Digital Infrastructure Monthly Insights

Equinix has launched its Enhanced Performance Innovation Centers (EPIC), a global network of production-ready liquid cooling environments designed to accelerate enterprise AI infrastructure deployments. Now operational in Frankfurt, London, Osaka, Paris and Sydney, the centers enable organizations to test high-density GPU and CPU platforms, validate liquid cooling technologies and evaluate AI training and inference workloads under real-world conditions. Working alongside leading hardware partners and Equinix infrastructure experts, enterprises can assess cooling, power, networking and operational requirements before scaling into production. Building on Equinix's existing innovation and solution validation facilities, EPIC aims to reduce deployment risks, improve infrastructure readiness and accelerate the adoption of high-performance, liquid-cooled AI data centers. Additional locations are planned for 2027.

Taiwan's Chief Telecom has broken ground on a new AI data center at the Central Taiwan Science Park, investing approximately NT$3.3 billion (US$101.7 million) to expand its AI infrastructure capacity. Scheduled to begin operations in the second half of 2028, the facility represents another step in the company's growing AI data center business. The investment builds on increasing demand for AI computing, cloud infrastructure and data center services, strengthening Taiwan's capacity to support enterprise AI adoption and high-performance computing workloads.

Zayo is scaling its digital infrastructure strategy to support growing AI and data center demand. With plans for more than 8,000 miles of fiber backed by its Nvidia partnership, a long-term supply agreement with Corning, and new agentic AI network monitoring tools, Zayo is positioning itself for the next wave of AI-driven connectivity growth.

Cisco is expanding its AI networking portfolio with new silicon, SONiC-based networking capabilities and cross-site deployment solutions designed for large-scale AI infrastructure. As AI clusters grow, the company is addressing rising demands for network bandwidth, power efficiency, cooling and operational coordination across distributed data centers. These innovations aim to simplify AI network management, improve infrastructure scalability and support high-performance AI training and inference workloads, strengthening Cisco's role in next-generation AI data center networking.

GlobalFoundries and TSMC have signed a five-year, $2 billion agreement to manufacture silicon interposers at GlobalFoundries' Malta, New York facility. These critical components enable high-speed communication between AI processors and memory chips. With volume production expected to ramp up in the first half of 2028, the partnership aims to strengthen US semiconductor manufacturing and expand advanced packaging capacity to meet growing AI infrastructure demand.

HPE will provide AMD Helios AI racks with AI processors, along with HPE networking switches and software, to expand Vultr’s United States data center infrastructure under a $1.2 billion agreement.

Finnish telecom operator Elisa is accelerating its AI infrastructure and data center connectivity strategy despite softer profitability in Q3 2026. The company reported revenue of €561 million and comparable EBITDA of €210 million, while its International Software Services business grew 10.9% on a comparable basis. Elisa secured approximately €100 million in cumulative data center connectivity orders during the first nine months, supported by an expanded Google partnership covering facilities in Hamina, Kajaani, Muhos and Vaala. With 5G subscription penetration reaching 61%, growing fiber broadband adoption, and AI-RAN collaboration with Nokia and Nvidia, Elisa is strengthening its position in Finland's expanding AI and digital infrastructure ecosystem.

Hewlett Packard Enterprise (HPE) is expanding its Private Cloud AI platform in collaboration with Nvidia to accelerate enterprise agentic AI adoption and high-performance computing. The enhanced solution supports mixed Nvidia H200 and RTX PRO 6000 Blackwell GPU clusters, with KAI Scheduler optimizing resource allocation and workload placement. HPE also introduced a liquid-cooled Nvidia Vera CPU rack featuring 256 Vera processors, BlueField-4 DPUs and support for 22,500 AI agent sandboxes. According to HPE, the architecture delivers up to 80% faster agentic CPU performance, while Nvidia OpenShell and Sentry technologies strengthen security, governance and workload isolation. These innovations enable enterprises to scale AI inference, autonomous agents and hybrid AI infrastructure while maintaining control over sensitive data and operations.

NVIDIA introduced Minerva, an AIOps platform that unifies telemetry from GPUs, networking, storage, schedulers, and AI workloads into a topology-aware knowledge graph. It applies supervised and unsupervised ML to learn baselines, detect anomalies, predict failures, and correlate events into root-cause incidents. A native AI assistant and integrations with external agentic tools and automation APIs enable natural-language investigations and guided remediation. Initial users and planned adopters include IREN, Aolani, Bitdeer AI, IBM, Jio, Lambda, Quake AI, and Tesla, with OEM distribution via Dell Technologies, Lenovo, and Supermicro. The platform targets improved AI factory yield metrics (GPU uptime, goodput, cluster uptime, token production).

ZTE has launched its ZXEDC M8000 Intelligent Computing Micro-Module at Data Center World Asia, introducing a modular infrastructure solution designed for high-density AI data centers. The platform integrates hybrid air-liquid cooling, 800V high-voltage DC power distribution, intelligent busbars and AI-powered infrastructure management. Supporting rack power densities of 20–60kW and peak capacity of 250kW per rack, the M8000 delivers power supply efficiency of up to 98% and power usage effectiveness (PUE) as low as 1.15, according to ZTE. Its AI-driven optimization can reduce cooling energy consumption by more than 10%. With 345 modules already delivered for a major Chinese computing center project, the solution supports scalable, energy-efficient AI infrastructure and faster data center deployments.

NVIDIA announced the DSX POD reference design for the Vera Rubin platform, enabling partners to preintegrate compute and supporting infrastructure (assembly, rack-to-rack cabling, wiring, piping, and functional validation) in the factory. Standardized, containerized units connect onsite to prepared power, water, and networking and scale into larger AI factories. The design includes DSX MaxLPS for policy-driven power control, dynamic workload orchestration, and high-fidelity telemetry. The announcement was highlighted alongside GTC Berlin updates and leverages an ecosystem of ODM and container solution providers.

NVIDIA announced DSX POD, a factory-built reference design for modular, containerized Vera Rubin AI factories. The units are pre-integrated and validated (compute, networking, cabling, rack power, liquid cooling) and arrive ready to connect to site utilities, reducing deployment from ~6 months to as little as 14 days. A global partner ecosystem will manufacture and deliver starting in early 2027.

The article highlights that AI infrastructure spans compute, memory, storage, and networking, and argues for heterogeneous accelerators and flexible deployment across edge, cloud, and on-prem environments rather than standardizing on one GPU class or a single deployment model.

Dell expanded its AI Factory offering by integrating Nvidia-based acceleration across deskside workstations and data center infrastructure to support enterprise AI workflows.

NVIDIA signed capacity agreements with IBIDEN, Samsung Electro-Mechanics, and Unimicron to raise substrate output 2.5x over three years. Amkor will expand interposer wafer advanced packaging capacity in South Korea by >300% by end-2028. KYEC will add new U.S. testing facilities backed by a $1.1B capacity reservation advance. SPIL targets a 15x increase in optical-engine capacity for co-packaged optics by mid-2027 and plans to double again by end-2028. NVIDIA expects total semiconductor testing capacity to double from 2026 to 2027. Additional regional expansions include SEMCO factories in Sejong and Busan (South Korea) and Unimicron KF substrate capacity in Taiwan, with IBIDEN’s Ono Plant (Japan) ramping additional capacity starting Q1 2028.

Nvidia is streamlining deployment processes to reduce the time customers wait for AI infrastructure to become operational, indicating faster setup for its AI data center offerings.

NVIDIA committed a multibillion-dollar capacity reservation advance to IBIDEN, as disclosed by IBIDEN earlier this year, to secure and expand supply of advanced IC packaging substrates used for AI GPUs and data center accelerators.

Germany-based Agile Robots is expanding its collaboration with Nvidia to accelerate humanoid robotics development, targeting production of 1,000 to 2,000 robots in 2027. Announced at Nvidia GTC Berlin, the partnership introduces the Agile ONE S and ONE SN humanoid platforms, with production planned to begin by the end of 2026. The compact Agile ONE SN integrates Nvidia's Jetson Thor computing platform and Isaac GR00T foundation model, providing advanced AI capabilities for robotic manipulation, learning and physical interaction. Designed for universities and corporate research laboratories, including potential customers Stanford University and ETH Zurich, the platform aims to simplify robotics development and accelerate real-world deployment. Nvidia also introduced Isaac OS and Isaac Lab 3.0, strengthening its physical AI ecosystem for next-generation autonomous machines and industrial robotics.

Dell Technologies announced enterprise AI offerings spanning deskside systems to data center infrastructure, integrating NVIDIA AI hardware/software to support agentic and generative AI workloads on-premises.

Carrier Global is investing more than $600 million in a new advanced manufacturing and engineering facility in Fort Worth, Texas, to meet surging demand for cooling infrastructure from AI data centers. The more than 1-million-square-foot plant will manufacture energy-efficient commercial chillers, with production expected to begin in 2027. The investment will more than triple Carrier's existing commercial chiller production capacity, targeting approximately 3,000 units annually and creating 1,500 jobs. With data center equipment sales projected to reach $2 billion in 2026, Carrier is strengthening its position in AI infrastructure cooling, advanced thermal management and sustainable data center operations as high-density computing workloads drive increasing cooling requirements.

Nscale is experiencing operational issues at certain data center sites, including project delays, equipment failures, and limited power availability, alongside investor-related pressures.

Business fixed investment rose at an 11% annualized rate in Q1 2026, largely attributed to AI-related spending, with a notable uptick in data center construction.

VB Pulse reports that 81% of companies with AI in production use half or less of their GPU capacity, while 71% still plan to invest in new AI infrastructure. The findings indicate broad underutilization alongside continued infrastructure expansion.

HUMAIN, led by CEO Tareq Amin, is scaling AI infrastructure in Saudi Arabia with an 84MW NVIDIA-based deployment and a 50MW offtake agreement with Nebius, located at Oxagon.

AMD is advancing agentic AI infrastructure with Helion, an open, modular rack-scale architecture designed to balance CPU-intensive AI agent execution with GPU-powered inference. Built on Open Compute Project (OCP) standards, Helion offers hyperscale configurations supporting up to 32,768 sixth-generation EPYC CPU cores and enterprise systems integrating EPYC processors with Instinct MI350P accelerators. The platform supports liquid and air cooling, flexible CPU configurations and intelligent token routing across AI models. With initial availability targeted for November 2026, Helion aims to improve computing efficiency, resource utilization and infrastructure scalability as AI agents drive growing demand for processing power, memory, networking and data center capacity.

AMD introduced the Instinct MI355X data center GPU featuring 288GB of HBM3E memory and 8 TB/s memory bandwidth, targeting AI infrastructure deployments.

Cortex DC, a subsidiary of The Chatterjee Group, announced a program to develop 2 GW of AI/HPC-ready data centre capacity across India. The first facility targets operational readiness within 18 months of groundbreaking, with a commitment to source 51% of energy from renewable sources to support high-performance AI and cloud workloads.

India’s data center market is seeing rising capital needs linked to AI workloads, with funding sourced through structured debt, sale-leaseback transactions, and direct hyperscaler participation.

Gartner estimates data-center owners need $8.5T in cumulative revenue by 2030 to cover AI infrastructure capex and a 12% ROIC hurdle, but only ~$3.6T is attributable to those assets, implying a ~$4.9T gap. AI capex is forecast at ~$7.5T for 2025–2030 (nearly $1T in 2026; +$6T during 2027–2030), with annual spend nearing $2T by 2030; up to ~$3.5T of assets could be impaired by 2030 and >$4T by 2032. No more than 30% of AI revenue is booked against underlying infrastructure, pushing hurdle-rate attainment to 2032 at the earliest.

The rapid expansion of hyperscale AI data centers is reshaping global digital infrastructure while intensifying debates over electricity consumption, water resources, environmental impacts and community development. As AI computing demand accelerates, governments and local authorities across the United States, Europe and Asia are introducing new policies, construction restrictions and grid connection requirements. From New York and Ireland to Denmark and Thailand, growing scrutiny of large-scale facilities highlights the importance of balancing AI investment with energy availability, sustainability and community interests. The developments underscore emerging opportunities for energy-efficient computing, renewable power integration, advanced cooling technologies and responsible AI infrastructure development.

At OCP Global Summit, NVIDIA contributed Vera Rubin rack and compute tray designs via its open MGX architecture (v1.2) supporting up to 250 kW racks with liquid cooling and standardized interfaces, and introduced a Vera CPU rack for AI agent workloads. NVIDIA joined AMD, Google, and Meta to publish Open Data Center Specification v0.7 and, with Google, Meta, and Microsoft, released the OCP Grid Resilience Specification targeting sub-2s recovery from grid events. Meta is adopting NVIDIA Spectrum-6 (102.4 Tbps) Ethernet switches and advancing Spectrum-X Multiplane (up to 8 planes) to scale beyond 100,000 GPUs. NVIDIA and Cisco unveiled the USFP optical standard targeting 400 Gbps per lane.

The company states its software platform connects silicon design with advanced packaging to streamline development of multi-die systems, supporting chiplets, high-bandwidth memory (HBM), and high-performance computing (HPC) devices targeted at data centers and AI factories.

Nutanix announced partnerships with NVIDIA, Supermicro, and Netris to offer AI infrastructure that integrates NVIDIA Spectrum-X Ethernet networking, aligning compute, networking, and software for enterprise AI deployments.

Crusoe obtained a $700M asset-backed financing facility from Neuberger Specialty Finance to purchase power generation equipment (behind-the-meter gas generators) for Crusoe-built and owned AI data centers. The facility is secured by recently ordered equipment and supports scaling of Crusoe’s AI factory platform, following a $3.9B Series F in Sep 2026 and over 6 GW of gross AI infrastructure capacity under contract. Customers cited include Microsoft, Google, Oracle, and OpenAI’s Stargate initiative.

World Wide Technology (WWT) and German AI infrastructure provider Polarise have announced a strategic partnership to accelerate sovereign AI data center development across Europe, beginning with a 5MW facility in Oslo, Norway. The new AI Factory will integrate Nvidia accelerated computing, liquid cooling and Polarise's full-stack AI cloud platform, with WWT serving as systems integrator. Building on Polarise's earlier AI cluster deployment in Munich, the collaboration aims to expand European-controlled AI computing capacity while supporting data sovereignty, enterprise AI inference and large-scale AI workloads. The partnership establishes a foundation for additional sovereign AI Factory deployments across Europe, strengthening regional AI infrastructure and digital independence.

TIM Enterprise will host a sovereign AI platform in Italy that integrates TIM’s centralized cloud and distributed edge network with NVIDIA’s full-stack AI, including GPUs, Nemotron models, and NeMo Guardrails. The platform provides managed services, data localization, access governance, security, and regulatory controls, and is supported by IPCEI CIS deployments of Edge Cloud nodes equipped with NVIDIA GPUs across Italy.

Crusoe announced a partnership with Araico to build a dedicated AI cloud facility adjacent to the Itaipú substation in Paraguay, powered by 100% renewable hydroelectric energy. Initial service is targeted for mid-2027 with 106 MW of IT capacity and long-term options to expand to approximately 280 MW. Crusoe will operate the AI compute for Crusoe Cloud, while Araico handles local development and site infrastructure, serving U.S. and Latin American customers.

NVIDIA introduced the DSX AI Factory Platform in Europe, featuring closed-loop liquid cooling and dry coolers to enable near-zero facility cooling water use and DSX MaxLPS software to deliver up to 30% lower power for the same AI token output. DSX Flex coordinates workload power with grid conditions, while SimReady Central launches on Hugging Face for validated power/cooling digital twin models. DSX OS is released as open source to automate infrastructure lifecycle and resilience. CoreWeave, Firmus, Google Cloud, IREN, Lambda, Nebius, Nscale and Sharon AI plan to use DSX MaxLPS. Industrial ecosystem partners include AVEVA, Dassault Systèmes, Siemens, ABB, Eaton, GE Vernova, Hitachi Energy, Schneider Electric, Siemens Energy, Trane Technologies and Vertiv.

Red Hat and Nvidia extended their collaboration to focus on security and deployment support for autonomous AI agents.

VivoPower executed a definitive heads of agreement with Open Innovation AI to jointly explore and develop sovereign AI data centers across multiple regions, targeting at least 1GW of capacity over three years. VivoPower will lead site development, power procurement, and facility operations; Open Innovation AI will supply software for compute orchestration, AI operations, and security. The agreement is a framework; each project will proceed via separate definitive project agreements. Initial deployment of Open Innovation AI software is planned across VivoPower’s portfolio and Nordic pipeline.

Lumen is integrating Eminence Grey’s sovereign AI platform with its physical infrastructure and programmable network to provide secure connectivity and orchestration for AI workloads spanning edge and cloud.

Cisco is expanding its Secure AI Factory with Nvidia through a strategic collaboration with Supermicro, introducing rack-scale AI infrastructure designed for enterprises, neocloud providers, and sovereign cloud environments. The architecture integrates Nvidia NVL72 systems, HGX and MGX GPU servers, Cisco high-performance networking, distributed security, and Splunk observability. Supporting both liquid- and air-cooled configurations, the solution enables organizations to scale AI training, inference, and agentic AI workloads with greater efficiency, security, and operational control. The expanded portfolio strengthens Cisco's position in full-stack AI infrastructure and accelerates deployment of next-generation AI data centers.

Huawei and Alibaba are advancing China's AI infrastructure capabilities through innovations in processors, computing clusters, networking, and foundation models. Huawei introduced its Peerium Computing Architecture, UnifiedBus interconnect, and new AI SuperClusters designed to support trillion-parameter models. Meanwhile, Alibaba unveiled its Zhenwu V900 AI processor, upgraded supernode servers, and next-generation Qwen AI model roadmap. Together, these developments highlight growing momentum in China's domestic AI semiconductor ecosystem and large-scale computing infrastructure.

CBRE Group has announced an agreement to acquire London-based Onnec Group in a transaction valued at up to $1 billion, strengthening its global data center infrastructure services portfolio. The deal includes an initial $610 million cash payment and up to $400 million in performance-based payments. Onnec brings specialized expertise in fiber networking, structured cabling, data center infrastructure deployment and maintenance across Europe and Asia Pacific. The acquisition will expand CBRE's ability to support hyperscale cloud providers and AI data center operators as demand grows for high-density connectivity and advanced infrastructure. CBRE expects its Critical Infrastructure Services business, including Onnec, to generate nearly $4.5 billion in revenue in 2027, reinforcing its expansion into AI-driven digital infrastructure services.

SK Hyper, a unit launched in July by SK Telecom, is promoting South Korea as a location for AI model training to draw international technology firms.

Nebius closed a $3.5 billion 364-day senior secured facility led by Citi to accelerate global AI infrastructure build-out, including data center development and GPU procurement, while preserving flexibility for longer-term and asset-backed financing.

Announced a new investment round for Peak, with additional investors participating. Proceeds are intended to scale the company’s AI capabilities and product development.

Intel installed the first major structural truss at its Ohio chip manufacturing site, signaling progress on a facility announced in 2022 with an initial production target of 2025. The company positions the project as part of expanding US leading-edge semiconductor capacity.

Singapore-based DayOne Data Centers is preparing for a Nasdaq IPO to accelerate its global AI and hyperscale data center expansion, supported by strong demand from ByteDance and other major technology companies. Formerly GDS International, DayOne has assembled 4.6GW of data center resources, including 2.3GW of reserved capacity across Southeast Asia, Japan and Europe. The company generated $484 million in revenue in 2025 and $512 million in the first half of 2026, driven by growing demand for AI computing and digital infrastructure. With ByteDance accounting for nearly 70% of revenue and committing to 838MW of capacity, DayOne is also expanding partnerships with Oracle and emerging AI cloud providers. The company is reportedly considering an IPO fundraising target of $2.5 billion to $3 billion to support new developments amid evolving AI infrastructure investment conditions.

Singapore-based SC Capital Partners has invested in a 40MW AI data center development in Pohang, South Korea, in partnership with NeoAI Cloud and government-backed investors. The high-density facility, scheduled for completion by the end of 2027, will feature liquid-cooled infrastructure designed to support next-generation Nvidia Vera Rubin GPUs and advanced AI workloads. With construction underway, regulatory approvals secured and an anchor tenant confirmed, the project will initially offer colocation services before expanding into GPU-as-a-Service (GPUaaS) and AI computing platforms. NeoAI Cloud is also planning a second-phase expansion of 260MW, potentially bringing total capacity to 300MW and strengthening South Korea's position in Asia-Pacific's growing AI infrastructure ecosystem.

NVIDIA signed a $1.3 billion agreement with Amkor to scale supply for semiconductor substrates, advanced packaging, test services, and optical engines. The deal is aimed at increasing manufacturing throughput and assuring capacity for NVIDIA’s AI accelerators and related optical interconnect components.

Digital Realty plans a three-building, 3M sq ft data center campus on 476 acres in Kokomo, Indiana, targeting ~1GW IT capacity by 2033. Investment is estimated between $1B and $3B. The site is expected to draw ~1.5GW from Duke Energy (grid agreements pending) and use closed-loop air-cooled cooling with water recirculation; Digital Realty will fund water infrastructure upgrades. Zoning for data centers has been approved; no tax abatement is being sought. Projected property tax revenue is $44M–$55M annually, with a noise cap of <50 dB at the property line.

Japan will fund $20B for two new semiconductor fabs at GlobalFoundries’ Malta, New York campus. GlobalFoundries is selected to build and operate the sites, as part of a broader agreement in which Japan plans $550B of investment tied to reduced US tariffs. The initiative targets expanded, trusted semiconductor manufacturing capacity in the United States.

MRO-TEK Networks is expanding its focus on India's growing data center market with plans to develop indigenous network operating system capabilities and integrated security platforms. Building on its Made-in-India switching portfolio spanning 1G to 100G, the company aims to strengthen domestic networking innovation and support rising demand for AI-ready data center infrastructure.

Asia-Pacific's rapidly expanding data center market is navigating new financial, regulatory, and energy challenges. Australia's Firmus withdrew its planned $5 billion IPO, while Philippine telecom operator PLDT postponed its Vitro data center REIT listing until 2027. Meanwhile, Indonesia and Thailand have paused major data center developments amid regulatory reviews, and Malaysia is addressing rising electricity demand. Despite these challenges, the region remains a major growth market for AI and digital infrastructure investment.

Financial Times reports SpaceX seeks approximately $40B in funding, with Apollo Global Management leading, to purchase Nvidia AI chips for expanding internal AI compute capacity.

Filed a US prospectus for a Nasdaq IPO to raise up to US$5B at ~US$20B valuation. Proceeds to fund new data center development/construction, working capital, and general corporate purposes. H1 revenue was US$512M with a net loss of US$81.9M; total bookings ~2.3 GW (1.4 GW in Johor, Malaysia). Largest customer contributed 69.2% of H1 revenue; 87% of H1 revenue from Malaysia.

AWS will invest more than $1 billion over five years to fund programs for communities hosting its US data centres, aiming to mitigate opposition and enable continued AI-focused capacity expansion. The move was outlined by AWS CEO Matt Garman, who argued data centre buildout is critical for US competitiveness in AI.

Bell Canada and the Saskatchewan government emphasized the rapid commissioning of an AI data centre outside Regina, adding regional AI compute capacity and localized processing in Canada.

Globe Telecom plans to invest a minimum of $1 billion to expand data center capacity in the Philippines to attract AI workloads and hyperscaler tenants as the country targets a regional AI hub position.

TSMC is evaluating a new multibillion-dollar manufacturing campus in Texas as part of its multi-year US expansion to boost output of AI-focused semiconductors, reflecting sustained demand from customers such as Nvidia and Apple.

SK hynix outlines how AI performance depends on full data center integration rather than individual chips, highlighting a design shift from server-level to rack-scale and facility-wide architectures. The piece emphasizes co-design of compute, memory (including HBM), storage, networking, and power/cooling, and calls for early ecosystem collaboration and shared standards to address bandwidth, data movement, and thermal/power constraints.

Jensen Huang stated that a three-month delay in AI data center infrastructure can forgo roughly $10B in revenue and indicated the potential to compress data center construction timelines.

Blackstone President Jonathan Gray outlined a plan to expand the firm’s India portfolio from about $50B to $100B over five years, prioritizing data centers and energy along with healthcare, housing, and financial services. He addressed AI infrastructure demand and said Indian IT services could benefit from AI, while commenting on private credit risks.

Editorial analysis arguing that AI deployments are moving from centralized training to distributed inference and agentic AI, requiring open standards, interoperable software, and heterogeneous compute across edge, cloud, and sovereign/on-prem environments.