Digital Infrastructure Monthly Insights

Coverage points to a transition from traditional model-centric AI to agentic AI workflows, with interest in broadening compute options beyond Nvidia accelerators to support these workloads.

Equinix launched the first phase of an AI-ready data center in Hong Kong, providing 1,000 cabinets with an initial investment of US$124 million.

Alibaba Cloud opened a fully operational cloud region in France with two availability zones in Paris, adding to existing European infrastructure in Germany and the UK. The region offers enterprise cloud services across compute, storage, containerization, networking, security, databases, and developer tools, and supports a forthcoming rollout of enterprise-grade agentic AI services in European markets later this year.

A 360MW data center in Batam, Indonesia is under construction to accommodate up to 170,000 Nvidia accelerator chips, with operations expected by Q1 2027.

Taiwan plans a $200 million investment to build a data center in Paraguay, targeting 10MW computing capacity with operations expected by end-2027. The facility will be constructed on government land and is reported as a diplomatic initiative, with backing cited from the Trump administration.

Google has reportedly selected Intel to fabricate more than 3 million custom Tensor Processing Units (TPUs) for delivery in 2028 after months of process validation, shifting part of its TPU manufacturing from TSMC. The TPUs support AI training and inference for Google Cloud.

A telecom-related project, part of broader business deals in the country, targets sovereign, physical and agentic AI services with a national deployment scope.

Nvidia approved Samsung, SK Hynix, and Micron as HBM4 suppliers for its Vera Rubin platform, with production in progress and first shipments expected in Q3. The move diversifies supply for scarce high-bandwidth memory, addresses AI data center bottlenecks, and coincides with a new R&D center in South Korea.

Megaport announced plans for a globally distributed AI inference cloud built around an on-demand GPU pool funded by A$350 million, offered to enterprise customers via contracted and consumption-based pricing models.

IBM disclosed a sub-1nm (0.7nm) silicon design employing a 3D stacked architecture that fits nearly 100 billion transistors in a fingernail-sized area and claims roughly double the density of its 2nm demonstration from 2021. The work targets continued gains in performance and efficiency as physical scaling nears atomic dimensions.

Telefónica finished rolling out 17 distributed edge computing nodes across Spain, creating a low-latency compute layer integrated with its network to support digital services and data residency requirements.

HPE introduced AI data center networking and routing updates, Agentic AIOps enhancements, and a unified AI-native SASE platform. The release adds support for HPE Networking CX wired access switches in the HPE Mist platform, expands Marvis insights and self-healing in Aruba Central, and applies agentic reasoning for faster root-cause analysis. HPE launched HPE Juniper Networking QFX switches optimized for inference and scale-up architectures, integrated into the HPE AI Data Center Solution and managed via HPE Networking Data Center Director.

Nvidia has told Chinese customers its Vera CPU for AI data centers could be available in August, positioning it as an alternative after H200 GPU shipments were stalled by U.S. export restrictions, per Reuters.

Vertiv announced SmartRun, a modular prefabricated overhead infrastructure solution intended to speed the build-out of high-density AI data centers.

CoreWeave announced it has brought a fully validated and operational Vera Rubin NVL72 system online in its AI cloud.

Apple implemented approximately 20% price increases on MacBook and iPad lines, attributing the move to rapid rises in memory component costs linked to AI infrastructure demand. iPhone pricing remains unchanged.

Following probes into Amazon Web Services and Microsoft Azure, the European Commission issued a preliminary decision to classify their parent companies as Digital Markets Act gatekeepers for cloud computing, signaling forthcoming interoperability, data portability, and anti-self-preferencing obligations pending final designation.

OpenAI introduced 'Jalapeno', a custom inference processor co-developed with Broadcom. The chip is tailored to OpenAI’s inference systems and is the first release in a jointly developed multi-generation compute platform.

Nvidia announced agreements in South Korea: a multi-year technology partnership with SK Hynix to advance next‑generation memory for AI data centres; cooperation with Naver and Doosan to build AI data centres; and SK Telecom plans to deploy a gigawatt‑scale AI cloud using Nvidia technology with the first data centre targeted for 2027. Doosan expects its energy solution to integrate with Nvidia data centre platforms and contributes materials used in Nvidia’s Blackwell chips. Deal values were not disclosed.

PLDT filed with the Philippine SEC to rebrand its data center subsidiary as Vitro REIT and plans to raise about $395 million via a REIT initial public offering.

Verizon Business, via comments from SVP and Chief Product Officer Scott Lawrence, indicated it is conducting a GPU-as-a-Service proof of concept with a partner.

SK hynix filed with the US SEC to list American Depositary Receipts on Nasdaq, planning to issue 17.79 million new shares to raise about KRW45.5 trillion (~$29.4B), implying a company valuation near $1.2T.

The fragment references artificial intelligence applications tied to growth in a national digital economy and includes an unrelated listing for a local sour gas plant operator in Alberta. No identifiable telecom operator, vendor, deployment details, or network specifications are provided.

Verizon Business disclosed a partnership to test a GPU-as-a-Service proof of concept aimed at supporting AI workloads. The excerpt does not provide partner identity, deployment model, or technical specifications.

Meta Platforms expanded its partnership with Reliance Industries. Reliance will build a 168MW AI-enabled data centre in Gujarat, which Meta will lease to scale its AI infrastructure in India.

Supermicro plans to raise approximately $7.0 billion via equity and equity-linked securities to finance working capital and inventory for AI system orders and related supply chain needs.

LG Uplus designated its forthcoming 200 MW Paju AI data center as the centerpiece of its AI data center (AIDC) infrastructure strategy, prioritizing inference-driven AI workloads.

Plans indicate a move from basic space leasing to an 'AI Factory Operator' approach during the operational phase, integrating strategic and operational services for AI workloads.

Generative AI vendors are increasing investment to build presence in India, leveraging a user base exceeding one billion internet users.