Broadcom’s VMware acquisition ended perpetual licenses and collapsed the catalog into per-core VCF/VVF bundles with a 16-core-per-socket minimum — that, not performance, is the migration driver. Free ESXi is back as 8.0U3e but caps at 2 physical CPUs per host and 8 vCPUs per VM. Proxmox VE 9.2 has no caps; use it for new builds.
Proxmox vs ESXi in 2026 comes down to one question that has almost nothing to do with raw speed: after the Broadcom licensing shake-up, which hypervisor is actually worth building on? Broadcom removed the free ESXi tier in early 2024, the homelab world rebuilt on Proxmox, and then — quietly — Broadcom brought free ESXi back.
Free ESXi is available again, specifically as ESXi 8.0 Update 3e (Build 24677879), per Broadcom's own KB. So the natural question returns: should you go back? The short answer for most people is no — but the long answer involves the real feature caps, the current Proxmox release, actual migration experience, and a few scenarios where ESXi still wins. Here's the breakdown.
TL;DR · The 2026 hypervisor call
- Fresh homelab or small-server build, 2026: Proxmox VE 9.2. Free, no caps, clusters out of the box, KVM + LXC, and the post-Broadcom community owns the playbook.
- On a working paid vSphere license until renewal: Stay put, then re-evaluate against the per-core VCF/VVF math below.
- Paying Broadcom renewal sticker shock: Migrate to Proxmox. The built-in VMware import wizard makes it a few-step job — just read the migration gotchas first.
- VMware certification practice / ESXi muscle memory: Free ESXi 8.0U3e is fine — just know the 2-CPU-per-host and 8-vCPU-per-VM caps and that there is no vCenter.
- Don't: Try to cluster free ESXi hosts. The free tier has no vCenter, no vMotion, no HA.
Free ESXi 8.0U3e: unlimited VMs, max 2 physical CPUs/host, 8 vCPU/VM cap, no vCenter/vMotion/HA/API · Proxmox VE 9.2: no caps, free clustering, KVM + LXC, full API · Paid VMware: per-core with a 16-core-per-socket minimum.
The 2026 Reality: Licensing Changed Everything
The reason "Proxmox vs ESXi" reads completely differently in 2026 than it did two years ago is licensing, not technology. After Broadcom acquired VMware, perpetual licenses ended and the a-la-carte catalog collapsed into two bundles: VMware vSphere Foundation (VVF) and the fuller VMware Cloud Foundation (VCF).
Both are priced per physical core with a mandatory 16-core-per-CPU minimum. As Broadcom's "Counting Cores" KB states, "you must license a minimum of 16 physical cores for each CPU in your ESXi hosts, even if a CPU has fewer than 16 cores." A dual-socket host with 8-core CPUs is billed as 2 × 16 = 32 cores. The two bundles differ mainly in bundled vSAN storage — VCF grants 1 TiB of vSAN per core, VVF grants 0.25 TiB per core.
What that does to renewals is the whole story. Practitioners on Hacker News self-report the damage in hard numbers: one Broadcom customer's renewal jumped from 8 million dollars to 100 million, and another commenter reported going from 66 dollars last year to 3,600 this year. Those specific figures are self-reported anecdotes, not audited. At enterprise scale the pattern is documented on the record: AT&T's court filing claimed Broadcom offered it a 1,050 percent increase. The direction is unambiguous, and it is the single biggest reason shops are evaluating Proxmox at all.
Proxmox flips the model. The software is free and open source; the optional subscription is per CPU socket (not per core) and buys the stable enterprise repository plus support, not a license to run. Per the official pricing page, tiers are Community EUR120, Basic EUR370, Standard EUR550, and Premium EUR1,100 per socket per year. Every feature works without any subscription.
What Free ESXi Actually Gives You (and What It Doesn't)
Free ESXi 8.0 Update 3e is a real type-1 hypervisor. Unlimited VMs, no time limit, no license fee. For a single-host lab that just needs to run VMs, that is a legitimate offering.
The caps are where it bites. Per Broadcom's KB, the free build is limited to a maximum of 2 physical CPUs per host and a maximum of 8 vCPUs per virtual machine. It cannot be managed by vCenter Server. There is no vMotion, no DRS, no HA, and no VADP-based backups. API-based host management is not supported — which kills Terraform, Ansible, and most automation. There is no Broadcom support. Broadcom positions it explicitly for home labs and dev/test, not production.
The 8-vCPU-per-VM ceiling is the one that bites hardest in practice. A domain-controller test VM is fine on 8 vCPUs. A Plex transcoding VM or a build server that wants 16+ threads hits a hard wall the free tier will not bend on. And because there is no free ESXi 9 tier, the free path dead-ends at 8.0U3e.
Proxmox VE, by comparison, has no VM limits, no vCPU caps, and no feature lockouts on the free tier. The subscription buys the enterprise repo and support — nothing functional is gated behind it.
What You Get With Current Proxmox VE 9.2
The current release is Proxmox VE 9.2, dated May 21, 2026, per the official roadmap. It is built on Debian 13.5 "Trixie" with Linux kernel 7.0 as the new stable default, QEMU 11.0, LXC 7.0, and ZFS 2.4, and it ships Ceph Squid 19.2.3 (with Ceph Tentacle 20.2.1 available as a stable option).
The headline 9.2 features close some of the historical gaps against vSphere: a Dynamic Load Balancer via the Cluster Resource Scheduler that auto-migrates HA-managed guests to reduce node imbalance; software-defined-networking additions including WireGuard and BGP fabric protocols, route maps, and prefix lists; management of custom CPU models from the web UI; and the ability to disarm and re-arm the HA Manager cluster-wide for safer planned maintenance.
None of that is behind a paywall. Clustering, live migration, HA, the full REST API, and LXC containers alongside KVM VMs are all in the free edition — the exact set of capabilities the free ESXi tier withholds. (New to running both? Our LXC vs VM in Proxmox guide has the decision tree for picking one per workload.)
| Free ESXi 8.0U3e | Proxmox VE 9.2 (free) | Paid VMware (VVF/VCF) | Proxmox (subscribed) | |
|---|---|---|---|---|
| Price | $0 | $0 | Per-core, 16-core/socket min | EUR120-1,100/yr/socket |
| Physical CPUs/host | Max 2 | No cap | No cap | No cap |
| vCPU cap per VM | 8 vCPUs | No cap | No cap | No cap |
| LXC containers | No | Yes | No | Yes |
| Centralized mgmt | No vCenter | Built-in cluster UI | vCenter included | Built-in cluster UI |
| Live migration | No | Yes (free) | vMotion | Yes |
| API / automation | Not supported | Full REST API | Full API | Full REST API |
| HA / clustering | No | Yes (3+ nodes) | vSphere HA | Yes (3+ nodes) |
| Vendor support | None | Community only | Broadcom | Proxmox GmbH |
Does Proxmox Actually Run Faster Than ESXi?
For most homelab workloads, no meaningful difference — and that is the honest answer. If you are running Jellyfin, Home Assistant, Pi-hole, Nextcloud, and a stack of Docker containers on a mini PC, the hypervisor overhead is invisible next to your disk, RAM, and the applications themselves.
Where a real gap shows up is storage throughput on the Linux stack. The most-cited data point is from storage vendor Blockbridge, whose published NVMe/TCP testing had Proxmox VE beat VMware ESXi in 56 of 57 individual tests, with roughly 50 percent IOPS gains. Read that with the obvious caveat: it is vendor-authored testing with a Proxmox integration, so the scenarios favor Linux's storage path. Take it as a genuine signal about Linux NVMe/TCP performance, not a universal hypervisor verdict. If you are building a storage-focused node serving fast NFS over a 2.5GbE or 10GbE network, that advantage is real and measurable.
One structural note that matters more than benchmark percentages: Proxmox ships with conservative defaults and expects you to tune the CPU governor, I/O scheduler, and NUMA settings, whereas ESXi arrives tuned. That is solvable work, covered in our Proxmox setup guide — not an architectural ceiling.
Migrating From ESXi to Proxmox: What Practitioners Hit
Migration in 2026 is far easier than it was, thanks to the official Proxmox VMware import wizard — the Hacker News discussion of the wizard frames it as roughly a three-step job (add ESXi as a source, fill the wizard, start the VM). But real production migrators keep hitting the same three traps. Plan around them and the move is smooth; ignore them and you will lose a weekend.
Prep VirtIO before you migrate Windows VMs. The single most-repeated gotcha: move a Windows VM without preparing VirtIO and it blue-screens with INACCESSIBLE_BOOT_DEVICE because the guest has no driver for the new virtio-scsi disk controller. The safe workflow is to uninstall VMware Tools and install the VirtIO drivers first, import with a SATA/IDE controller so the VM boots the first time, then attach a small dummy virtio-scsi disk to load the driver before converting the boot disk to virtio-scsi.
Shared SAN storage historically loses efficient snapshots. On FibreChannel or iSCSI shared block storage, Proxmox traditionally supported only thick LVM, not lvm-thin — so a snapshot was not thin and needed as much free space as the original VM (a 5TB VM needed 5TB free just to snapshot). This is the biggest architectural adjustment for shops coming from VMFS. Proxmox VE 9 added a snapshot-as-volume-chain option — external QCOW2 overlays on thick LVM over FC/iSCSI SAN — but it is still a technology preview in the 9.x line and not recommended for production, so design around the thick-LVM limit before you commit.
Clustering needs three votes, not two. A 2-node ESXi mental model backfires on Proxmox HA, which needs three voting members (or a QDevice tiebreaker) to keep quorum. Give corosync its own low-latency network too, or a node that thinks it is isolated will self-fence and reboot, taking its VMs down.
Despite all that, the dominant sentiment from people who actually migrated is relief: more capable, easier to manage, and Proxmox Backup Server makes backups straightforward. Low regret is the recurring theme.
Who Should Use What
Use Proxmox VE 9.2 if:
- You're building a new homelab or small server from scratch in 2026 — it's the default for a reason
- You want both KVM VMs and lightweight LXC containers from one interface
- You plan to add a second and third node and want free clustering, live migration, and HA
- You want API access for Terraform, Ansible, or scripted provisioning without paying a license fee
- You're facing a Broadcom renewal and the per-core, 16-core-per-socket math no longer makes sense
Use free ESXi 8.0U3e if:
- You're actively studying for a VMware certification and need hands-on vSphere 8 experience
- You have a working paid vSphere environment and aren't up against a renewal yet
- You're running enterprise hardware with VMware-certified drivers that would be painful on Linux
- You'll never need more than 2 physical CPUs, 8 vCPUs per VM, an API, or a second clustered node
The Verdict
Free ESXi coming back is good for the ecosystem — competition keeps Proxmox honest about its own rough edges, and Proxmox has them: the conservative performance defaults, the thick-LVM snapshot limit on shared SANs (only partly addressed by the VE 9 external-QCOW2 snapshot preview), the ZFS ARC memory surprise on fresh installs. Those are real.
But the feature gap between free ESXi and free Proxmox is enormous. Proxmox VE 9.2 gives you containers, clustering, live migration, a full API, and no CPU or vCPU caps for $0. Free ESXi 8.0U3e gives you a single-host VM platform capped at 2 physical CPUs and 8 vCPUs per VM, with no vCenter, no automation, no HA, and no free version 9 ahead of it. On the paid side, Broadcom's per-core VCF/VVF pricing with its 16-core-per-socket floor is exactly what is pushing shops to evaluate the switch in the first place.
For a new build in 2026, Proxmox is the answer — not because it wins every benchmark, but because it is more capable, more flexible, and the KVM, LXC, ZFS, and Linux-networking skills you build transfer straight to real infrastructure work. If you're coming from ESXi, read the migration gotchas above, prep your VirtIO drivers, and the move is a well-trodden path.
Install Proxmox. Tune the defaults, size the host with our Proxmox hardware guide, and build something real.
Speccing a fresh Proxmox host? The PC Builder cross-checks socket, RAM, and PSU compatibility from a CC BY 4.0 parts dataset.
Sources and methodology
This comparison is grounded in official documentation from Proxmox and Broadcom/VMware, plus cited community practitioner threads. No first-party benchmarks were run.
Primary sources cited:
- Broadcom KB — ESXi 8.0 Update 3e free hypervisor — free-tier availability and caps (2 CPUs/host, 8 vCPUs/VM, no vCenter/vMotion/HA/API)
- Broadcom KB — Counting Cores for VCF/VVF — per-core licensing with 16-core-per-socket minimum; vSAN entitlements
- Proxmox VE Roadmap — VE 9.2 (2026-05-21), Debian 13.5, kernel 7.0, QEMU 11.0, LXC 7.0, ZFS 2.4, feature highlights
- Proxmox VE pricing — subscription tiers per socket/year (EUR120-1,100)
- Blockbridge — Proxmox VE vs VMware ESXi NVMe/TCP benchmarks — vendor-authored storage testing (56 of 57 tests, ~50% IOPS gains)
- Blockbridge — QCOW2 snapshots on thick LVM — VE 9 snapshot-as-volume-chain on FC/iSCSI SAN (technology preview)
- The Register — AT&T claims 1,050% VMware price rise — AT&T court filing on Broadcom licensing
- Hacker News — VMware import wizard + renewal cost reports — migration tooling and self-reported licensing cost anecdotes
Email corrections to hello@techfuelhq.com.
Frequently asked questions
Why is VMware better than Proxmox?
Is Proxmox completely free?
Why are people moving away from VMware?
Can I run Proxmox without a subscription?
What are the disadvantages of Proxmox?
Can I run ESXi on Proxmox?
Does free ESXi 8.0U3e still get patched, and is there a free ESXi 9?
Evidence ledger
- Last updated
- Methodology
- This guide was written and edited by Lowell K. Wood IV. Specs and prices verified against vendor and project documentation current on the date above. Full editorial standard: methodology.
- Update log
- 2026-07-04 — 2026 SERP-intent rebuild + verify pass: reframed to the head 'proxmox vs esxi' intent, corrected Proxmox version to VE 9.2 (2026-05-21, Debian 13.5 / kernel 7.0), grounded all licensing to Broadcom + Proxmox primary pages (free ESXi 8.0U3e caps, 16-core/socket paid minimum, EUR120-1100 Proxmox tiers). Benchmark data point cites the primary Blockbridge NVMe/TCP technote (56-of-57, ~50% IOPS). Three migration gotchas kept as documented Proxmox behavior (VirtIO boot / INACCESSIBLE_BOOT_DEVICE, thick-LVM SAN snapshots with the VE 9 QCOW2-chain tech-preview caveat, 3-node quorum). Adversarial verify: two Reddit threads (1qa8rca, 1kdvay8) could not be re-confirmed via any mirror or archive, so their verbatim quotes, upvote counts, thread citations, and the thread-only NTFS-GUID gotcha were stripped to unattributed textbook substance. AT&T 1,050% figure re-attributed from the wrong HN link to The Register (2024-10-01) as primary. HN dollar figures kept as self-reported anecdotes. Removed the fabricated-precision performance chart.
- 2026-06-19 — 2026 reality refresh: added free-ESXi patch-freeze context, no-free-ESXi-9 dead-end, and Broadcom subscription/bundle licensing context.
- 2026-05-06 — Voice cleanup as part of AdSense hardening pass.
- 2026-04-04 — Initial publish.
- Corrections
- Spotted an error or a stale number? Email hello@techfuelhq.com. Confirmed corrections are added to the update log above.