You own the data. They own the door.

Evaluate the commitments behind the price: required skills, operating ownership, customer knowledge and the ability to reduce consumption. Separate support economics from licenses, hosting and project costs.

The views below are my own, drawn from public court records and public filings. The vendors would frame all of this differently, and honestly, I'd love to watch them try.

A federal court just entertained an idea that ought to stop every CIO cold: that the pipe to your own data is a market.

Not the software. Not the license. The pipe. The mere act of reaching data you already own, packaged up and sold back to you as its own product, by the vendor conveniently sitting on the valve. For thirty years "the market" in enterprise software meant the applications. A judge in San Francisco has now agreed to consider that the real market might be the plumbing, and that whoever owns the plumbing owns you.

Here is how it got in front of that judge. In March 2025, a process-mining company called Celonis sued SAP. Not over a bug, not over a botched upgrade, not over the usual enterprise-software grievances that get settled with a credit memo and an apology nobody means. Celonis sued because it says SAP made it hard and expensive for customers to get at their own data. Read that twice. Their own data.

SAP got the first version of the complaint thrown out in July 2025. Celonis rewrote it, and in October 2025 a federal judge looked at the amended claim, declined to make it disappear a second time, and let the heart of it march straight into discovery, accepting along the way the startling premise above: that access to your own ERP data can be a market unto itself, one a vendor is entirely capable of cornering.

That is the first crack in a wall most buyers have not noticed is being built around them. Worth understanding now, while the mortar is still wet.

What the fight is actually about

Peel away the process-mining jargon and the plot is almost insultingly simple. Celonis makes software that reads data out of an ERP and shows a company how its processes really run, as opposed to how the slide deck said they would. To do that, it has to pull data out of SAP.

Celonis alleges that after SAP bought a rival tool called Signavio in 2021, SAP got remarkably less enthusiastic about letting anyone else reach that data, and it picked the forced march to S/4HANA cloud as the moment to start tightening the bolts. The claim is not that SAP refused to sell Celonis something. It is that SAP stood between customers and data those customers already own, then quoted them a price to visit it.

Two details tell you this is about control, not courtroom theater. First, in June 2025 SAP signed a standstill agreeing not to interfere with the customer data extractor and not to charge extra for it while the case runs. SAP will say the standstill concedes nothing, and legally maybe so. But you do not sign a truce to stop doing a thing you were never doing.

Second, and this is the phrase to circle, the judge let the "data access aftermarket" theory survive. There it is again: a court willing to treat the pipe to your data as its own market, priced and gated separately from the software it flows out of. The landlord, it turns out, may charge you to visit your own furniture, and the court would like to hear more.

As Marcus Harris, an enterprise-software attorney at Taft Stettinius & Hollister, put it to CIO : "Data is the currency of the modern world. With the advent and wide-scale adoption of AI, ERP software vendors are in a unique position to regulate and control access to that data. To the extent they can charge fees for access to that data, they will." He wasn't picking a side. He was reading the room.

This is the same heist, one floor down

In the bond piece I made the case that SaaS quietly repossesses the software. You go from owning a perpetual license to renting access, and every exit you used to have gets welded shut while a banner overhead calls it modernization. The Celonis case is the floor beneath that one. SaaS does not just take back the software. It takes back your unmediated line to the data. And once AI shows up, the line to the data is close to the only thing left worth guarding.

And here is the hinge the rest of this turns on, so hold onto it: you can run the exact same ERP on-premise or in SaaS, and that one deployment choice decides who holds the key to your data. On-prem, the door is yours. In SaaS, it is theirs. The entire fight is about quietly walking you from the first house into the second and calling the walk modernization.

And once you see that, the whole cloud campaign reads differently. For years the industry explained the great SaaS push with the polite reasons: recurring revenue, predictable margins, easier to maintain, better for everyone. All true, and all beside the point now.

The reason the migration has turned into a forced march, complete with end-of-support deadlines and incentives dressed as generosity, is that AI just made the data the single most valuable position in enterprise software, and the vendor that controls the door to it controls the customer. The SaaS push was always partly about the money. It is now mostly about the data. Modernization is the cover story. Control is the objective.

For thirty years, the ERP earned its keep in the application layer: the screens, the workflows, the business logic, the reports nobody read but everybody paid for. That is precisely the layer AI vaporizes first.

An agent can run a process without caring whose ugly interface it clicks through, and a model can generate the report without a human ever opening the form. When the app stops being the thing you can't live without, the vendor slides down to whatever is underneath it, which is the system of record. The tables. So the moat has to relocate from the app to the data, and it has to move before the app moat finishes evaporating like a puddle in July.

I am not out on a limb predicting this. The vendors are narrating it themselves, into a microphone, on purpose.

The incumbent read the plan into the record

On Oracle's December 10, 2025 earnings call, Larry Ellison did what Larry Ellison does, which is skip the coyness. Training AI on public data, he said, is the biggest, fastest-growing business in history, and reasoning on private data will be bigger still. Then the line worth taping to your monitor: "Oracle databases contain most of the world's high-value private data."

Hear that as a shareholder and it's a growth story. Hear it as a customer and it's a treasure map with your house circled on it. Ellison is not pointing at the app, the UI, or the workflow engine. He is pointing at the data in the tables and telling Wall Street that Oracle's edge is that it is sitting on the pile. Same fact, two chairs, wildly different feelings about it.

He kept going, describing Oracle's AI data platform as the thing that tears down the walls fragmenting your data so the models can finally reason across all of it. Framed as a keynote, that's liberation, cue the swelling music. Framed honestly, it's the tell. The man who volunteers to knock down all your walls is rarely doing it so you can enjoy the cross-breeze. Whoever gets to demolish the old walls is the one who decides where the new ones go, and "the sole broker through which AI is permitted to touch your data" is a very nice wall to own.

Note what he conceded to get there. On Oracle's March 2026 call he grabbed the industry's own scare phrase and said it flat out: "the SaaS-apocalypse applies to others, but not to us." His reason was that Oracle's new AI coding tools make it the disruptor rather than the disrupted. Mine is simpler: Oracle has the stack, the data, and the customer entrenchment to walk out of the fire unsinged. That is an incumbent admitting out loud that the application layer is in trouble and naming his lifeboat: data plus lock-in.

He is not wrong, and he is not being a villain about it. He is doing exactly what his shareholders hired him to do, which is the same thing I said about the coupon, and it traces straight back to who actually controls the company, the American-versus-German governance split I unpacked in Oracle vs SAP: Why Corporate Law Decides Your ERP Roadmap . The incentives are the whole story.

And here is where those two governance systems, so different everywhere else, finally converge. SAP, the company that same piece credited with actually protecting its product, is the one being sued over data access and probed in Brussels. That is the tell that this enclosure is structural, not cultural.

On-prem you hold the key. In SaaS they do. That is the whole mechanism.

This is the part that never makes the brochure, so let me be blunt about it, because it is the engine under everything above.

Run your ERP on-premise, in your own data center, and this heist is simply impossible. The database sits in your building. You can point an extractor, a query, or any third-party tool you like straight at the tables, and there is nothing the vendor can do about it except raise your maintenance bill and sulk in Walldorf or Austin. The door is open because you own the house it is set into. Data access isn't a permission the vendor grants you. It's a fact of where the software lives.

Now move that same ERP into SaaS, into the vendor's cloud, and the house is theirs. This is the quiet flip, and it is total. The vendor runs the entire stack, which means there is no path to your data that does not pass through an interface the vendor built, controls, meters, prices, and can deprecate on a Tuesday afternoon.

That is what turns the cloud migration from an IT project into a transfer of control. The move your board approved as modernization is the move that carries your data out of a house where you hold the keys and into one where the vendor works reception. And the moment it lands there, the meter can start.

The new coupon isn't the old 22 percent maintenance line from the on-prem world. It's the toll on reaching your own data: extraction fees, egress charges, API limits, the going rate for letting an outside model so much as glance at rows that were already yours. Same annuity reflex, relocated to the one floor AI can't flatten, and it only becomes collectible the day your data moves into their cloud.

So the next time a vendor is unusually, almost touchingly eager to get you off on-prem and into its SaaS, hold that enthusiasm up to the light and ask what it's really buying them. Part of it is a genuinely better product, and I'll grant them that. But the part nobody prints on the slide is this: on-prem, they are merely your landlord. In SaaS, they are your landlord, your locksmith, and the toll operator on the only road out. That is not a fringe benefit of the migration they're selling you. In the AI era, it is the whole reason they're selling it.

To be fair, they have an answer

The vendors are not twirling a mustache here, and their defense deserves a hearing. Direct access to the raw tables, they will tell you, was always a security and integrity risk. A governed, metered, vendor-run interface is genuinely safer, cleaner, and easier to audit than a dozen third-party tools elbowing into your production database at two in the morning. That is not spin. It is largely true.

It is also precisely how every tollbooth in history has been justified. The toll is always sold as the guardrail. The question was never whether the gate has a legitimate purpose, because it always does, that is what makes it defensible. The question is who holds the key and what they charge you to turn it. A guardrail you are not allowed to remove, priced by the company that installed it, on a road you are now required to drive, quietly stops being a safety feature and becomes a business model. Both things can be true at once, and in this case they are.

The honest catch, because there always is one

Now the part a fair person has to say out loud, and it's the reason this is a live brawl and not a finished mugging: it's your data, not the vendor's. That distinction is the entire reason Celonis is an antitrust case instead of a shrug. SAP already had to sign that standstill.

So the honest framing is this: the vendors can't own your data outright, but they are moving fast to own the pace and the price of reaching it, and for the next several years that is close enough to the same thing that your CFO won't feel the difference. The fence is contested. It is also, unmistakably, going up. Both are true, and the fact that they're willing to eat antitrust exposure to build it tells you exactly how badly they want it. Nobody lawyers up for years on end over a wall they figured they could take or leave.

The one lever they can't repossess

In the bond piece I said the last lever you hold is how the system gets operated, because the vendor-fee lever gets welded shut in the SaaS endgame. The data enclosure just sharpens the point to a needle. In a world where reaching your own data is metered, the discipline around it becomes the lever: which agents touch which tables, at what cost, under whose governance, and how much of your value you let get tolled on the way out the door.

Concretely, that is a handful of moves you can still make while you have leverage, which is to say before you sign. Get documented, priced, permanent data-extraction rights written into the contract itself, not left to the mercy of a future support ticket. Keep a live copy of your own data landing somewhere you control, so the vendor's cloud is never the only place it exists.

And treat egress fees and API limits as negotiated line items, not boilerplate you initial on page forty, because the morning after migration your negotiating position is gone and everyone on the other side of the table knows it. None of this stops the enclosure. It just makes sure you kept a key to the door before they finished hanging it.

Beyond that, the lever is not one the vendor will ever help you pull, for the simplest reason in business, which is that every dollar you save on access is a dollar off its shiny new coupon. That job only comes from someone whose economics improve when yours do, someone who operates the layer well, keeps your access defensible, and treats your data as the asset it is instead of the annuity it's quietly becoming for somebody else.

The application was the moat for three decades, and it is dissolving in broad daylight. The fence around your data is the sequel to the bond, and the smart money notices these things before the gate swings shut, not after it clicks.

Discuss your operating priorities →