EU buyers now demand portability, and the single-vendor rep loses

The European enterprise buyer changed what they are buying this June, and most AI sales reps have not updated their pitch. They are still selling depth on one model, deep integration, deep lock-in, the standard land-and-expand story. The buyer is now asking the opposite question: if your provider gets cut off tomorrow, what happens to me. The rep who answers “that won’t happen” loses to the rep who answers “here is how you stay running with a different model in a week.” My claim is that portability has become the feature that wins EU enterprise AI deals, and single-vendor depth has flipped from a strength into a liability you have to overcome.

The trigger is in the June reporting and it is specific. When U.S. export controls forced a frontier provider to suspend access to two of its models overnight, the abstract idea of “AI sovereignty” stopped being a conference panel and became an operational fact for European buyers. Reuters reported on June 22 that large firms including Siemens, Renault, Orange, and ChapsVision were already mixing U.S., Chinese, and European models specifically to reduce dependency risk. That is not a policy aspiration. That is named enterprises rewiring their procurement because they watched a provider get pulled and decided they would not be caught single-sourced when it happens again.

What the buyer is actually afraid of

The fear is not that any particular model is bad. The fear is concentration risk on something outside the buyer’s control. A European enterprise that built a critical workflow on one U.S. model just learned that the model’s availability is subject to a government in another country making a decision the buyer has no say in. From the buyer’s seat, that is the same category of risk as building your supply chain through a single port. The component might be excellent. The single point of failure is the problem, and no amount of quality on the component fixes the single point of failure.

This reframes every lock-in mechanic that AI vendors have spent two years building. Deep integration, proprietary formats, model-specific tuning, the whole apparatus designed to make switching costly, was sold internally as defensibility. To a 2026 EU buyer, that apparatus now reads as captured. The deeper you are wired into one model, the more catastrophic the day that model becomes unavailable, and the buyer has just seen that day happen to someone else. The lock-in you built to protect your revenue is the exact thing the buyer’s risk owner is now screening deals to avoid.

The shift is worth stating precisely because it inverts a decade of SaaS strategy. The entire playbook of the last ten years treated switching costs as the asset. You wanted the customer wired in deep enough that leaving was painful, because painful-to-leave meant predictable revenue, and predictable revenue was the whole game. That logic held as long as the thing customers feared most was their own inertia. It breaks the moment the thing customers fear most is being trapped when an external shock hits, because now the same switching cost you engineered as a moat is read by the buyer as a hostage situation. The buyer is no longer asking “how committed am I willing to be.” They are asking “how fast can I get out if I have to,” and every mechanism you built to slow their exit is now a mark against you. The asset became the liability without the vendor changing anything, because the buyer changed what they were afraid of.

I have argued before that in enterprise deals the risk owner outranks the budget owner, and this is that principle playing out at continental scale. The budget owner might still prefer the cheaper single-vendor option. The risk owner, who now has a live precedent for provider-level disruption, will veto it. And in the current European climate, the risk owner is winning these internal arguments, because nobody wants to be the person who signed the single-source deal right before the next disruption.

The asymmetry that drives this is the same one that makes risk owners powerful in any enterprise deal, only sharpened by a fresh, public precedent. The person who pushes for single-sourcing to save money gets a modest, diffuse win if nothing goes wrong, a slightly better price that nobody remembers. The person who pushes for it and then gets caught when the provider is cut off owns a very specific, very visible disaster that everyone remembers. The payoffs are wildly lopsided, and a rational risk owner reads that lopsidedness and concludes that the multi-vendor path is the only defensible one regardless of cost, because the cost of being wrong about single-sourcing is now a thing they have personally watched happen to a peer. Before June, advocating for portability over price required imagination, you had to convince people of a hypothetical. After June, it requires only memory. The disruption did the persuasion that the risk owner used to have to do themselves, which is why these internal arguments are suddenly short.

Sell portability, and mean it

The selling move is to stop pitching depth-on-one-model and start pitching the buyer’s ability to leave. This feels backwards to anyone trained on stickiness, and that is exactly why most reps will not do it, which is why the ones who do will win the deals.

Concretely, the portability story has a few load-bearing pieces, and they have to be real or sophisticated buyers will catch the gap. The first is a provider abstraction layer, the ability to swap the underlying model without rebuilding the application on top of it. Arvid Kahl’s advice to solo builders applies just as hard at enterprise scale: abstract your LLM provider behind a config toggle so you can swap. A vendor who can credibly say “your workflows run on whichever model you choose, and switching is a configuration change, not a rebuild” is selling the thing the EU buyer is now actively shopping for. A vendor whose product only works on one model is selling the thing they are actively avoiding.

The second piece is data portability, which is the part that actually owns the buyer’s long-term commitment. Models will come and go. The buyer’s accumulated data, configurations, and history are the thing they cannot afford to lose, and a vendor who makes that layer portable and provider-independent is offering durability the single-model competitor cannot match. The honest version of stickiness in 2026 is not “you can’t leave.” It is “your data and your workflows survive any model you run them on,” and that is a stickiness the risk owner will approve because it reduces their concentration risk rather than increasing it.

This is the piece reps most often underweight, because it sounds like giving away the moat, when it is actually relocating the moat to higher ground. If your defensibility lives in switching cost, the buyer’s risk owner is your enemy, because their whole job this year is reducing switching cost. If your defensibility lives in being the system of record for the buyer’s accumulated work, the risk owner is your ally, because portable data that survives any model is exactly the resilience they are mandated to buy. You keep the customer not by trapping them but by being the layer their data lives in regardless of which model sits underneath. They stay because leaving would mean abandoning years of accumulated configuration and history, not because you bolted the door. That is a stickiness that survives a provider getting cut off, which is the only kind of stickiness a 2026 EU buyer will accept, and it happens to be a stronger one, because a buyer who could leave and chooses not to is a far more durable customer than one who wants to leave and cannot. Arvid Kahl’s broader argument about data being the only real moat lands here directly: the model is replaceable and the accumulated human data is not, so the vendor who owns the data layer and frees the model layer is holding the part that actually appreciates.

The third piece is the sovereignty option itself: the ability to run on a European or open-weight model, or on-premise, when the buyer’s regulatory or political posture demands it. This is where the single-vendor U.S.-only pitch is weakest, because there is no answer to “what is our European fallback” if the architecture only speaks to one U.S. provider. The vendor with a multi-model story has an answer. The vendor without one is asking the buyer to bet that the geopolitics will stay calm, and the buyer just watched the geopolitics not stay calm.

The named examples in the Reuters reporting make this concrete in a way an abstract sovereignty pitch never could. Siemens, Renault, and Orange are not fringe buyers experimenting at the margin. They are exactly the large, regulated, politically visible European enterprises that the rest of the continent’s procurement teams watch and copy. When ChapsVision and firms of that profile start deliberately mixing U.S., Chinese, and European models to spread their dependency, they are not making a one-off hedge, they are setting a reference behavior that smaller European buyers will cite in their own internal arguments. A risk owner at a mid-size European firm now gets to say “Siemens is multi-sourcing, why aren’t we,” and that sentence ends the debate. The single-vendor rep is no longer arguing against an abstract fear. They are arguing against the visible, named behavior of the most respected industrial buyers on the continent, and that is an argument they lose.

Where I would temper this

I do not want to overstate how clean this is, because multi-model architecture has real costs and the slogan hides them. Supporting many models means you optimize deeply for none of them, and there are genuine workflows where single-model depth produces a meaningfully better result than a lowest-common-denominator abstraction. A buyer who picks portability is often accepting somewhat worse performance per task in exchange for resilience. That tradeoff is real, and a rep who pretends portability is free will get caught the first time the buyer benchmarks the abstracted version against the native one.

The honest pitch acknowledges the tradeoff and argues the resilience is worth it for the buyer’s specific posture, which means you have to actually know whether this buyer is in a posture where resilience trumps peak performance. A research lab pushing the frontier might rationally accept lock-in for the best model. A regulated European bank that cannot tolerate a provider getting cut off will pay the performance tax for portability every time. Selling portability to the lab is as much a mismatch as selling lock-in to the bank. The point is not that portability always wins. The point is that the EU enterprise buyer has, this June, moved decisively into the posture where it does, and the rep who has not noticed is still selling to a 2024 version of that buyer.

The way to handle the tradeoff in the room is to name it before the buyer does, because a buyer who raises the performance gap themselves has caught you hiding it, while a rep who raises it first has demonstrated they understand the engineering honestly. The line is something like: “Running across multiple models means we are not squeezing the last few points of performance out of any single one, and for a workload where peak quality is everything, a single-model competitor will beat us on that specific benchmark. What you get in exchange is that no government decision in another country can take your system offline. For your situation, which of those matters more.” That framing does two things at once. It concedes the real cost, which builds the credibility every subsequent claim rides on, and it forces the buyer to articulate their own posture, which is the thing you needed to know anyway. A regulated buyer who hears that question almost always answers in favor of resilience, because they have already lived the alternative in their imagination since June, and now you have helped them say out loud the thing they were going to screen for regardless. You have not argued them into portability. You have handed them the language to argue themselves into it.

There is also a credibility trap. Portability is about to become a claim everyone makes, the way “secure” and “enterprise-ready” became claims everyone makes, and most of the claims will be thin. A vendor who says “multi-model” but whose switching actually requires a six-week reimplementation is going to get found out, because the EU risk owner is now specifically testing this. They will ask you to demonstrate a model swap, and “it’s on the roadmap” is a losing answer. Build the portability before you sell it, because this is a buyer who has learned to check.

The shift here is structural and I do not think it reverses soon. Once a buyer has been burned by concentration, or watched a peer get burned, single-sourcing becomes a decision they have to actively justify rather than a default they fall into. The vendors who spent two years deepening lock-in are about to spend the next year explaining it away in European deals. The ones who can hand the buyer a credible exit are going to find that offering the door is what gets them invited in.