The State of the VFD 2026

Variable frequency drives (VFDs) control roughly 40 percent of industrial motors, which makes their current state a question about equipment nearly every plant already owns rather than about what a catalog advertises.

This report assesses that state from public documents only. No distributor portal, no login, no non-disclosure material, no vendor briefing, no sample unit, no supplier conversation. Three frontier research instruments were run independently against one structured prompt, with declared source ranks and excluded domains, and a separate scholarly harness was run against an academic works index with live document identifier resolution. A finding reached the report only when it survived disagreement between the instruments, and where they split, the split is reported rather than averaged away.

No manufacturer paid for placement, reviewed a draft, or was contacted during the work.

Five findings both editions carry

Each of these carries a document number and a section that shows the work.

• No drive from any manufacturer appears in the ISASecure certified components register. A specification clause reading "component shall hold ISASecure CSA certification" is currently unmeetable by any drive on the market. If that clause is in your standard specification today, it is either being waived quietly on every job or it is excluding every bidder.
• Field return rates: zero of the 19 brands reviewed publish them. Thirteen companies scored, and not one will say how often the product comes back. That is the finding underneath the whole report.
• IEEE 519-2022 states its limits apply at the point of common coupling and are not intended for evaluating equipment. The wording changed more than the numbers did, and the wording is the part that matters.
• Safe Torque Off is base build rather than a priced adder, down to the value tier.
• Silicon carbide has not shipped in any general purpose low voltage drive. The silicon shipped and the drives did not, and seven years of module launch coverage reading almost identically to drive launch coverage has produced a widespread belief otherwise.

Then it forks, because the job is not the same job

Both editions share the method, the 19 brands, the rubric, and the five findings above. What differs is which argument gets opened and which matrix sits at the back. Pick the one that matches what you have to defend.

Plant Edition

For the person who has to put a number in a specification, or explain to a plant manager why a drive that was supposed to last did not.

Two findings appear only here. One mid-tier manufacturer publishes wear-component life with four stated conditions, and it is the only one that does, while three of the largest companies in the industry publish nothing comparable. And drive-resident bearing diagnostics rest on a single seeded-fault dataset: a drive that counts run hours, load and heatsink temperature is doing something real, but a drive claiming to read your motor's bearings from current signature alone is making a claim the published evidence does not yet support.

It closes with 15 questions to put to a vendor that have public answers you can check before the quote arrives.

Channel Edition

For the person who has to put a number in front of a customer who checks.

Two findings appear only here. The low voltage AC drive market is approximately $15 billion growing at about 3 percent, not the $30 billion at 6 percent in wide circulation, which describes a much wider basket. And global heat pump sales fell about 2 percent in 2025, which matters to anyone who has built a pitch on that segment. It also carries the shared-platform argument that turns a rebadge accusation into a dead end, and a matrix showing exactly which questions have public answers and which do not.

That matrix is the part worth having. Every cell marked as gated is a document you can pull and your customer cannot.

Most of it is a correction

Nearly every section reports something widely believed that is not supported by the public record, or is supported by a document saying something narrower than the claim built on top of it. That is deliberate. A report confirming what you already believe is not worth an email address, and the test each finding had to pass was whether it changes what a competent engineer would write in a specification, ask a supplier, or put in a budget.

What it costs you is five or six comfortable talking points. What you get back is a verification queue you can work through in an afternoon.

This is not a product recommendation and it is not a market forecast. It does not replace a vendor's own current documentation, which is why every document number and revision date appears in full.

The method is published, not summarized

The last third of either edition is the apparatus rather than the argument. The nine-dimension rubric appears in full, with the rule that governs every cell. So does the disclosure of where AI research instruments were used and at what level of influence, the competing interests statement, and a section on errors and omissions and why they form part of the finding rather than an embarrassment to it.

There is also a section telling you how to check the document itself. A report built on the claim that manufacturers should show their evidence does not get to hide its own.