Trillion-Dollar AI Capex, Power-Secured Cloud Capacity, and Control-Plane Recovery

TrendForce Puts Hyperscaler AI Spend on a $1 Trillion Track

TrendForce’s latest readout adds a new layer to the backlog story by quantifying how far the cloud buildout has already been pre-committed: AWS has increased purchases of NVIDIA GB300 and V200 rack-scale systems, Microsoft is still procuring NVIDIA rack-scale hardware, and Alphabet/Google, Meta, and Oracle are all expanding GPU deployments alongside data center buildouts. TrendForce now estimates the top eight cloud service providers will spend more than $710 billion on AI capex in 2026, with Alphabet/Google above $178.3 billion and Meta above $124.5 billion; aggregate hyperscaler capex forecasts have moved above $1 trillion.

The mix matters as much as the total. TrendForce says GPUs will account for nearly 60% of AWS’s AI server build-out in 2026, while GPU-based systems will exceed 80% of Meta’s build-out, signaling committed hardware programs rather than generic cloud inventory. Oracle’s GPU rack-scale expansion tied to Stargate/OpenAI shows the same model extending beyond the largest hyperscalers.

The scarce asset is now delivered rack-scale AI capacity: GPUs, HBM, interconnect, power, and ready sites. That extends the earlier shift from reservation and financing into execution, raising the value of supply lock-up, allocation discipline, and contracted utilization, while custom-chip efforts from AMD, Broadcom, and Marvell compete for share without removing the need to secure physical capacity early.

Microsoft’s Pullbacks Show Where Power-Secured Cloud Capacity Breaks First

Microsoft’s reported pullbacks this week made the constraint explicit: Bloomberg-linked coverage says the company paused or re-scoped cloud and AI data center plans in Ohio’s Licking County, Wisconsin, Illinois near Chicago, North Dakota, the UK/London corridor, and Jakarta after reassessing whether usable grid capacity and power delivery could actually be secured. In Ohio, Microsoft suspended initial rural land projects and repurposed two of three sites for agriculture; in Wisconsin, reporting described a later-stage suspension and a hold on expansion at the Mount Pleasant campus; near Chicago, it exited a space deal; in North Dakota, talks with Applied Digital slowed after an exclusivity clause expired. These were not demand cancellations. They show AI buildout being stopped where land, campuses, or partner discussions meet deliverable megawatts.

That pushes the story one step further from policy pressure into portfolio triage: cloud supply is becoming location- and energy-specific capacity. The Ola Electric–Axis Energy 20 GWh battery storage deal points in the same direction: industry sources say BESS can smooth volatile load, provide UPS-like support, enable peak shaving, and bridge shortfalls when utilities cannot meet demand. The build pattern is increasingly grid plus behind-the-meter generation plus storage, with gas advantaged where speed and scale matter. For operators, the unit of competition is secured uptime; for vendors and investors, value is moving toward BESS, microgrids, power infrastructure, and efficiency tools that stretch scarce megawatts.

FireMon and NetApp Push the Control Plane Into Recovery

FireMon’s completed integration with Palo Alto Networks’ Strata Cloud Manager is the latest sign that the control-plane story is moving beyond governance alone. FireMon now ingests, normalizes, and analyzes policy across environments, while Palo Alto retains deployment and enforcement. That split matters because it separates the system of record for governance from the systems that push controls into production, giving enterprises a cleaner way to maintain consistent oversight as Panorama and Strata Cloud Manager run in parallel.

The bigger shift is that control planes are now absorbing resilience. NetApp extended hybrid cloud data protection for Red Hat OpenShift with incremental-forever backups using Change Block Tracking and added a public preview of NetApp Disaster Recovery for OpenShift, moving Kubernetes protection from backup efficiency toward guided recovery workflows. Red Hat’s third straight year as a Leader for hybrid platforms reinforces buyer preference for standardized operating layers across datacenter, cloud, and edge.

For operators, this means fewer disconnected consoles for governance, backup, and recovery. For vendors and investors, the next layer of value is in integrated control-plane software that normalizes policy, orchestrates resilience, and sits above fragmented estates.