The Great Un-Defaulting: AI Is Coming Home, and Print Should Too

Why the cloud-repatriation movement reshaping AI infrastructure in 2026 is the same reckoning print management has been avoiding — and what it means for data sovereignty, security, and control.

Every era of technology has a default. An answer so reflexive that people stop noticing it is a choice. For the last decade, that default was “put it in the cloud.” It was rarely a decision. It was an assumption, a posture, a thing you said in a meeting to sound modern. And like all defaults, it quietly made a thousand architecture decisions on your behalf that you never actually made.

In 2026, that default is breaking. The most advanced organizations in the world are pulling their most demanding workloads: AI training, inference, sensitive data and putting them back onto physical infrastructure that they own and control. Call it cloud repatriation if you want the analyst term. I’d call it something more interesting: the great un-defaulting. The moment a generation of buyers stopped letting the industry’s reflex pick their architecture and started picking it themselves.

This piece is about why that is happening, why it is not a step backward, and why the same reckoning is overdue in the least glamorous corner of the stack: print.

The default always overstays its welcome

Defaults are useful right up until they aren’t. They let you move fast without thinking, which is exactly the problem. They keep working long after the conditions that justified them have changed, because nobody is paying attention to a decision nobody remembers making.

Cloud-first earned its default status honestly. For variable, experimental, globally distributed workloads, it was, and is, genuinely the right call. But “right for some workloads” silently became “right for all workloads,” and the bill came due, literally. By the time enterprises noticed, AI-heavy cloud spend was climbing fast and only a minority of finance leaders could point to clear ROI.¹ The default had stopped serving the business and started taxing it.

So the smart money un-defaulted. Not by swinging to the opposite extreme with the notion that “cloud is bad, on-prem is good” but by doing the thing defaults exist to prevent: thinking. Matching each workload to the deployment that actually fits it. The emerging best practice in AI is explicitly hybrid: sensitive, steady, high-volume work on owned infrastructure; bursty, low-sensitivity, experimental work in the cloud.² That is not a side in a war. It is the refusal to fight the war on the war’s terms.

What un-defaulting looks like

You can see the un-defaulting in the questions that now open every serious infrastructure evaluation. A regulated enterprise assessing AI no longer asks “is it cloud-native?” It asks:³

• Where does our data physically reside?
• What certifications does it carry?
• What is our exit path?
• Can we audit it?
• Can we inspect it?

These are not cloud questions or on prem questions. They are control questions. And control is precisely what defaults erode because a default, by definition, is control you handed to someone else without noticing. Regulators have now codified this: DORA, the EU AI Act, GDPR, and HIPAA do not ask whether your vendor is secure. They ask whether you are in control.⁴ The un-defaulting isn’t a trend. It is becoming the law.

Printing: the most un-examined default of all

Now consider printing: the workload no one ever consciously architects.

The traditional print server is a perfect fossil of default thinking. Nobody designed it. It accreted: a box, then drivers, then scripts, then GPOs, then a decade of “don’t touch it, it works.” It became load-bearing by neglect. And it is, by any honest measure, bad. It is brittle, costly, insecure, the single most common source of helpdesk pain.⁵ So far, so deserving of the wrecking ball.

But here is where printing made its great mistake. When the industry finally turned on the print server, it did not un-default. It re-defaulted. “Eliminate your print servers” became the new reflex — go serverless, go cloud-native, do it because everyone is doing it. Buyers fled one unexamined architecture straight into another.

And the new default turned out to carry its own sins. The loudest “eliminate your print servers” platform was found to contain 83 documented vulnerabilities — unauthenticated remote code execution, cross-tenant takeover where one customer’s breach could reach another’s data, hardcoded keys — across a four-year disclosure saga, with several findings the vendor classified as “feature requests” rather than fixing.⁶ The researcher opened his report by quoting the vendor’s own marketing promise to secure your environment by eliminating print servers. The irony writes itself: the cure for one default became a new default with new wounds.

Print didn’t need a new default. It needed to un-default.

The same reasons, not the same hardware

Be careful with the parallel, though, because the technology doesn’t actually rhyme. AI is repatriating partly because the hardware is exotic with racks of cutting-edge GPUs, and a model you run yourself is a genuinely different animal from one you rent from a frontier lab. Nobody is hauling a Windows print server home for the silicon. A print server is the opposite of cutting-edge: it’s the workload time forgot, and the on-prem box and the cloud one run nearly the same decades-old technology. The only thing that changes between them is who holds the keys. So forget the hardware. The analogy that actually holds is about motive — and when you list why organizations are pulling AI back home, the same drivers map onto print almost one for one:

• Control of sensitive data. AI moved home so training data and prompts stay inside the perimeter. Print jobs are PHI, financials, and signed records — the same case for keeping them on your own turf.
• Predictable cost. Repatriation tamed runaway, metered cloud bills. A serverless print contract is the same open meter, billed per device and seat instead of per token.
• Owning the risk. Enterprises pulled AI back to stop renting their risk from a black box. Outsourcing print to a single cloud vendor hands that vendor your blast radius (see the 83 vulnerabilities above).

The data backs the AI side cleanly: 91% of enterprises would keep sensitive AI data on-prem, private, or hybrid, and 93% are repatriating or evaluating it.⁷ The vendors who can prove control are winning; the ones who can only point to an SLA are losing.⁸

Same drivers, same destination: infrastructure you designed, secured, and optimized on purpose. For printing that points to a clear, un-defaulted architecture: print data that stays inside your fence line, every job authenticated to a real identity, hold-and-release at the device, and an immutable audit trail that satisfies HIPAA, SOX, and GDPR.⁹ Not “keep the old server.” Not “trust the new cloud.” Design the thing.

Three doors, one of them on purpose

In practice, every print environment walks through one of three doors. Two of them get chosen by default. Only the third is designed.

• The traditional print server. The fossil: a box that accreted drivers, scripts, and GPOs until it became load-bearing by neglect. It still works, which is exactly why nobody secures it: an unsegmented, over-credentialed machine with a line of sight to your file shares and identity systems, and the single most common source of helpdesk pain. Choosing it isn’t a decision so much as a failure to make one.
• Cloud-only / serverless. The rebound default: “eliminate the print server” taken on faith. It hands your print data, and your blast radius, to a vendor whose track record includes 83 documented vulnerabilities. You trade a server you control for a dependency you don’t.
• Intentional print architecture. The designed option. Printing is treated as critical infrastructure: always on, with designed failover so a single point never takes printing down. You consolidate the printing infrastructure by up to 99%, then secure and optimize what’s left: data inside your perimeter, jobs released only to an authenticated identity at the device. Less infrastructure, by intention, and every piece of it deliberate.

That third door is the whole point. Intentional design isn’t “on-prem” or “cloud” — it’s control: a single, deliberate way to govern how every page is printed, who can release it, and how you prove it later. Done right, it shrinks the print attack surface dramatically while making print boring again, the way infrastructure you actually designed is supposed to be.

The point was never the server

The cloud-versus-server debate was always a distraction, in AI and in print alike. The server was never the villain, and the cloud was never the hero. The villain was the default, the decision you let the industry make for you while you were busy.

AI infrastructure is un-defaulting in public, at the scale of billions of dollars and the urgency of regulators. Print can do it quietly, on its own terms, before the next breach forces the issue. Either way the principle is identical, and it is the only principle that has ever actually mattered: stop inheriting your infrastructure. Start designing it.

AI is coming home. Printing should too. Not because home is where servers live, but because home is where intention does.