Three Dates, One Decision: VMware's Endgame Meets a 270% Hardware Cost Shock

Week of 31 August 2026 | Compiled by Team Paringa

Enterprise Linux, virtualisation and platform news for infrastructure directors, heads of platform and CTOs. Real stories, real numbers, and what they mean for your renewal calendar.

What Happened This Week

🔴 VMware Explore opens with three deadlines and no rescue plan

VMware's annual user conference began on 31 August, and The Register's APAC editor Simon Sharwood used the moment to declare the end of VMware's two-decade run as the undisputed leader of server virtualisation. His argument is not that VMware is failing. It is that Broadcom has deliberately narrowed the customer base and is content to let the rest go.

The numbers behind that are specific. Pre-acquisition VMware had roughly 350,000 customers and more than half the server virtualisation market. Broadcom has all but said it wants 10,000 to 30,000 of them, the ones who will go all in on VMware Cloud Foundation as their private cloud and DevOps platform. vSphere Standard and Enterprise Plus are no longer sold standalone. VCF is the only door, and for most buyers it costs considerably more than their last bill, including for components they will never switch on.

Three dates now force the decision. 22 November 2026 is the third anniversary of the acquisition, when the multi-year subscriptions many customers rushed to sign before the deal closed start expiring. 31 March 2027 is when VMware's contracts with many Cloud Service Provider partners end, stranding customers who buy VMware-powered VMs from a hoster. 11 October 2027 is the last day of support for VCF 8. Miss the planning window on any of these and you are negotiating under duress.

The migration market is already visible in the results. Red Hat has taken over $680 million in virtualisation orders from a standing start. Nutanix is winning hundreds of former VMware customers per quarter. HPE reports high double-digit new logo growth on VM Essentials. Rimini Street says third-party VMware support is its fastest-growing service. And Adam Centorrino, CEO of Australian services firm Centorrino, moved his own business from VMware to SUSE ahead of the March 2027 partner cut-off and is taking clients with him. Sharwood's blunt conclusion: none of the alternatives fully match VMware technically, but plenty are good enough, and the real fight is now over who hosts the AI and Kubernetes workloads.

Sources: The Register

Treat those three dates as procurement deadlines, not technical ones. The mistake we see repeatedly is teams starting the platform evaluation six months before the contract ends, which hands Broadcom the pricing power. The commercially useful move is to segment the estate now: which VMs genuinely need VCF, which are ordinary Linux workloads that can move to KVM or SUSE, and which are containerised already. Most estates split further than the CIO expects. And note the pattern in the Centorrino example: the migration is being led by service partners protecting their own margin, not by vendors. If your provider has not told you where they land on 31 March 2027, ask this week.

🟡 Kubernetes 1.37 starts binning the plumbing your clusters still run on

Kubernetes v1.37, released 26 August and nicknamed Garhwal, carries 67 changes and a deprecation list that will cost real engineering hours. kube-dns is retired in favour of CoreDNS, and clusters still running it stop working at Kubernetes 1.40. kube-proxy drops support for Linux IPVS from 1.43, with nftables named as the path forward. cgroup v1 is on the way out too, and since v1.35 Kubernetes has refused to initialise nodes still relying on it, with only a temporary workaround.

There is also a governance signal worth reading. The Metrics API finally reached general availability after nine years stuck in beta, which the release notes describe as formally recognising an API already proven in production. CNCF survey data puts 82% of container users running Kubernetes in production, up from 66% in 2023, and 66% of organisations hosting generative AI models use Kubernetes for at least some inference workloads.

The release was built by over 1,700 contributors from 212 companies across a 15-week cycle. Also new: KYAML, a strict YAML subset that kubectl already understands, plus pod-level checkpoint and restore in alpha and scale-to-zero horizontal pod autoscaling in beta.

Sources: The Register

Deprecation lists are budget events, not release notes. If your distribution vendor is doing the backporting and lifecycle work, these three removals cost you a change window. If you are self-managing, they cost you a project. That gap is the entire commercial case for a supported Kubernetes distribution, and it is a better argument than any feature comparison. Worth checking now: how many of your clusters still resolve via kube-dns, and does anyone own that answer.

🟡 Server DRAM up 270%, and it lands on your renewal not your hardware line

TrendForce forecasts cloud provider capex will nearly double this year, up 98%, then grow another 50% in 2027. The striking part is the mix. DRAM and NAND flash will account for 47% of total hardware spend this year and 68% in 2027. Server DRAM components are set to be up 270% year-on-year by the end of 2026, with enterprise SSD prices up 235% over the same period.

This is already flowing downstream. PC prices have climbed by double-digit percentages, PC shipments are down 5%, and smartphone shipments are forecast to fall 15% this year. OVHcloud warned earlier in August that it plans to raise charges by up to 87% to cover memory costs. Nvidia is reportedly telling customers to expect 15% more for Vera Rubin and Grace Blackwell products.

For infrastructure directors the practical consequence is simple. Hardware refresh, cloud consumption and AI capacity are all getting more expensive at once, and none of those three levers is under your control. The lines you can still move are software support, licensing and the operating model around them.

Sources: The Register

This is the context that makes support renewals interesting again. When hardware and cloud are both inflating at double or triple digits, a 50% reduction on Linux support is one of the few savings a platform lead can actually deliver this financial year without a migration programme. SUSE Multi Linux Support replaces Red Hat support at roughly half the cost, and it does not require you to touch the operating system underneath. A global financial institution we work with is taking over £500k a year out of its run cost this way. That is the same order of magnitude as an entire refresh cycle's inflation, and it lands in months rather than years.

What infrastructure directors are dealing with

Three cost shocks are converging on the same budget. Broadcom is repricing virtualisation upward and shrinking the customer list. Memory pricing is repricing every piece of hardware you buy. And AI workloads are arriving with capacity demands nobody budgeted for two years ago. The result is that infrastructure teams are being asked to absorb structural price rises while their own headcount stays flat.

The underlying shift is that negotiating power has moved to the vendors on hardware and hypervisor, and away from them on the operating system. Enterprise Linux is the one layer of the stack where there are genuinely interchangeable, fully supported alternatives with a decade-plus of runway. That makes it the obvious place to take cost out first, and it buys the political capital to fund the harder virtualisation decision that 2027 is going to force anyway.

Three Things Worth Watching Next Week

1. VMware Explore announcements — Watch whether Broadcom acknowledges the November 2026, March 2027 and October 2027 dates at all. Silence confirms the strategy and means your renewal conversation gets no easier.

2. Cloud provider price notices — OVHcloud has signalled up to 87%. If a second European or UK provider follows in September, memory-driven repricing becomes an industry norm rather than one operator's problem.

3. Kubernetes 1.37 adoption guidance from distribution vendors — SUSE, Red Hat and the managed Kubernetes providers will publish their kube-dns, IPVS and cgroup v1 migration timelines. Those timelines set your change calendar for the next two quarters.

Questions? hello@paringa.ai

Paringa are Infrastructure and AI Security, Cost and Governance Specialists. We find the cost and risk that sits between your platforms, your licensing and your partners, then we take it out. | 31 August 2026 | © Paringa Ltd