AI Infrastructure Scales Out: 2GW AI Factories, Sovereign Data and 1.6T Optical Links

network-switch.comReviewed by David Lorame

By: Network-Switch.com | September 15, 2026

Table of contents
Engineers monitoring gigawatt-scale AI data center infrastructure with high-density GPU racks, optical networking, enterprise data platforms and governed local AI systems.

Quick Summary

Enterprise infrastructure developments from September 7 through September 14 show AI scaling in two directions at once: outward into gigawatt-class data center ecosystems and inward into more controlled enterprise deployments built around governed data, custom accelerators and local infrastructure.

Seven new developments were selected for this edition. Previously covered September 1-7 topics were excluded, and older Juniper or HPE Aruba announcements were not reused simply to fill the list.

NVIDIA and Australian Partners Plan Up to 2GW of New AI Infrastructure

NVIDIA announced on September 9 that it is working with Firmus, Sharon AI, IREN, Megaport, ResetData, CDC, NEXTDC and AirTrunk to expand AI infrastructure capacity across Australia, with the ecosystem targeting up to 2 gigawatts of new capacity by 2027.

The buildout is based on NVIDIA DSX AI factory architecture. Sharon AI plans deployments of up to 68,000 NVIDIA GPUs connected with Quantum InfiniBand and Spectrum-X Ethernet, while IREN is applying the DSX blueprint across large power and data center sites. The infrastructure story is clear: AI expansion is becoming a coordinated land, power, network and compute project rather than a server procurement exercise.

NVIDIA and Palantir Tie Sovereign AI to the Physical Supply Chain

NVIDIA and Palantir announced a sovereign AI stack for critical supply chains on September 10, combining NVIDIA Nemotron open models with Palantir Foundry, AIP and Ontology. The first deployment is inside NVIDIA's own supply chain.

NVIDIA says a Vera Rubin rack involves about 1.3 million parts, making supply-chain coordination a direct infrastructure constraint. The joint architecture can run on-premises, in colocation facilities or in cloud environments while organizations retain control of proprietary operational data. For infrastructure operators, this is another sign that sovereign AI is moving from policy language into production architecture.

Dell and Aible Put Tighter Controls Around Enterprise Agentic AI

Dell published a new Aible on Dell AI Factory deployment blueprint on September 10. The architecture uses Aible's Minimum Viable Agency approach to limit the permissions, data access and reasoning scope assigned to individual AI agents.

Dell also describes physically air-gapped deployment on Dell Pro Max with GB10, with agent isolation, permission controls and NVIDIA NeMo Guardrails used to keep workflows bounded and observable. The important infrastructure shift is from simply running agents to operating them with predictable cost, governance and deployment controls.

Dell PowerScale for Azure Targets the Persistent Data Layer Behind Enterprise AI

Dell detailed PowerScale for Microsoft Azure on September 9 as a native cloud file-storage layer for enterprise AI. The service supports NFS, SMB and S3 access with unified permissions and can scale to 8.4PB of usable capacity in a single namespace.

Dell positions PowerScale as a persistent corpus layer for Azure AI Foundry and retrieval-augmented generation pipelines, including delta updates when source documents change. This reflects a broader trend in enterprise AI: storage architecture and data traceability increasingly determine whether AI systems can move from experiments into governed production.

HPE ProLiant and NVIDIA Blackwell Power Distributed VFX Workflows at DNEG

HPE published a new DNEG infrastructure case study on September 8 covering globally distributed visual-effects and animation workloads. The environment uses HPE ProLiant DL385 Gen11 servers with AMD EPYC 9475F processors, HPE DDR5 Smart Memory and NVIDIA RTX PRO 6000 Blackwell Server Edition GPUs.

The platform is designed for dense graphics and compute workloads and supports both advanced air cooling and direct liquid cooling. Although this is a customer deployment rather than a new server launch, it is a useful example of how enterprise GPU infrastructure is spreading beyond AI model training into high-resolution simulation, rendering and remote production.

Qualcomm and Amazon announced a long-term AI data center collaboration on September 8 covering custom AI inference silicon and high-speed optical connectivity. Reuters reported that Amazon could purchase up to $60 billion of Qualcomm AI data center chips and related products under the agreement.

The companies also plan optical connectivity solutions extending to 1.6 terabits per second. That combination matters because custom accelerators cannot scale independently of the network: inference density is pushing compute silicon, SerDes, optical DSPs and data center fabrics into the same design cycle.

Arista Brings Mission-Critical IP Media Networking to IBC 2026

Arista participated in IBC 2026 in Amsterdam from September 11 through September 14, with its media and entertainment engineering team meeting broadcasters and infrastructure operators at Hall 10.D40.

The event focus includes IP network modernization, high availability, multicast, precision timing, orchestration and monitoring for live and post-production environments. This is not a new switch launch, but it highlights a market where deterministic network behavior and operational visibility matter as much as raw bandwidth when legacy broadcast systems migrate onto Ethernet.

Frequently asked questions (FAQs)

What were the most important AI infrastructure developments from September 7 to September 14, 2026?

Key developments included NVIDIA-backed plans for up to 2GW of AI infrastructure in Australia, a sovereign supply-chain AI stack from NVIDIA and Palantir, new Dell enterprise AI deployment and data-layer architectures, an HPE GPU server deployment at DNEG, Qualcomm and Amazon's custom AI silicon and 1.6T optical connectivity agreement, and Arista's IBC 2026 media networking activity.

How large is NVIDIA's planned AI infrastructure expansion in Australia?

NVIDIA says its Australian cloud and infrastructure partners are working toward up to 2 gigawatts of AI infrastructure capacity by 2027, using NVIDIA DSX-based AI factory designs.

What is Dell PowerScale for Azure used for in enterprise AI?

Dell positions PowerScale for Azure as a persistent enterprise data layer for AI pipelines, including retrieval-augmented generation. It supports NFS, SMB and S3 and can scale to 8.4PB usable capacity in a single namespace.

What does the Qualcomm and Amazon AI infrastructure deal include?

The agreement covers custom AI inference silicon and high-speed optical connectivity, including solutions extending to 1.6 terabits per second. Reuters reported that Amazon could purchase up to $60 billion in Qualcomm AI data center chips and related products over the life of the partnership.

Were there new standalone Juniper or HPE Aruba hardware launches during this review window?

No qualifying standalone Juniper or HPE Aruba switch, router or wireless hardware launch was confirmed for this edition. Rather than reuse older announcements, this roundup focuses on newly verifiable infrastructure, deployment and technology developments.

Sources

  1. NVIDIA: Australia AI Infrastructure Capacity Expansion
  2. NVIDIA: Sovereign Intelligence With Palantir
  3. Dell: Aible on Dell AI Factory
  4. Dell: PowerScale for Microsoft Azure and Enterprise AI
  5. HPE: DNEG Scales VFX Workflows With HPE Servers
  6. Reuters: Qualcomm and Amazon AI Chip and Optical Networking Deal
  7. Arista: IBC 2026