2026; 10,000 GPUs allocated to the India AI Mission.
• Firebird Armenia project$500 million first-phase investment; more than 400 MW plannedPhase one of sovereign AI infrastructure has come online.
• Era4 UK GPU deployment1,500 to more than 10,000 GPUs within 12 monthsDeployment has access to more than 40 UK sites.

Impact & implications

The report indicates that AI infrastructure demand is extending beyond GPU purchases into integrated server, networking, liquid-cooling, power and data-center systems. Goldman Sachs specifically sees SMCI’s order outlook as supportive of GPU-server demand for DELL and SMCI and CPU-server demand for DELL and HPE, while regulatory requirements may weigh more on speculative greenfield data-center development than on large, well-capitalized projects.

Risks

• Texas and Pennsylvania’s tighter regulatory, energy and environmental guardrails could create uncertainty and slow new data-center project development, particularly for speculative or under-capitalized projects.