Cloud Computing Watch: AWS, Azure, Google Cloud Multicloud Interconnects, VMware Private AI Cloud & the End of Cloud-First Strategy
Multicloud Interconnects, Private AI Factories & the End of Cloud-First — Weekly analysis of hyperscaler competition, AI GPU infrastructure and the structural shift away from blanket cloud-first strategies.
Two Shifts That Define Cloud Right Now: Native Multicloud and Private AI as Primary
Enterprise cloud infrastructure spending jumped more than $43 billion YoY to reach $143 billion in Q2 2026 — the 11th consecutive quarter of rising growth per Synergy Research — yet the strategic conversation has moved from “how fast to public cloud?” to “where does each workload actually belong?” Two product moves define the answer: AWS and Microsoft putting AWS Interconnect – multicloud with Azure into Public Preview — a purpose-built, open-spec, managed L3 fabric with MACsec and a 100 Gbps path to GA — and Broadcom launching VMware Private AI Cloud on VCF 9.1 as the primary home for production AI and agents.
Hybrid remains the dominant architecture; value and governance metrics are rising relative to pure cost savings. The structural trend is cloud maturing from elastic public utility into a differentiated, multi-environment fabric optimized for AI — shaped as much by physics, power and policy as by software features.
01 — Cloud at a Glance
The biggest developments this week crystallize three interlocking shifts: hyperscalers are finally making multicloud networking native and managed; private-cloud platforms are being repositioned as the primary home for production AI and agents; and enterprises are formally retiring “cloud-first” as a default in favor of deliberate, workload-by-workload placement driven by cost, performance, regulation and data gravity. Hybrid remains the practical majority.
02 — Hyperscaler Watch
AWS, Azure, GCP — Growth and the Interconnect Milestone
AWS, Microsoft Azure and Google Cloud remain the clear leaders, with recent quarterly growth rates of approximately 37%, 43% and 82% respectively — illustrating both AWS scale advantage and the relative velocity of the smaller two. Combined Big Three share stays in the high-60s percent range even as absolute spend explodes.
Standout move: Public preview of AWS Interconnect – multicloud with Microsoft Azure and reciprocal Azure Multicloud Interconnect. Provision private, resilient, MACsec-encrypted Layer-3 connectivity through a single managed experience, initially in select regions (US East N. Virginia, US West, Sydney, Frankfurt) at preview bandwidths, with paths to 100 Gbps at GA. AWS already offers equivalent managed interconnects to Google Cloud (GA) and Oracle Cloud Infrastructure (Preview).
This is the first purpose-built, open-specification multicloud networking product of its kind and materially lowers operational friction. Google Cloud was again named a Leader in the 2026 Gartner Magic Quadrant for Strategic Cloud Platform Services, positioned furthest for Completeness of Vision, emphasizing its co-designed silicon-to-agent stack.
03 — AI Cloud Watch
AI remains the primary growth engine. AWS and NVIDIA expanded their partnership with plans to deploy an additional 2 million NVIDIA GPUs across AWS infrastructure in 2027–2028, on top of earlier multi-million-GPU commitments, plus deeper collaboration on CPUs, networking, and secure AI factories for government workloads.
Capacity constraints (especially power) continue to dominate. Pre-leasing of new data-center capacity remains extremely high, nuclear and other firm-power deals proliferating, and “neoclouds” plus specialized AI infra providers capturing meaningful share of GPU-hungry training and inference demand.
04 — Cloud Infrastructure
VMware Private AI Cloud — Bring the Model to the Data
Broadcom used VMware Explore 2026 (late August) to launch VMware Private AI Cloud and VMware AI Factory. Built on VMware Cloud Foundation (VCF) 9/9.1, the stack runs traditional workloads, inference, and agentic applications on a single private platform. Features include heterogeneous accelerator support, validated open and commercial models (150+), memory tiering and storage optimizations claimed to cut hardware costs significantly, AgentMinder for agent governance, and tighter integration with data and security services.
Networking enhancements in VCF 9.1 (transit gateways, segmentation, native EVPN VXLAN) and partnerships (e.g., expanded Kyndryl collaboration) reinforce the private-cloud push for AI. VCF 9.1 delivers up to 40% reduction in server costs through memory tiering and up to 39% lower storage TCO.
05 — Cloud-Native Watch
Platform engineering and internal developer platforms continue toward ubiquity. Kubernetes remains default orchestration, with hyperscalers and private-cloud vendors investing in simpler upgrade/rollback paths, AI-assisted operations, and tighter integration of agents into the control plane. Serverless and container services are increasingly tuned for AI inference and agent runtimes rather than pure general-purpose workloads.
06 — Cloud Security
Agentic AI is forcing new security and governance models. Broadcom’s AgentMinder and related Tanzu capabilities aim to bind agent authority to specific missions with least-privilege and auditability. Sovereign and regulated environments drive demand for private or hybrid AI stacks that keep data and models under tighter control. Multicloud interconnects with built-in encryption reduce some surface area but expand the need for consistent identity, policy,y and observability across providers.
07 — FinOps & Cloud Economics
The economic narrative has shifted. Organizations still chase efficiency, but “value delivered to the business” is rising as a primary metric while pure cost-cutting declines in relative importance. AI token and GPU costs remain highly variable and difficult to forecast; private AI platforms are positioned partly as a way to regain predictability and utilization control. Marketplace tools and commitment trackers show customer cloud commitments now measured in the low trillions of dollars, locking in multi-year spend while capacity remains the binding constraint.
08 — Enterprise Cloud
“Cloud-First Is Out”
A Kearney CIO survey and multiple analyst notes confirm large enterprises are replacing blanket public-cloud defaults with workload-by-workload placement that weighs cost, latency, data gravity, regulatory risk,sk and AI readiness. Hybrid is the practical majority architecture. Migration tooling is incorporating more AI-driven assessment (e.g., Google’s Migration Center enhancements), but strategic emphasis is modernization and selective placement rather than wholesale lift-and-shift.
09 — Edge & Distributed Cloud
Sovereign and regional cloud expansion continues (Oracle in the Middle East, various national AI factory programs). Edge inference for agents and physical AI is rising in priority, but the bulk of new capital is still flowing into large centralized AI factories and hyperscale campuses constrained by power and interconnect.
10 — M&A & Competitive Moves
IBM completed its ~$11 billion acquisition of Confluent earlier in the year; integration is now focused on feeding real-time, governed data streams into watsonx and hybrid environments to power generative and agentic AI. The deal underscores that data-in-motion platforms are strategic infrastructure for AI, not just application plumbing. Broader M&A remains selective: capital flowing heavily into data-center, power, and specialized AI infrastructure assets rather than pure software roll-ups.
11 — The Cloud Shift
The structural trend is the maturation of cloud from primarily elastic, public, general-purpose utility into a differentiated, multi-environment fabric optimized for AI. Multicloud is becoming operationally feasible rather than merely strategic; private cloud is being re-architected as preferred home for sensitive, high-utilization or regulated AI workloads; and economics plus sovereignty are forcing explicit placement decisions. Data gravity and agentic workloads are the forcing functions.
Three Cloud Signals
- Multicloud plumbing is now a product. AWS Interconnect – multicloud with Azure, GCP, OCI on one open spec removes the carrier middleman and makes private 100 Gbps paths a managed primitive.
- Private AI is now positioned as primary, not fallback. VCF 9.1 / Private AI Cloud's 40% server-cost reduction claim and AgentMinder target FinOps and sovereignty gaps public GPU can't solve alone.
- Placement replaces migration. With $143B quarterly spend and power-constrained factories, the CIO question is explicit workload economics across public, private, and edge — not lift-and-shift.
12 — Strategic Takeaway
For CIOs: Treat placement as a continuous discipline, not a one-time migration. Invest in observability, FinOps, and policy engines that work across public, private,e and edge. Evaluate private AI platforms seriously for production inference and agents where cost predictability or data control matter.
For developers and platform teams: Expect platform engineering to absorb more of the AI runtime and agent governance surface. Multicloud networking is improving, but consistency of identity, networking, and data services remains the hard problem.
For investors: The highest-conviction themes remain AI infrastructure (GPUs, power, data centers), specialized cloud and networking software that reduces multicloud friction, and platforms that improve utilization and governance of AI capacity. Pure hyperscaler share battles matter less than absolute AI-driven spend growth.
13 — What to Watch Next Week
- Further details and early customer traction on AWS–Azure Interconnect and any additional provider announcements
- Follow-on product and partnership news from the VMware Private AI Cloud ecosystem
- Any incremental CapEx or capacity guidance from hyperscalers or major neoclouds
- Regulatory or sovereignty developments that further shape placement decisions (especially in Europe, Middle East, Asia)
- Earnings or guidance updates that quantify AI contribution to cloud growth rates
Cloud is no longer primarily a destination. It is becoming a managed, multi-environment operating system for the AI era — shaped as much by physics, power and policy as by software features.
Editorial Note: Cloud Computing Watch is The CODEW's recurring intelligence series tracking cloud infrastructure, hyperscaler economics, and the specialist providers reshaping how AI compute gets bought and sold. This edition prioritizes architecture and economics over pure product launches.
Reviewed by Erwin Castro
on
Tuesday, September 08, 2026
Rating:
