Slow to Adopt, Hard to Displace

By Thorsten Meyer

This is the last piece in my cloud-to-AI series, and it’s the one that resolves a contradiction I’ve been circling for seven articles. In the last piece, I argued that enterprises are painfully slow to adopt AI — 95% of pilots delivering nothing, the internal customer fighting the change, the whole institution built to resist. And yet those same slow, lumbering incumbents turn out to be remarkably hard to displace — the AI-native disruptors keep discovering that the giants they expected to overrun are still standing, still winning deals, still absorbing most of the money.

How can both be true? Slow to adopt and hard to displace? The answer, once you see it, ties the whole series together: they are the same fact wearing two faces. The very inertia that makes an incumbent slow to change is the moat that makes it hard to dislodge. You cannot have one without the other, and mistaking the first for weakness is the most expensive error a disruptor can make.

The slowness is real — and so is the durability

I won’t relitigate the slowness; the last piece did that. Enterprises are genuinely bad at absorbing AI, for reasons that are 80% organizational and deeply human. That part is not in dispute.

What's easy to miss, standing in the wreckage of all those failed pilots, is how durable the slow incumbents remain. The evidence is unambiguous. The platforms absorbing most enterprise AI investment aren't the disruptors — they're the established players. Microsoft Copilot, embedded across Microsoft 365, represents what analysts call the deepest enterprise AI lock-in currently available. Salesforce's Agentforce, ServiceNow, SAP's Joule (40-plus agents shipped, cloud revenue growing 27% year over year) — the incumbents didn't get displaced during the AI transition; they became, in the words of one analysis, the "operational control planes" for enterprise AI. BCG's own read is blunt: in an AI-first world, incumbents have critical structural advantages, and the ones that move in time have "a clear right to win."

The most telling detail of 2026 is that the big enterprise vendors stopped trying to differentiate from each other and converged — all of them shipping the same architecture: agents acting on trusted enterprise data, wrapped in governance. The disruption everyone predicted didn't unseat the systems of record. It got absorbed into them.

Why it's the same coin

Here is the mechanism, and it's the crux of the whole thing. Every source of an incumbent's slowness is also a source of its durability, because they're the same underlying property viewed from different sides.

Switching costs make you slow to adopt a new vendor's AI — and slow for your customers to leave you for one. The data gravity, the compliance lineage, the deep workflow integration I described back in the SaaS piece: those raise the cost of change in both directions. An enterprise that takes two years to adopt AI because ripping open its systems is hard is, for exactly that reason, an enterprise a competitor cannot rip away in two years either.

Being the system of record makes you conservative and slow — and makes you the only place the AI can plausibly live. This is the deep point: AI has to be grounded in trusted, governed, real institutional data before an enterprise can use it at scale. And who holds that data? The incumbent. As one observer put it, for the finance team living inside SAP or the service desk running on Copilot, the AI is just infrastructure — they want inventory forecasts and resolved tickets, not model portability. Vendor capture is the price of seamless integration, and for most enterprises it's a price worth paying. The slowness and the stickiness are the same fact: the incumbent is embedded, and embedded things move slowly and leave slowly.

Trust and distribution make you cautious and slow — and make you the default. Regulated, compliance-heavy buyers are structurally inclined toward established vendors; the bias toward buying from who you already trust is real and it's an asset. The incumbent's conservatism, which looks like a weakness in the adoption race, is the same conservatism its customers are paying for.

The disruptor's category error

Now the mistake, because it's the practical payoff. The AI-native challenger looks at a slow incumbent and reads "vulnerable." Sees the failed pilots, the internal resistance, the two-year timelines, and concludes the giant is ripe for the taking. This is almost always wrong, and it's wrong for a reason that echoes the tech-giants piece: distribution beats invention, and the incumbent's slowness is the visible surface of an invisible moat.

The incumbent doesn't have to be first to AI. It has to be hard to leave while it catches up — and it is. So the disruptor invents the category, proves the demand, and then watches the incumbent bundle a "good enough" version into the billions of seats it already owns, grounded in the data it already holds, sold to the customers who already trust it. The pioneer did the hard work of creating the market; the incumbent, slow but immovable, captures much of the value anyway by being the place the customer was never going to leave. Reading "slow" as "prey" is how disruptors walk confidently into that trap.

The synthesis — what this whole series was about

Step all the way back, because this finale is really the thesis of the entire arc. Across eight pieces I've traced the same shape through cloud, SaaS, energy, fintech, talent, the fallen giants, and the internal customer — and it always resolves the same way, between two wrong extremes.

The disruption-maximalists are wrong: AI does not cleanly sweep away the incumbents, because the incumbents are slow-but-immovable, and the slowness is a moat, not a death sentence. The incumbent-forever camp is also wrong: value genuinely relocates, frontiers genuinely move, and giants genuinely fall when they miss a platform shift. What actually happens is the thing I keep landing on — an oligopoly plus a long tail, a pie that expands rather than a fixed one that gets divided. The incumbents survive, slow and durable, at the center. The disruptors grow enormous on top of them rather than replacing them — Snowflake on AWS, the whole pattern from the very first piece. And the value moves to whoever builds the neutral layer, owns the proprietary workflow data, and stays dense enough to ride the jagged edge.

"Slow to adopt, hard to displace" is the honest final word on all of it. It's why neither the doom nor the hype was ever right. Markets get mispriced in both directions — that was the cloud lesson, the fintech lesson, the energy lesson, the talent lesson — and the truth lives in the uncomfortable middle where an institution can be genuinely bad at change and genuinely impossible to kill at the same time.

The honest bear case — against my own thesis

I'll close the way I've closed every piece: by arguing against myself, because "hard to displace" is not "impossible to displace," and forgetting that is its own trap.

The durability is conditional. BCG's warning sits right next to its bullish case: incumbents who delay too long risk not just erosion but "full displacement." Slow-but-durable only works if you eventually move — and some incumbents won't. Some will be Intel from the giants piece: so committed to protecting the old franchise that they miss the platform shift entirely, and discover too late that the moat became an anchor. The system-of-record advantage holds right up until the definition of the system of record changes underneath it, and then all that embeddedness is embedded in the wrong thing.

So the real claim is narrower and more honest than "incumbents win": the inertia buys them time and optionality, not immortality. Whether they use that time to genuinely reinvent — cannibalizing themselves while cash-rich, as the survivors always did — or merely to defend, decides which incumbents are hard to displace and which merely look it right up until they aren't.

The part that's mine to make

Here's where the whole series lands for me, as a builder and a post-labor economist. The lesson of "slow to adopt, hard to displace" is not to bet on incumbents falling, nor on disruptors winning cleanly. It's to bet on the relocation itself — and to position where the value actually moves, which is the thread through all eight pieces: build the neutral layer on top of whoever wins; own the proprietary workflow data no foundation model can reach; stay small and dense enough to ride the jagged edge; and hold the alignment and governance touchpoints yourself.

Satya Nadella made the point that closes the loop for me: firm-level sovereignty — a company's ability to control and embed its own proprietary knowledge into AI rather than renting it — is becoming one of the defining questions of this moment. That's the whole case for the local-first, sovereign approach I build around, stated by the CEO of the most durable incumbent of them all. Because here's the final turn of the screw: the thing that is truly hard to displace is not a vendor. It's ownership. You cannot be displaced from infrastructure you control, from data that lives on your own machines, from capability you don't rent. The incumbents are hard to displace because their customers are embedded in them. The most durable position of all is to be embedded in nothing but what you own.

That's the series. Cloud taught us the shape; AI is running the same play, faster and with higher stakes; and the winners, as ever, will be the ones who stopped asking who wins the whole thing and started building the layer the whole thing runs on. Slow to adopt, hard to displace — and hardest of all to displace is the builder who owns their own ground.

Analysis and opinion from a builder, founder, and post-labor economist running a local-first inference operation. Figures verified at time of writing against 2026 sources (BCG, Futurum, SAP and Microsoft disclosures, and enterprise-AI landscape analyses) for incumbent AI positioning, Copilot lock-in, SAP cloud growth, and the vendor convergence on governed agentic architectures; the Nadella reference reflects his reported WEF 2026 remarks on firm-level AI sovereignty. Interpretation and the post-labor framing are the author's own. This is analysis, not investment advice. Part 8 of 8 — the finale of the cloud-to-AI series. Point-in-time as of 18 August 2026.