The Quiet Case for Getting Your Business Off Windows Before 2028

This is a labeled opinion piece — a grounded prediction, not a migration guide. Most small businesses shouldn’t switch operating systems today. The argument here is about the next three years, and about buying optionality cheap while you still have the room.

Nobody writes procurement policy off a Hacker News thread, and they shouldn’t. But every once in a while, three unrelated stories in three weeks land in the same direction, and the interesting move isn’t to jump — it’s to ask the question early, before it becomes obvious.

Three stories this July did that for me. In the middle of the month, Microsoft confirmed to ghacks.net that Windows now carries a device identifier known as GDID that cannot be disabled through any user-facing setting. Days later, videocardz.com and others reported that LG monitors had been caught silently installing companion software through Windows Update itself, with no consent prompt and no visible entry in the usual install list. And earlier in the same month, linuxiac.com noted that Windows’ share of the global desktop market had slipped below 60% for the first time in years — a trend, not a stunt, that has been continuing quietly for eighteen months.

Consumer tech press covered each of those as a separate outrage cycle. Different websites, different comment sections, different weeks. I think they add up to something more specific and less dramatic: a slow, quiet repricing of what “we just run Windows on everything” actually costs a small business — and, more interestingly, the beginning of a case that the hardware choices your business is about to make for entirely unrelated reasons will start pushing you off Windows even if the privacy stories never touch a hair on your head.

I want to be clear about what this piece is and what it isn’t, right up front, because the topic invites both overreaction and dismissal.

It is a labeled opinion piece. Prediction sections are marked as such. It is not a “switch to Linux this weekend” evangelical post — most small shops should not switch today, and I’ll steelman that hard below. It is a case that the cost of Windows-by-default is rising in ways that don’t show up on the license invoice, and that spending ninety days buying yourself cheap optionality is a bet that pays off in every scenario I can think of, whether or not the pessimistic version of the story comes true.

Signal 1 — The device identifier you can’t turn off

Windows has always had telemetry. That isn’t new. What is new — and quietly significant — is that Microsoft has now confirmed the GDID is a per-device identifier that persists across resets, cannot be reset by the user, cannot be disabled through Group Policy in any documented way, and is tied to the physical hardware in a way that survives reinstallation of the OS. Before this confirmation there was a widespread assumption that the various Windows advertising and telemetry IDs were, at least, resettable. That assumption is now wrong.

For a consumer running one laptop, this is annoying and creepy but not immediately expensive. For a business running any number of Windows machines — five, fifty, five hundred — it’s a different animal. You now have an unresettable hardware identifier associated with every machine on your books, and you cannot honestly tell an auditor, a client, or a compliance questionnaire that you have full control over what identifiers your endpoints emit. That answer used to be, at worst, complicated. It’s now definitively no.

That change is small in isolation. It compounds against the next one.

Signal 2 — Silent installs via Windows Update

The LG story is important not because it’s about LG monitors but because of the channel it exposed. Software installed through Windows Update carries the trust of Windows Update. Users — including IT staff — treat that channel as the OS updating itself. When third-party vendor software rides along that channel without a consent prompt, the mental model of “Windows Update updates Windows” is quietly, materially wrong.

The specific incident affected a specific monitor line and, as of this writing, appears to be limited in blast radius. But the fact that it could happen without breaking any of Microsoft’s rules or triggering any warning tells you something important about the channel. Every business running Windows Update on autopilot — which is essentially every business running Windows — has now watched a demonstration that “trusted vendor software you didn’t consent to” is a fully supported category on the platform.

Again, small in isolation. Compounds against the third one.

Signal 3 — The share number

Windows crossing under 60% of the global desktop market share is not a signal to switch. It’s a signal that the ecosystem assumption — the one under every third-party vendor’s product roadmap and every job posting for internal IT — has started to shift. When Windows was 90% of the desktop, “we’re a Windows shop” was a decision that made itself. When it’s 58% and drifting, the “and drifting” part matters more than the exact number. Software vendors are quietly increasing their investment in Mac and Linux support because they have to. Cloud services are neutralizing OS assumptions because their customers are. The moat around Windows-by-default is getting narrower every year.

None of this individually is a “get off Windows” argument. All of it together starts to change the answer to a question that, up until recently, didn’t feel like a question at all: why are we a Windows shop?

Why this hits businesses differently than consumers

Consumer tech coverage of these stories treats them as personal privacy issues, and for consumers, that’s about right. Business exposure works differently, and it’s worth naming the specific ways.

Compliance questionnaires have started asking. Insurance renewals, enterprise vendor onboardings, and — increasingly — mid-market client due-diligence questionnaires now include line items about endpoint telemetry, device identifier controls, and data flows out of your fleet. Two years ago, “we run Windows” was a complete answer to those. Today, “we run Windows” is a question that gets follow-ups.

Cyber-insurance is pricing this in. Premiums for small businesses have been climbing across the board, and one of the quiet drivers is that carriers are increasingly asking about telemetry surface area and vendor sub-processing. You do not need this to be a huge factor in your premium to notice that it wasn’t a factor at all three years ago.

Client data obligations translate down. If any of your clients are under HIPAA, GDPR-style regimes, financial-services rules, or a large-enterprise vendor code of conduct, obligations flow to you as their vendor. “We have no control over the identifiers our endpoints emit” is a genuinely uncomfortable position to be in when a client asks a pointed question. It doesn’t matter that this is also true of every other Windows shop; the question is being asked more often, and you’d rather have a better answer than everyone else than the same answer as everyone else.

Audit trails get harder. When software can be installed on your fleet through the OS update channel without going through your normal endpoint management, your ability to reconstruct “what was on this machine on this date” becomes worse. Not catastrophically worse. Meaningfully worse.

None of these are apocalyptic. They’re all instances of the same pattern: Windows-by-default is quietly costing you more than it did three years ago, and the cost is going up rather than down.

The prediction (labeled opinion)

Here is my grounded prediction, labeled explicitly as opinion.

By 2028, “what OS does your fleet run” becomes a standard line item in small-business risk reviews. Not a compliance failure — nobody’s going to tell you Windows is disqualifying. But a justified choice. If you run Windows, you’ll be expected to be able to articulate why, what your telemetry controls look like, and what your fallback plan is if the vendor’s policies drift further. “It’s what we’ve always run” will stop being a sufficient answer for anyone above a very small size.

Corollary prediction: the small businesses that are quietly evaluating alternatives in 2026 and 2027 will have massively more optionality — and better negotiating leverage — than the ones who wait until it’s obvious. By the time an insurance carrier or a large client asks you the pointed question, you either have a story or you don’t. Building the story after you’re asked is much harder and much more expensive than building it now, while there’s no pressure.

I want to be honest that this is a prediction, not a fact. It could be wrong. Microsoft could change course on the identifier question; the vendor-silent-install channel could be tightened; the share number could stabilize. If any of that happens, some of the specific arguments above weaken. What doesn’t weaken is the general observation that “our OS choice is a real decision now, not a default,” and that observation is worth acting on regardless of which scenario plays out.

The honest steelman

Now the other side, and I mean it.

Most small businesses should not switch today. Windows still wins, decisively, on the things a small business actually spends its time doing. Line-of-business Windows-only software is real and often irreplaceable — accounting packages, tax software, industry-specific tools, POS integrations, payroll systems. Staff familiarity is enormous; the cost of retraining ten people on a new desktop paradigm dwarfs any theoretical privacy gain. Hardware compatibility is broad and cheap. Support labor is abundant and inexpensive. Almost every vendor you buy from tests on Windows first. The macOS alternative is expensive per-seat and has its own set of vendor-lock-in concerns. The Linux alternative is powerful and free but comes with a support-and-familiarity tax you should not underestimate.

If you are running a fifteen-person operation on Windows today, with staff who are productive, software that works, and vendors who support you — switching operating systems in Q4 of 2026 would be an unforced error. That is not what I’m suggesting.

What I am suggesting is that you spend a small amount of time and money buying optionality, so that if the pessimistic scenario plays out you’re not caught flat-footed, and so that when you replace hardware in the ordinary course of business you have real information about what your alternatives would cost. The steelman for staying on Windows is strong today. It has been getting slightly weaker every year for four years. Optionality is the hedge.

The twist that changes the math

Here’s the interesting piece that consumer coverage of the Windows stories missed entirely.

The most exciting stories in small-business AI infrastructure this month are about running capable models on hardware you already own, or on hardware that’s cheaper than a laptop. In the past two weeks alone we’ve seen an 80-billion-parameter model running in 4.3 GB of RAM on a Mac, a 26-billion-parameter model running in 2 GB on modest hardware, and a serious open-weights model running on a single GPU. Cloudflare has publicly committed to running open-weight models like Kimi and GLM in production at scale, calling out “smaller, faster, safer” as the direction. Big infrastructure is voting with real dollars that open models on capable hardware are the near-future baseline.

Look at the platforms those stories run on. Overwhelmingly Mac and Linux. Not because those platforms are politically preferred, but because the toolchains (MLX on Apple Silicon; the entire quantization and inference ecosystem on Linux; container tooling; GPU driver stacks) are more mature and more open there. Windows works for some of this, especially with WSL, but it is not where the frontier is developing.

Now project forward eighteen months. If your business ends up wanting to run any meaningful local AI — internal document Q&A, email triage, customer support first-line, contract summarization, whatever the practical use case is — the hardware you buy for that job is going to make the OS choice for you. A Mac Mini or a mid-tier Linux workstation running local inference for your team is going to be an obvious purchase for a lot of small operations by the end of 2027. And the moment you own one of those, “we’re a Windows shop” starts being “we’re a Windows shop, except for the AI box in the corner that we also depend on.” That’s not a stable equilibrium.

The Windows privacy stories give you one reason to build optionality. The local-AI hardware trend gives you a completely orthogonal, much more practical reason to build the same optionality. The two arguments meet in the same place: over the next two years, having a working non-Windows footprint in your business will go from unusual to normal.

A no-regrets 90-day plan

This is what I would actually do if I ran a small business today. It is designed to be cheap, low-risk, and to produce useful information regardless of which future scenario plays out.

Days 1–14 — Dependency inventory. Walk your operation. For every workflow that requires a computer, write down: what software is being used, what specifically it does, whether it runs only on Windows, and what the switching cost would look like. This is a one-time exercise that most small businesses have never done. The output alone is worth the ninety days — it’s your first honest map of your OS exposure, and it will surprise you.

Days 15–30 — Pick one non-critical role. Identify a job function that is Windows-by-inertia rather than Windows-by-necessity. A marketing coordinator, a bookkeeper who lives in a browser and a spreadsheet, a small-business owner who mostly does email and video calls. Not your accountant. Not your production line. A role where an OS trial won’t take the business down if it goes sideways.

Days 31–60 — Trial one non-Windows machine. A Mac Mini or a modest Linux workstation for that role. Not to replace their Windows machine on day one — as a parallel setup they can use for real work for a month. Buy a laptop for the person if the role is mobile. This is the point where the theoretical becomes concrete. You will learn things about your actual workflows that no amount of research would have surfaced.

Days 61–75 — Document what breaks. Keep a running list. Which vendor logins don’t work as smoothly. Which shared documents render differently. Which peripheral needs a different driver. Which workflow needs a workaround. This is your real dependency map — not the imagined one, the actual one. Most businesses discover the list is far shorter than they feared and far more concentrated (three or four specific vendors, not fifty).

Days 76–90 — Decide. You now have a real answer to “could we operate a mixed-OS environment for the next hardware refresh cycle?” and, more importantly, you have a real answer to “what specifically would we need to solve for?” Whether the answer is “we’re staying all-Windows and here’s why with specifics” or “we’re going mixed on the next refresh” — either answer is better than the answer you had ninety days ago. That’s the definition of no regrets.

Cost of this plan: one workstation you keep either way, and roughly one person-day per week of attention for a quarter. Value of this plan: the map of your own operation you never bothered to make, plus concrete information you don’t currently have, plus the optionality of being three months ahead if the pessimistic scenarios materialize.

What I’m not asking you to do

I am not asking you to switch. I am not asking you to buy into a free-software worldview. I am not predicting Windows will collapse or Microsoft will suddenly become worse in some dramatic way. I don’t think any of those things are likely. What I’m suggesting is more modest and, I think, more defensible: the default nature of Windows-by-default has quietly ended, and the small businesses that treat their OS choice as a real decision — with a real map, a real fallback, and a real answer — will be better positioned than the ones who don’t, whether or not any of the specific concerns above ever become acute.

The three-week window of stories that prompted this piece will fade. The pattern they represent won’t. Nobody is going to tell you at a specific moment that the deadline is here. That’s exactly why the boring, cheap, ninety-day version of the work is worth doing now, quietly, before it’s a story.

Sources

• ghacks.net — coverage of Microsoft’s GDID device identifier confirmation, published mid-July 2026 (~3 weeks old at publication). Primary source for Signal 1.
• videocardz.com — reporting on LG monitor silent-install incident via Windows Update, July 18, 2026. Primary source for Signal 2.
• linuxiac.com — Windows desktop share reporting, July 8, 2026, referenced for trend framing on Signal 3.
• Cloudflare — blog on running open-weight Kimi and GLM at scale, August 3, 2026, cited in the local-AI hardware section.
• Swiftlet — 80B Qwen on Mac in 4.3GB RAM demonstration, August 3, 2026, as evidence of the hardware trend.
• Related project — Gemma 4 26B in 2GB demonstration, “turbo-fieldfare”, July 29, 2026 (~1 week old at publication), same trend line.

Freshness blip: Signals 1–3 are approximately three weeks old at publication, right at the freshness boundary; local-AI hardware evidence is under two weeks old. This is a labeled prediction/opinion piece — the underlying repricing question is durable regardless of when specific stories fade.