AECP 7.0: Elevating One Platform Beyond VMware, Ready for Cloud and AI

Replacing VMware is a platform decision, not a hypervisor swap. AECP 7.0 widens what the platform covers and deepens what each layer can carry, from the storage you already own to agentic AI.

In our VMware Alternative series, Beyond the Hypervisor Swap, we argued that a VMware replacement should be judged on two axes.

• Breadth: does the platform cover every layer and every scenario your VMware estate covers today, from storage and networking to desktops, backup and AI?
• Depth: does each of those layers hold up under enterprise production load, with the availability, performance, security and operability Tier-1 workloads need?

AECP was built to answer both: an enterprise-ready, converged platform for VMs, containers and AI, managed from one console and recognized by Gartner, GigaOm and DCIG.

AECP 7.0 takes the whole platform a big step further on both axes:

Here are the highlights, section by section.

1. Keep the array: AECP Compute Edition

Most VMware alternatives ask you to replace the hypervisor and the storage at the same time. AECP Compute Edition separates the two decisions. It runs the AVE hypervisor on the enterprise SAN you already own, over Fibre Channel or iSCSI, so there is no storage purchase to begin.

At its core is VCFS, a clustered filesystem that formats a LUN and lets every host in the cluster mount it as a shared datastore. VM disks are files on that datastore, much as VMDKs are files on VMFS, so the operating model will feel familiar to vSphere teams.

What ships in 7.0:

• FC and iSCSI connectivity, storage multipathing, and storage adapter and device management.
• VCFS datastore management, thin and thick provisioning, and automatic space reclaim (UNMAP) back to the array.
• VM and virtual-volume I/O QoS.
• VM snapshots and snapshot schedules, crash-consistent and file-system-consistent.
• DRS adapted to external storage, and cross-cluster live migration of compute.
• Datastore capacity, performance and health monitoring with alerting.

Your array, your fabric and your storage team’s runbooks stay as they are. Compute and storage scale independently, and the same AOC console manages Compute Edition and converged clusters together.

2. More from the same hardware, with stronger security

Memory overcommit. AECP 7.0 adds per-VM memory overcommit to raise VM density on the hosts you already own. Each overcommitted VM gets its own isolated swap volume instead of contending for shared host swap, and policy is set at cluster, host and VM level. Critical databases keep their reserved memory while desktop, office and test workloads share.

Boost mode, per VM. On Compute Edition, the Boost mode I/O acceleration can now be switched on for individual VMs rather than whole clusters, so the CPU cost is spent only where the workload needs it.

VM Secure Boot and vTPM. Both are new in 7.0 and available on external storage as well as converged clusters. Secure Boot protects the guest boot chain; vTPM provides a virtual Trusted Platform Module for BitLocker, Credential Guard and Windows 11.

Host firewall. Cluster services now add their own entries to the host access whitelist automatically, rather than an administrator adding them by hand.

VMTools decoupled from the host. VMTools now upgrades independently of the platform, so guest tooling no longer waits for a platform upgrade window.

3. Simpler day-2 operations and migration

The services around the hypervisor move forward with AECP 7.0 too:

• AOC 4.10 manages Compute Edition (non-HCI) and HCI clusters from one console and can itself run on VCFS external storage.
• Japanese language: AOC 4.10 adds a Japanese user interface for key parts of the console, for customers and partners in Japan.
• Networking: custom system networks support multiple storage networks with static routes, alongside distributed virtual switches, VLAN and trunk. Traffic visualisation covers the custom system networks.
• Observability 1.5 adds capacity forecasting.
• Migration Tool 2.1, estimated arriving in early November, performs agentless warm migration from ESXi 5.0 to 8.0 and vCenter 6.0 to 8.0.

4. Networking: layer 7 load balancing

ANS has offered layer 4 load balancing until now. ANS 3.6 adds layer 7 load balancing to complement it, so application traffic can be balanced on the platform itself as well as connection-level traffic.

5. Keep your desktops: the new AECP VDI Standard edition

AECP VDI Standard brings two enterprise-grade building blocks together in one edition for Omnissa Horizon: the Arcfra AVE hypervisor and ABS (Arcfra Block Storage). AECP carries the Omnissa Ready — Horizon badge, and deployments have a joint Omnissa and Arcfra support path.

• Omnissa-certified hypervisor and storage. AVE and ABS are each certified by Omnissa for Horizon, so desktops run on an enterprise-grade hypervisor and enterprise-grade block storage that are validated component by component.
• Licensed by CPU core, with unlimited storage capacity. Growing the desktop estate does not trigger a storage licence true-up.
• Built for boot and login storms. Data locality keeps read I/O on the local host, so login peaks do not flood the network.
• Lower cost. Up to 67% lower infrastructure software cost than Broadcom VMware in HCI mode, in a 500-seat office-worker reference architecture.
• Zero-trust desktops. The distributed firewall applies default-deny east-west policy to every desktop and can quarantine a compromised one in seconds.

VDI Standard runs in HCI mode. Teams that want to keep their vSphere licences while replacing vSAN can start with VDI Essential and move to VDI Standard later.

6. Keep your backup software

Changing hypervisors doesn’t mean you have to change your data protection stack. Your backup product, repositories and retention policies stay where they are.

7. Agentic AI on the same platform: Enhanced Model as a Service via Neutree

AI agents do not call one model. They call many, from every team, all day. Without a shared model service, each team stands up its own GPUs, model copies and API keys, and cost and risk grow with every new agent.

Neutree is Arcfra’s Model-as-a-Service platform: one model registry, one inference gateway, and per-team quotas and access logs, running on-premises on the same AECP platform as your VMs.

Neutree 1.2.1, released earlier this month, added model load balancing and support for the vLLM 0.31.0 inference engine. Neutree 1.3, arriving on 16 November, adds more to enhance the platform for MaaS:

• Multi-tenancy with OIDC and Active Directory integration
• End-to-end observability
• KV-cache offloading
• High availability

What comes next

AECP 7.0U1 continues the Compute Edition roadmap with local disks alongside external storage, shared volumes for Oracle RAC, cross-cluster migration that includes storage, and single-node and two-node clusters.

Replace VMware. Run Private AI. One Platform.

Read more: Arcfra Releases AECP 7.0 with AECP Compute Edition, Extending VMware Migration to Existing SANs