Why Do Businesses Still Run DOS? Legacy Software vs SaaS Costs

Why do businesses still run DOS and legacy software instead of SaaS?

Businesses still run legacy software because it's paid off, it works, and replacing it costs more in money, risk, and downtime than keeping it running. That's the whole answer. The interesting part is what that decision reveals about the alternative most companies are choosing instead: a subscription stack that gets more expensive every year it stays in place.

Which businesses are still quietly running DOS?

This isn't a museum piece. These are live systems running the show right now:

• An insurance company's entire IT infrastructure has reportedly run on air-gapped DOS for more than 15 years, with only two system collapses in that time and no plan to migrate to Windows.
• Many ATMs worldwide still operate on legacy Windows builds, including Windows XP and even Windows NT from 1993, because rewriting proprietary ATM software and requalifying it for regulatory compliance costs more than it's worth.
• San Francisco's Muni Metro won't start its trains in the morning until someone feeds a 5.25-inch floppy disk into a control system installed in 1998. The agency has approved a decade-long overhaul estimated at $700 million overall, starting with a $30 million first contract, not because the DOS software failed, but because the manufacturer told them replacement parts can no longer be sourced.
• Deutsche Bahn has advertised for an IT administrator fluent in Windows 3.11 and MS-DOS to maintain driver's cab displays that have run reliably for up to 30 years and get replaced only when they stop being safety-compliant.
• A San Diego fine-art print shop keeps Windows 2000-era servers alive to drive discontinued LightJet printers because upgrading the full licensing chain was quoted at $50,000 to $60,000, more than the value of switching.

FreeDOS, the open-source MS-DOS successor, is still under active development more than 30 years after DOS left retail shelves, specifically to keep cash registers, PBX phone systems, CNC controllers, and point-of-sale terminals running on hardware whose applications talk directly to the BIOS and can't be ported without breaking.

None of this is inertia dressed up as charm. It's a pattern, and the pattern has a name: ownership.

Why does 'obsolete' software keep winning?

Once a business owns a system outright, either through a perpetual license or software it built and controls, there's no renewal date. No vendor can revoke access. No vendor can raise the price. No vendor can strip a feature into a higher tier next quarter. The system does one job, and it keeps doing that job until the hardware physically fails or a regulator forces a change.

The economics back this up. Owning software via a perpetual license or lifetime deal typically breaks even against an equivalent subscription somewhere between month 12 and month 26. After that point the two cost curves diverge hard: ownership stays flat, while subscription costs compound with every renewal and every added seat. Renting software also carries four structural risks that owned software simply doesn't have: pricing pressure at renewal, feature degradation as capabilities get pushed to pricier tiers, disruption when the vendor gets acquired or pivots, and outright sunset of the product a business depends on. The businesses running DOS in 2025 have been on the flat side of that curve for two and three decades.

What is the SaaS alternative, and why does a rented stack renew against you?

Here's what the DOS-runners are opting out of. SaaS prices rose roughly 11.4% year over year in 2025, nearly five times the G7's 2.7% inflation rate. Half of software companies are preparing further price hikes, and 60% of vendors deliberately obscure those increases through mechanisms like AI-feature bundling and 'credit multiplier' changes.

Average SaaS spend per employee climbed from $7,900 in 2023 to about $9,100 by the end of 2025, a jump of more than 27% in two years, while corporate IT budgets grow only about 2.8% annually. The math doesn't close. Subscription costs are structurally outpacing what businesses can absorb.

Meanwhile the average organization now manages around 305 SaaS applications, and large organizations manage closer to 696, but IT departments directly oversee only about 13% of that footprint. The rest gets bought and run by individual teams, compounding cost nobody is tracking centrally. The renewal pain shows up everywhere:

• 79% of IT leaders reported a price increase at renewal in the past year.
• 78% got hit with unexpected AI or consumption charges.
• 61% of organizations had to cut a planned project because of unplanned SaaS cost increases.

That's the exact opposite of a Muni train that just needs a floppy disk and a clean power supply.

Are COBOL and mainframes the same bet at enterprise scale?

The logic isn't confined to quirky DOS anecdotes. It runs the financial system. COBOL, a language dating to 1959, still underpins more than 40% of all online banking systems, supports 80% of in-person credit card transactions, handles 95% of ATM transactions, and powers systems moving more than $3 billion in commerce daily. Nobody keeps COBOL running out of sentiment. They keep it running because the accumulated business logic inside it, not the language itself, is the real asset, and replacing that logic is an operating-risk project, not a coding exercise.

The federal government tells the same story at a larger scale. A 2025 GAO review of 69 legacy IT systems found 11 in urgent need of modernization, spanning Treasury to Health and Human Services. Eight of those 11 run on outdated programming languages, four have unsupported hardware or software, and seven carry known cybersecurity vulnerabilities. Three years after GAO flagged ten systems for modernization in 2019, only three had actually been fully replaced.

Why don't businesses just upgrade?

The honest reasons legacy software survives have little to do with laziness:

• Replacement cost dwarfs the upgrade's benefit. The San Diego print shop's $50,000-$60,000 licensing quote is the norm, not the exception.
• Embedded business logic is the real product. Decades-old code encodes undocumented rules for exceptions, calculations, and account states that a rewrite has to rediscover before it can replace anything.
• Vendor lock-in cuts against switching, not toward it. Microsoft and Oracle win 25% to 30% of their federal contracts non-competitively, and one agency paid an extra $112 million just to keep Microsoft Office instead of moving to Google Workspace.
• Sunk-cost dynamics protect what's already funded. Public-sector buyers in particular treat replacing an already-paid-for system as waste, even when it's the riskier choice long term.
• Skills and parts scarcity raise the price of change. Muni's overhaul only got approved after the agency was told replacement parts for its 1998 system could no longer be sourced at all.

Add it up and 'legacy' stops looking like a failure to modernize and starts looking like a rational bet that paid off.

Is the anti-SaaS stack a strategy, not an accident?

Here's the reframe: some of this persistence is deliberate. A system that's paid for, self-contained, and not connected to a vendor's billing cycle can't be repriced, degraded, or sunset out from under the business running it. That's not a bug in these old systems. It's the design.

The same instinct is resurfacing across modern infrastructure, just with newer hardware. Teams are buying enterprise machines to run open-weight AI models instead of metering every token through a cloud API, for the same reason a print shop keeps Windows 2000 alive: predictable cost beats a renewal notice. When GitHub had three major outages in six weeks, the fix teams reached for wasn't a better vendor, it was owning more of the stack themselves. And the same lesson about remote control cuts both ways: when a vendor can remotely lock hardware you thought you owned, the DOS shop that never networked its floppy-boot control system starts looking less quaint and more prescient.

AI coding tools are lowering the cost of building and owning that kind of internal software again. A team that once had to buy a vendor's package because writing it in-house was too expensive can now sketch a working internal tool in days. That gives operators a way to build and keep the agents and workflows they actually need, instead of renting a seat license for features they'll never fully use.

What does this mean for founders and operators?

The decision isn't 'legacy versus modern.' It's 'own versus rent,' and it should be made line by line. Before renewing a SaaS contract, ask what breaks if the vendor doubles the price next year, what breaks if the feature you rely on moves to a higher tier, and what breaks if the vendor gets acquired. If the answer to any of those is 'the business stops functioning,' that's the system to evaluate for ownership, whether that means a perpetual license, a self-hosted alternative, or building it internally. The insurance company running DOS for 15 years didn't get lucky. It made a bet that stability beats features, and the bill for renting instead of owning just kept proving it right.