Typical licensing increase after the Broadcom transition.
TRITON PRIVATE CLOUDRun your own cloud.
Talk to sales
Run your own cloud.
Not your own cloud stack.
Triton Private Cloud is the operator-grade replacement for VMware, a full cloud operating model on hardware you own, built on illumos, ZFS and bhyve, with no per-socket licensing.
LINEAGEJoyent → Samsung → MNX → Sovren
SUBSTRATEillumos · zones · ZFS · bhyve
IN PRODUCTIONFortune 500 customer under contract
THE THESISGARTNER300–400% CLOUDBOLT · JAN 202686% FORRESTER~25% GARTNER FORECAST70→40%
Broadcom is deliberately shedding 290,000 customers to serve 10,000. Somebody has to catch them.
Of enterprises are actively reducing their VMware footprint. Only 4% have finished.
Average share of existing footprint that actually renews with VMware.
VMware's share of server virtualization, 2024 → 2029.
The migration is real but slow, and that is precisely the opening. Displacement compounds over the 2026–2029 renewal cycle, and the incumbent has dismantled the very channel that would have defended it.
Gartner (P. Delory, May 2026) · CloudBolt Industry Insights, n=302, Jan 2026 · Forrester (N. Chhabra, Jun 2026) · Gartner share forecast, Sep 2025
PROBLEM · 1 OF 3
The bill went up by a multiple, not a percentage.
Broadcom floated a 4.5× increase in minimum license size in early 2025, 16 cores to 72, and retreated only after channel revolt. Intent, on the record.
JAN 2024Perpetual licenses withdrawn from sale.
MAR 202520% surcharge for missing a renewal anniversary.
OCT 2027vSphere 8 end of support, forced move to VCF 9.
We do not sell the 10× number. We sell the uncertainty: 88% of enterprises say they are shaping strategy around expected future increases.
The Register (AT&T filings Oct 2024; Beeks Dec 2024; Arrow Mar 2025; CISPE Mar 2026) · TechTarget (FedHIVE May 2026) · CloudBolt Jan 2026
PROBLEM · 2 OF 3THE STRATEGY, Optimize the top 10,000 THE OPENING, 290,000 accounts and no channel
The incumbent fired its own distribution.
Authorized VMware cloud service providersTHOUSANDSbefore Jan 202619 in the US · 9 in the UKafter the VCSP program endedVCF adoption among Broadcom's top 10,000 customers: over 87%
Broadcom's software revenue is up 26% and VCF penetration in the top decile exceeds 87%. The strategy is working, for 10,000 accounts.
VMware's base exceeds 300,000 customers. The partners who served the long tail have been de-authorized. Those customers now need a platform and a partner.
The Register, Jan & Mar 2026 · Network World, Jun 2026 · The Register, Aug 2026 (Azure VMware Solution)
PROBLEM · 3 OF 3GARTNER35% NUTANIX, APR 20261,000+ RED HAT, AUG 2026$680M+
A 30-point share transfer is already forecast.
0% · 20% · 40% · 60% · 80%
- VMware share is 70% in 2024, 67% in 2025, 60% in 2026, 50% in 2027, 45% in 2028, and 40% in 2029.
- Displaced to alternatives is 0% in 2024, 3% in 2025, 10% in 2026, 20% in 2027, 25% in 2028, and 30% in 2029.
of VMware workloads migrate to other platforms by 2028.
VMware migrants won in a single quarter, its strongest new-logo quarter in eight years.
in OpenShift Virtualization orders, from a standing start.
Gartner via Network World, Sep 2025 & The Register, May 2026 · SDxCentral, Apr 2026 · The Register, Aug 2026. Displacement line is the arithmetic complement of Gartner's VMware share forecast.
WHAT WE WILL NOT CLAIMWHY THAT IS THE WHOLE OPPORTUNITY
It is an erosion, not an exodus. That is better for us.
86%REDUCING FOOTPRINT4%FULLY MIGRATED
Ninety-six percent of the displacement has not happened yet. Contracts unwind over the 2026–2029 renewal cycle, one workload class at a time, which means this is a market being entered, not a market already lost.
"You will emerge with more complex and less capable infrastructure. No vendor offers a one-for-one replacement."PAUL DELORY, VP ANALYST, GARTNER, MAY 2026
Gartner's objection is aimed at the hypervisor swap, trading vSphere for another hypervisor and inheriting a worse operating model. It is a correct criticism of every alternative. It is not a criticism of replacing virtualization with a cloud.
CloudBolt Industry Insights, n=302 North American IT decision-makers, director+, companies 1,000+ employees, fielded Jan 2026 · Gartner via The Register, May 2026
THE GAP IN THE MARKET
Every alternative sells a hypervisor. Enterprises are trying to buy a cloud.
| Platform | What it replaces | Tenancy | Metering | Object storage | Licensing basis |
|---|---|---|---|---|---|
| Proxmox VE | The hypervisor. KVM + LXC, HA. | Limited | No | External (Ceph) | €370–1,100 / socket / yr |
| Microsoft Hyper-V | The hypervisor, in a Windows estate. | System Center | Partial | No | Windows Server / per core |
| Nutanix AHV | Hypervisor + HCI appliance. | Yes | Partial | Add-on | Node subscription |
| OpenShift Virt. | VMs inside Kubernetes. | Yes | Partial | Via ODF | Per core / socket-pair |
| XCP-ng · Scale | The hypervisor / the appliance. | Limited | No | No | Support subscription |
| Triton | The cloud. Hypervisor is one layer. | Silo→tenant→project | Native | S3, built in | No per-socket licensing |
Gartner counts 30+ alternative platforms, most "incomplete or still in development." The crowd is in the hypervisor lane. The unserved requirement is the operating model that made public cloud worth having: self-service, projects, quotas, APIs, cost attribution.
Proxmox published price list (per socket/yr, ex-VAT) · Microsoft Azure Local & Windows Server licensing pages · vendor documentation · Gartner via ITdaily, Nov 2025. Nutanix, Red Hat and Azure Local list prices are not published.
PRODUCTCLOUD ECONOMICS, Chargeback, not allocation PORTABILITY, Same software as the public region
Don't stop at the hypervisor.
ACCESSPortal · CLI · API · Automation
CLOUD MODELSilo · Tenants · Projects · Identity · Policy
RESOURCESCompute (hypervisor is one part) · Network · Block · Object
HARDWAREServers · NVMe · Network fabric · GPUs
Virtualization stops at the resource layer.
Operators define capacity, policy and boundaries; teams consume on demand.
Every instance, volume and object is metered and attributed. Finance gets chargeback against real consumption instead of a capex allocation spreadsheet.
Triton Private Cloud runs the same platform as Triton Cloud. Same software, same APIs, different operator. Burst and portability are architectural, not a roadmap item.
TECHNOLOGYSUBSTRATE CONTROL PLANE NETWORKING STORAGE
A new control plane on a twenty-year substrate.
illumos
Zones, ZFS and DTrace in the host OS. Virtual machines run on bhyve. Storage is ZFS all the way down, no bolted-on SDS tier.
tritond · tritonagent
One HTTP service owns all state and the entire API. One agent touches compute nodes. State in FoundationDB, metrics in ClickHouse. Installs with a single command.
Proteus
An illumos kernel module in Rust on Oxide's OPTE packet engine. Per-port flow tables, distributed firewall, routing, NAT, floating IPs, Geneve overlay, DHCP, IMDSv2.
Manta
Rebuilt on FoundationDB and ZFS. S3-compatible object storage end to end, clustering across nodes, with a POSIX filesystem shipping as a Triton zone.
"Same substrate, new codebase. Triton Cloud is not Triton DataCenter with a new coat of paint."
AVAILABLE Triton control planeIN DEVELOPMENT Manta data servicesPLANNED Managed Kubernetes
WHY US2009BUILT 2015NAMED 2016–22SCALED 2022–24ACQUIRED 2026REBUILT
Seventeen years of production cloud, and we own the codebase.
Joyent
SmartDataCenter · SmartOS · zones · ZFS
Triton DataCenter
Multi-tenancy · SDN · APIs · Docker-native containers
Samsung
Triton Private Regions in production · Samsung Cloud object storage on Triton
MNX → Sovren
Triton, Manta and SmartOS suite · 608 repositories acquired
Triton Cloud
New control plane · Rust networking
Competitors are assembling a cloud out of open-source parts. We acquired a cloud platform that ran a hyperscaler's object storage, kept the substrate that made it fast, and rewrote the control plane that made it hard to operate. First recognized OCP Solution Provider running Triton on an OCP-recognized hardware platform, MiTAC compute with Edgecore networking.
Open Compute Project solution registry · Sovren Technologies acquisition record
DIFFERENTIATIONWEDGE 01 WEDGE 02 WEDGE 03
Three things no one else brings together.
Rack to API, turnkey
Open Rack V3, 44 OU, 48 V DC bus, A/B Ethernet fabric, arrives configured, with the Triton control plane installed by a single command. Not a bill of materials and a services engagement.
The alternative for a 300-person IT org is standing up Ceph, an SDN, an IdP and a metering layer themselves. That is the project Gartner is warning them about.Built for GPUs, not retrofitted
Accelerated compute is a first-class tray in the rack design and a first-class resource in the API. Private AI, inference and training run beside the data they need, inside the tenancy and quota model, not in a separate silo.
Broadcom's own commentary attributes VMware growth to "rising CPU core deployments alongside GPUs."No per-socket licensing
Open-source platform, open hardware, no per-socket and no per-core licensing. Customers buy capacity and support, not the right to turn on the cores they already own.
This is the structural answer to the entire Broadcom story: there is no lever here for a future 300% increase, because the pricing is not attached to the customer's core count.Broadcom Q2 FY2026 commentary via Futurum, Jun 2026 · Triton Open Compute reference architecture
ECONOMICSTHE TRAP, 96 cores THE TRITON POSITION, Density is free
The core count stops being a pricing lever.
Platform cost as core density per socket rises
Illustrative of pricing structure, not a quoted TCO comparison. The shape is the point: every core a customer adds to a per-core platform is a price increase they did not negotiate.
Current-generation server CPUs. Under per-core licensing, a hardware refresh is a licensing event, the customer pays more to run the same workloads on fewer, denser boxes.
Consolidation improves the customer's economics instead of ours. That aligns our renewal with their efficiency, the exact relationship Broadcom inverted.
Proxmox published per-socket price list · Microsoft Azure Local per-physical-core model · VMware VCF subscription terms. Curves are structural illustrations, not vendor quotes.
WHO BUYSLAND EXPAND MULTIPLY PREMIUM
Four entry points, one platform.
Existing applications
Run the VMs they already have. The lowest-risk first milestone and the one every VMware refugee starts with.
Internal cloud
Platform teams give engineering self-service infrastructure with quotas and chargeback, the reason the workload went to AWS in the first place.
Service provider
The de-authorized VMware partners need a platform to resell. Multi-tenancy, metering and white-label are already in the product.
Private AI
Keep compute close to the data. GPU capacity inside the same tenancy, policy and metering model as everything else.
LAND AND EXPAND
A customer enters on the cheapest, most defensible motion, "run the VMs you already have", and the tenancy, metering and object-storage layers turn a hypervisor replacement into a platform relationship. The private-AI segment carries the highest revenue per rack and pulls forward the hardware line.
THE HONEST POSITIONWHERE WE WIN
Nutanix and Red Hat are winning today.
Both have enterprise sales motions we do not yet have at scale, and both are taking VMware workloads right now. If you are a large enterprise with an existing relationship and a preference for an appliance model or a Kubernetes-first stack, they are credible choices and you should evaluate them.
Neither sells the abandoned mid-market or the de-authorized service provider, both are priced and staffed for the enterprise. Nutanix ties the customer to appliance economics; OpenShift asks them to adopt Kubernetes to run a VM. Triton is complete at the cloud layer, which is an empty quadrant.
HOW A MIGRATION STARTSSTEP 01 STEP 02 STEP 03 STEP 04 TIMING
A migration that starts small enough to say yes to.
Map
Workloads, hardware and dependencies. Output is a defensible target-state design, not a quote.
Run a slice
Triton on a defined slice of infrastructure. One rack, one tenant, real workloads.
Validate
Workloads, operations and automation, against the customer's own runbooks and change process.
Move capacity
Expand where the platform and the workload fit. Renewal dates set the pace, not us.
vSphere 8 support ends Oct 2027; Azure's license-included VMware option stops selling Oct 2026. Every enterprise has a dated forcing event.
PROOFWHAT IT PROVES, 188 nodes, priced per node WHAT IT DE-RISKS, The channel motion works
Not a pilot. An executed purchase order.
End customerA Fortune 500 life sciences company
Reseller of recordA national IT solutions provider
Coverage termTwelve months, 2026–2027
Scope8 head nodes · 180 compute nodes
Contract value$600k+
Not a proof of concept, a 188-node production estate under contract. And the invoice is denominated in nodes, not cores: the customer can put 96-core CPUs in all 180 compute nodes without the licence changing.
The order came through a national IT solutions provider with an established enterprise book. We did not have to build a direct enterprise salesforce to land a Fortune 500 logo.
Executed purchase order, July 2026. Customer and reseller names withheld.
APPENDIXCITED EXCLUDED ON PURPOSE
Sources, and what we deliberately left out.
Gartner, Worldwide Sovereign Cloud IaaS Spending, 9 Feb 2026 · Gartner VMware share forecast via Network World, 9 Sep 2025 · Gartner (Paul Delory) via The Register, 12 May 2026 · Gartner 30+ alternatives via ITdaily, 17 Nov 2025 · CloudBolt Industry Insights, n=302, fielded Jan 2026 · Forrester (Naveen Chhabra) via Network World, 3 Jun 2026 · Nutanix Q4 FY2026 results, 27 Aug 2026 · SDxCentral, 8 Apr 2026 · The Register: AT&T filings Oct 2024, Beeks Dec 2024, Arrow Mar 2025, VCSP termination Jan 2026, CISPE Mar 2026, Azure VMware Solution Aug 2026 · TechTarget, May and Jun 2026 · Computer Weekly, 3 Feb 2026 · Futurum, 8 Jun 2026 · Proxmox published price list · Microsoft licensing pages · Open Compute Project registry · Executed purchase order, July 2026.
"74% EXPLORING ALTERNATIVES"
Traces to a self-selected Gartner Peer Community poll (n=246, date not published), not Gartner research. We use CloudBolt's 86% instead.
"72-CORE MINIMUM"
Announced by Arrow in Mar 2025 and withdrawn within two weeks. Presented as evidence of pricing intent, not current policy.
SERVER VIRTUALIZATION TAM
Available figures come from report-mill publishers whose estimates differ by 5–10× with unpublished methodology. Sized bottom-up instead.
CLOUD REPATRIATION STATISTICS
The widely-quoted repatriation numbers have no primary source. CloudBolt finds 72% of departing VMware workloads go to public cloud, which we address rather than hide.
Every figure on this page is attributed to a named analyst, a company filing, an executed purchase order, a published price list, or a dated trade report.
The window is the 2026–2029 renewal cycle.
The platform exists. The customer is under contract. The channel is available.
Talk to salesMarket data verified September 2026.