Sovereignty is the visible strategy. Discipline is the underlying cause. A response to Michael Gale — and to what the MIT study actually found.
Jon W. Hansen · Procurement Insights
Truth Is Believing. Accuracy Is Knowing.
Hand a vintage Fender to Paul McCartney* and you get something the rest of us will never get out of the same guitar. The instrument is necessary. It is also identical in every pair of hands. And it is not the cause of the song.
That distinction is the whole of this piece, and it is the one the current debate about enterprise AI keeps losing. The technology of any era — the internet, SaaS, the cloud, agentic AI — is the Fender. Necessary, roughly the same instrument available to everyone, and not what determines whether the song comes out. What determines that is the condition of the organization holding it: its people, its processes, its decision rights, the real conditions the technology is dropped into. Call that the substrate.
Michael Gale and I agree that AI is not failing. Where we part company is on what the winners prove.
What the MIT study actually found
In 2025, MIT’s Project NANDA published The GenAI Divide, and one number travelled fast: roughly 95% of enterprise generative-AI pilots produced no measurable return on the P&L, against tens of billions of dollars of spend.
That number produced two readings, and both walked past the part that matters.
The pessimist’s reading was AI doesn’t work. But that isn’t what MIT said. MIT was explicit that the cause was not model quality, not talent, not infrastructure. It was what the authors called the learning gap — brittle tools that never adapt to how the organization actually operates, pilots that never integrate into the work. A failure of fit between the technology and the organization, not a failure of the technology.
“It is our position that a true centralization of procurement objectives requires a decentralized architecture that is based on the real world operating attributes of all transactional stakeholders starting at the local or regional level. In other words, your organization gains control of its spend environment by relinquishing centralized functional control in favor of operational efficiencies originating on the front lines. This is the cornerstone of agent-based modeling.” — Jon Hansen, Acres of Diamonds, 2004 (published September 2005)
Gale’s reading was the mirror image: AI works — the laggards just failed to choose sovereignty. More optimistic, and useful for refusing the doom narrative. But it relocates the cause back into a purchase, and in doing so it walks past MIT’s actual finding, which points the other way. The divide MIT found is not in what enterprises bought. It is in whether the organization could absorb what it bought.
MIT located the divide in organizational absorption. Gale located it in sovereignty. I locate it in substrate readiness — and only one of those three is a thing you can buy.
Gale’s claim, at its strongest
His argument deserves to be put at full strength before it is tested. Drawing on a survey of more than 2,000 executives across thirteen countries, Gale reports that about 13% of enterprises are thriving — running far more deployments than the rest, at returns he places several times above their peers. His diagnosis is the sovereignty imperative: the winners made one choice, to own their AI and data platform, and that choice unleashes a flywheel. He names the proof — UniCredit, Union Insurance, Vattenfall, Verizon, Vezeeta, Vodafone España, Voith Hydro, Wells Fargo, among others — and closes with the cleanest version of the position: the 87% are not doomed; they simply need to make the same choice.
Two things I’ll concede right away, because conceding them makes the disagreement precise. The percentages are his survey’s, and I don’t dispute that he found them. And the companies are his — not a list I assembled to suit an argument, but the firms he offers as evidence. That second point matters, because the question I’m about to ask is put to the set he chose, not the set I chose.
The chicken-or-egg question
Here is the question his own list invites. Did sovereignty make these companies winners? Or are they serial absorbers — organizations that have repeatedly shown the discipline to take in each new generation of technology — for whom AI sovereignty is simply the most recent thing that discipline produced?
If it is the latter, sovereignty is not the cause. It is the visible symptom of a capability they already had — and “buy what they bought” is exactly backwards.
To test it, I plotted his named firms across four technology eras — internet, SaaS and cloud, digital and platform, AI and agentic — and asked whether the AI-era winners were also winning in the eras before. The honest version of that chart looks like this:
The Technology-Era Success Index. Solid markers are the only points sourced to Gale’s 2025 placement; hollow markers and dashed lines are working hypothesis; lines begin only where the company existed as a named entity. Illustrative — not a benchmark of any company.
I want to be plain about what this chart can and cannot carry, because the honesty is the point. Only the 2025 endpoint is sourced — that is Gale’s claim, citable to him. Every earlier point is hypothesis, which is why they are drawn hollow and dashed rather than solid. And some lines begin late on purpose: Vezeeta was founded in 2012, Verizon was formed in 2000, Voith Hydro took its present shape around 2000. They cannot carry an internet-era datapoint, because they did not exist to have one. Laid out truthfully, only two of Gale’s eight proof firms — Wells Fargo and Vattenfall — can show up in 1995 at all.
A chart that hides those seams asserts facts it doesn’t have. A chart that shows them is making a smaller, sturdier claim — and it is the claim I actually want to make. The strongest version of this exhibit does not prove that sovereignty causes success. It proves something more interesting and more defensible: the companies cited as AI-era winners were, in the main, already absorbing capability well before AI arrived.
Why “choose sovereignty” repeats a forty-year error
There is a deeper problem with the prescription than the chart. For forty years the major firms each published a model of why transformation succeeds, and once you strip the acronyms they all said the same thing — and it was never the technology. McKinsey’s 7-S put it in alignment. BCG’s DICE and Bain’s RAPID put it in execution and decision rights. Kotter put it in people. Every one of them named the right variable.
And the failure rate never moved. The famous “70% of transformations fail” turns out to be unsubstantiated, but every credible measure since lands in the same grim territory — McKinsey near 69%, a Bain study at 12% fully succeeding, MIT’s 95% of GenAI pilots returning nothing. Four decades, the right diagnosis, no improvement.
The error was not omission. It was sequence. Everyone says “people, process, and technology” — but the word in the middle is and, a conjunction that makes the three co-equal, a balanced trinity to optimize at once. They are not co-equal. People and process are the preconditions for technology. The correct word was never and. It was then.
Gale’s “own your AI and data platform” is that same flattening performed once more. It takes an organizational precondition — the discipline to govern data and align process across the enterprise — and renames it a thing you buy. But look at the winners’ own behavior as he describes it: open, hybrid data estates running across a dozen departments. No company does that unless it has already dismantled the silos and aligned the operating model. The sovereignty is downstream of the discipline. It is what the substrate work looks like once it is done — not the act that does it. Prescribing the purchase to a firm that hasn’t done the work is selling the trophy as if it were the training.
The evidence that actually controls for technology
A pattern across winners can’t prove cause; survivors survive for many reasons. To get at cause you need the same technology dropped into different organizations, producing different outcomes. There is such a case.
The Commonwealth of Virginia’s eVA program and a comparable Canadian public-sector initiative ran on the same procurement platform. Virginia aligned the substrate first — it treated reform as process understanding rather than platform compliance — and built one of the most durable public e-procurement programs on record. The comparison case modeled itself on Virginia’s outcome, imported the same technology, and skipped the substrate work. Same instrument. Opposite song. The variable that differs is not the platform. It is the sequence. This is where the causal claim actually lives — not in a seven-company chart, but in controlled pairs like this one, where the technology is held constant by fact rather than by assertion.
This was in print before the category existed
The usual objection to an argument like this is that it’s hindsight — a story told once the winners are known. The record answers that, because the substrate argument was published, in this exact sequence, long before MIT, before Gale, and before “AI sovereignty” was a phrase.
It begins in 1998, in an engagement partly funded under Canada’s SR&ED research program, in which technology was introduced only after a diagnosis of the spend had surfaced the dependency that actually governed the outcome. The documented result: delivery performance moved from roughly 51% to 97.3% within three months and held for years; the buyers needed to manage the contract fell from twenty-three FTE to three; the organization realized about 23% in sustained savings. The technology was primitive. It worked because it was last, not first.
By 2004, in a paper published the following year, the same idea had taken a form that reads strangely now. Arguing for low-dollar, high-volume spend as the overlooked source of sustainable savings, it concluded:
“It is our position that a true centralization of procurement objectives requires a decentralized architecture that is based on the real world operating attributes of all transactional stakeholders starting at the local or regional level. In other words, your organization gains control of its spend environment by relinquishing centralized functional control in favor of operational efficiencies originating on the front lines. This is the cornerstone of agent-based modeling.” — Acres of Diamonds, 2004 (published September 2005)
Agent-based modeling — written in 2004, two decades before “agentic AI” became the phrase of the moment. And the same paper was funded, under Canada’s SR&ED program, to develop a system it already called RAM. The vocabulary of the 2025 debate was being written, and funded, before the debate existed.
It was in print by 2007, in a column laying out a three-step method in exactly this order — understand the spend, align the process and the people, then introduce technology. And it was argued at length in a 2008 white paper, published through the CATA Alliance, that states the thesis of this piece almost verbatim seventeen years early: that the application ultimately selected has very little to do with the success of any initiative, that success tracks the alignment of technology with how the organization operates, and that failure originates in the absence of stakeholder collaboration before the platform is chosen. The same paper ran the same controlled comparison — Virginia succeeding while a centrally driven federal program struggled on identical software — and catalogued the era’s expensive failures: Hershey, Hewlett-Packard, FoxMeyer, Cadbury. Large, capable, well-resourced organizations, all defeated not by the software but by the sequence.
The HP case in that paper is the one worth pausing on. Hewlett-Packard was not a victim of unfamiliar technology — it was building a practice to rival IBM’s in implementing the very SAP systems it then failed to merge cleanly inside its own walls, at a cost the paper documented at roughly $400 million in lost revenue. An acknowledged expert in the product could not make the product work. The paper drew the only conclusion that mattered:
“…if a high technology company who has extensive experience with the product can’t succeed, what does this say in terms of any organization’s chances for success?” — SAP Procurement for Public Sector (CATA Alliance, 2008)
The expertise was never the variable – the sequence was and is. It wasn’t in 2008, and renaming the technology “AI” in 2025 does not change it.
By 2011, a conference series on “the end of functional silos” was making the point that has now resurfaced as “AI readiness.” The sovereignty conversation of 2025 is, in substance, the silo conversation of 2011 with a new noun.
Why you can’t fix it by trying harder
There is a reason “buy what the winners bought” fails that runs deeper than “you lack their conditions.” The moment an organization adopts the elite’s solution, it quietly swaps the question. It arrived with its own problem; it leaves working on someone else’s, wearing that solution inside a different world where it doesn’t fit. The failure isn’t that the imported answer was executed badly. It’s that the organization stopped solving its own problem the instant it accepted someone else’s answer as the goal.
Which raises the only question an executive actually has about “align the substrate first”: which substrate issues matter? You cannot align what you have not found — and the step the sequence has long assumed without naming is diagnosis. The 1998 result didn’t come from substrate alignment in the abstract; it came from a diagnosis that surfaced a hidden dependency, which then told the team what to align. A diagnosis belongs to one organization’s real conditions, which is precisely why no other organization’s solution can stand in for it. You can buy a platform. You cannot import a diagnosis.
What I’m claiming — and what I’m not
Two claims, and they are not equally strong, so I’ll keep them apart.
The first is a pattern: the firms cited as AI-era winners have track records of absorbing earlier technology eras, so their AI-era success is not, in the main, an AI-era event. Checked against the dated record — and with the younger firms’ lines honestly begun only when they existed — that is a strong, documented finding and a direct answer to the idea that AI-era success is an AI-era phenomenon.
The second is the cause: that organizational discipline, not sovereignty, is what produced the success. This is the more tentative claim, and I’ll state it as tentative. A chart drawn only from winners can’t escape two ceilings — correlation isn’t causation, and a list of survivors can’t show you the disciplined firms that lost or the undisciplined ones that won. So I hold the causal claim as the best-supported reading available rather than a proven one. It rests not on the company chart but on the controlled pairs, where the technology is held constant, and on the foresight record, where the sequence was specified before the outcomes were known. More controlled pairs would strengthen it. More lines on the chart would not.
I’ll add one thing, carefully. As far as I’m aware, no one else has made this specific connection — turning a sovereignty argument’s own named winners against it, and showing the discipline predates the strategy by decades from a contemporaneous record. The adjacent research exists and supports the pattern; MIT Sloan found in early 2026 that the firms seeing the strongest AI gains were the ones already digitally mature before adopting it. But the inversion itself — sovereignty as the manifestation, discipline as the cause — I haven’t seen elsewhere. I state that as “not that I’m aware of,” not as “no one has,” because the same discipline that caps the causal claim should govern the originality claim too.
The bottom line
Gale is right that AI isn’t failing, and right that the winners prove something. He’s wrong about what. They don’t prove that sovereignty is the choice that turns the flywheel. They prove that organizations which had already done the substrate work — silos down, processes aligned, decision rights settled — could absorb one more technology era, and that AI sovereignty is what that pre-existing discipline looks like in 2025. The choice did not create the capability. The capability made the choice possible.
So the 87% are not one purchase away from the flywheel. They are one diagnosis away from learning whether they can absorb anything at all. That’s not a verdict. It’s a starting point — and a more generous, more accurate, and more useful thing to tell them than buy what the winners bought. The winners were already winning. The honest question isn’t how to copy them. It’s whether you’re ready to win your own.
The full argument — the method, the confidence levels stated explicitly, the founding-date sourcing, and the complete record of the multimodel verification exchange that stress-tested this piece — is available as a companion working paper (63 pages). Anyone named here, Michael Gale included, is invited to challenge any attribution; corrections are made in public.
Jon Hansen is the founder of Hansen Models™ and publisher of Procurement Insights, an independent research practice operating on a nineteen-year archive with zero vendor sponsorships.
-30-
* For the musically precise: McCartney’s signature instruments are basses — the Höfner 500/1 “violin bass” and the Rickenbacker 4001S. He did play Fenders over the years (the Esquire and Telecaster among them), but Fender is the brand a general audience pictures, which is why it anchors the metaphor.
Related
The AI Winners Were Winning Long Before AI Arrived
Posted on June 7, 2026
0
Sovereignty is the visible strategy. Discipline is the underlying cause. A response to Michael Gale — and to what the MIT study actually found.
Jon W. Hansen · Procurement Insights
Truth Is Believing. Accuracy Is Knowing.
Hand a vintage Fender to Paul McCartney* and you get something the rest of us will never get out of the same guitar. The instrument is necessary. It is also identical in every pair of hands. And it is not the cause of the song.
That distinction is the whole of this piece, and it is the one the current debate about enterprise AI keeps losing. The technology of any era — the internet, SaaS, the cloud, agentic AI — is the Fender. Necessary, roughly the same instrument available to everyone, and not what determines whether the song comes out. What determines that is the condition of the organization holding it: its people, its processes, its decision rights, the real conditions the technology is dropped into. Call that the substrate.
Michael Gale and I agree that AI is not failing. Where we part company is on what the winners prove.
What the MIT study actually found
In 2025, MIT’s Project NANDA published The GenAI Divide, and one number travelled fast: roughly 95% of enterprise generative-AI pilots produced no measurable return on the P&L, against tens of billions of dollars of spend.
That number produced two readings, and both walked past the part that matters.
The pessimist’s reading was AI doesn’t work. But that isn’t what MIT said. MIT was explicit that the cause was not model quality, not talent, not infrastructure. It was what the authors called the learning gap — brittle tools that never adapt to how the organization actually operates, pilots that never integrate into the work. A failure of fit between the technology and the organization, not a failure of the technology.
“It is our position that a true centralization of procurement objectives requires a decentralized architecture that is based on the real world operating attributes of all transactional stakeholders starting at the local or regional level. In other words, your organization gains control of its spend environment by relinquishing centralized functional control in favor of operational efficiencies originating on the front lines. This is the cornerstone of agent-based modeling.” — Jon Hansen, Acres of Diamonds, 2004 (published September 2005)
Gale’s reading was the mirror image: AI works — the laggards just failed to choose sovereignty. More optimistic, and useful for refusing the doom narrative. But it relocates the cause back into a purchase, and in doing so it walks past MIT’s actual finding, which points the other way. The divide MIT found is not in what enterprises bought. It is in whether the organization could absorb what it bought.
MIT located the divide in organizational absorption. Gale located it in sovereignty. I locate it in substrate readiness — and only one of those three is a thing you can buy.
Gale’s claim, at its strongest
His argument deserves to be put at full strength before it is tested. Drawing on a survey of more than 2,000 executives across thirteen countries, Gale reports that about 13% of enterprises are thriving — running far more deployments than the rest, at returns he places several times above their peers. His diagnosis is the sovereignty imperative: the winners made one choice, to own their AI and data platform, and that choice unleashes a flywheel. He names the proof — UniCredit, Union Insurance, Vattenfall, Verizon, Vezeeta, Vodafone España, Voith Hydro, Wells Fargo, among others — and closes with the cleanest version of the position: the 87% are not doomed; they simply need to make the same choice.
Two things I’ll concede right away, because conceding them makes the disagreement precise. The percentages are his survey’s, and I don’t dispute that he found them. And the companies are his — not a list I assembled to suit an argument, but the firms he offers as evidence. That second point matters, because the question I’m about to ask is put to the set he chose, not the set I chose.
The chicken-or-egg question
Here is the question his own list invites. Did sovereignty make these companies winners? Or are they serial absorbers — organizations that have repeatedly shown the discipline to take in each new generation of technology — for whom AI sovereignty is simply the most recent thing that discipline produced?
If it is the latter, sovereignty is not the cause. It is the visible symptom of a capability they already had — and “buy what they bought” is exactly backwards.
To test it, I plotted his named firms across four technology eras — internet, SaaS and cloud, digital and platform, AI and agentic — and asked whether the AI-era winners were also winning in the eras before. The honest version of that chart looks like this:
The Technology-Era Success Index. Solid markers are the only points sourced to Gale’s 2025 placement; hollow markers and dashed lines are working hypothesis; lines begin only where the company existed as a named entity. Illustrative — not a benchmark of any company.
I want to be plain about what this chart can and cannot carry, because the honesty is the point. Only the 2025 endpoint is sourced — that is Gale’s claim, citable to him. Every earlier point is hypothesis, which is why they are drawn hollow and dashed rather than solid. And some lines begin late on purpose: Vezeeta was founded in 2012, Verizon was formed in 2000, Voith Hydro took its present shape around 2000. They cannot carry an internet-era datapoint, because they did not exist to have one. Laid out truthfully, only two of Gale’s eight proof firms — Wells Fargo and Vattenfall — can show up in 1995 at all.
A chart that hides those seams asserts facts it doesn’t have. A chart that shows them is making a smaller, sturdier claim — and it is the claim I actually want to make. The strongest version of this exhibit does not prove that sovereignty causes success. It proves something more interesting and more defensible: the companies cited as AI-era winners were, in the main, already absorbing capability well before AI arrived.
Why “choose sovereignty” repeats a forty-year error
There is a deeper problem with the prescription than the chart. For forty years the major firms each published a model of why transformation succeeds, and once you strip the acronyms they all said the same thing — and it was never the technology. McKinsey’s 7-S put it in alignment. BCG’s DICE and Bain’s RAPID put it in execution and decision rights. Kotter put it in people. Every one of them named the right variable.
And the failure rate never moved. The famous “70% of transformations fail” turns out to be unsubstantiated, but every credible measure since lands in the same grim territory — McKinsey near 69%, a Bain study at 12% fully succeeding, MIT’s 95% of GenAI pilots returning nothing. Four decades, the right diagnosis, no improvement.
The error was not omission. It was sequence. Everyone says “people, process, and technology” — but the word in the middle is and, a conjunction that makes the three co-equal, a balanced trinity to optimize at once. They are not co-equal. People and process are the preconditions for technology. The correct word was never and. It was then.
Gale’s “own your AI and data platform” is that same flattening performed once more. It takes an organizational precondition — the discipline to govern data and align process across the enterprise — and renames it a thing you buy. But look at the winners’ own behavior as he describes it: open, hybrid data estates running across a dozen departments. No company does that unless it has already dismantled the silos and aligned the operating model. The sovereignty is downstream of the discipline. It is what the substrate work looks like once it is done — not the act that does it. Prescribing the purchase to a firm that hasn’t done the work is selling the trophy as if it were the training.
The evidence that actually controls for technology
A pattern across winners can’t prove cause; survivors survive for many reasons. To get at cause you need the same technology dropped into different organizations, producing different outcomes. There is such a case.
The Commonwealth of Virginia’s eVA program and a comparable Canadian public-sector initiative ran on the same procurement platform. Virginia aligned the substrate first — it treated reform as process understanding rather than platform compliance — and built one of the most durable public e-procurement programs on record. The comparison case modeled itself on Virginia’s outcome, imported the same technology, and skipped the substrate work. Same instrument. Opposite song. The variable that differs is not the platform. It is the sequence. This is where the causal claim actually lives — not in a seven-company chart, but in controlled pairs like this one, where the technology is held constant by fact rather than by assertion.
This was in print before the category existed
The usual objection to an argument like this is that it’s hindsight — a story told once the winners are known. The record answers that, because the substrate argument was published, in this exact sequence, long before MIT, before Gale, and before “AI sovereignty” was a phrase.
It begins in 1998, in an engagement partly funded under Canada’s SR&ED research program, in which technology was introduced only after a diagnosis of the spend had surfaced the dependency that actually governed the outcome. The documented result: delivery performance moved from roughly 51% to 97.3% within three months and held for years; the buyers needed to manage the contract fell from twenty-three FTE to three; the organization realized about 23% in sustained savings. The technology was primitive. It worked because it was last, not first.
By 2004, in a paper published the following year, the same idea had taken a form that reads strangely now. Arguing for low-dollar, high-volume spend as the overlooked source of sustainable savings, it concluded:
Agent-based modeling — written in 2004, two decades before “agentic AI” became the phrase of the moment. And the same paper was funded, under Canada’s SR&ED program, to develop a system it already called RAM. The vocabulary of the 2025 debate was being written, and funded, before the debate existed.
It was in print by 2007, in a column laying out a three-step method in exactly this order — understand the spend, align the process and the people, then introduce technology. And it was argued at length in a 2008 white paper, published through the CATA Alliance, that states the thesis of this piece almost verbatim seventeen years early: that the application ultimately selected has very little to do with the success of any initiative, that success tracks the alignment of technology with how the organization operates, and that failure originates in the absence of stakeholder collaboration before the platform is chosen. The same paper ran the same controlled comparison — Virginia succeeding while a centrally driven federal program struggled on identical software — and catalogued the era’s expensive failures: Hershey, Hewlett-Packard, FoxMeyer, Cadbury. Large, capable, well-resourced organizations, all defeated not by the software but by the sequence.
The HP case in that paper is the one worth pausing on. Hewlett-Packard was not a victim of unfamiliar technology — it was building a practice to rival IBM’s in implementing the very SAP systems it then failed to merge cleanly inside its own walls, at a cost the paper documented at roughly $400 million in lost revenue. An acknowledged expert in the product could not make the product work. The paper drew the only conclusion that mattered:
The expertise was never the variable – the sequence was and is. It wasn’t in 2008, and renaming the technology “AI” in 2025 does not change it.
By 2011, a conference series on “the end of functional silos” was making the point that has now resurfaced as “AI readiness.” The sovereignty conversation of 2025 is, in substance, the silo conversation of 2011 with a new noun.
Why you can’t fix it by trying harder
There is a reason “buy what the winners bought” fails that runs deeper than “you lack their conditions.” The moment an organization adopts the elite’s solution, it quietly swaps the question. It arrived with its own problem; it leaves working on someone else’s, wearing that solution inside a different world where it doesn’t fit. The failure isn’t that the imported answer was executed badly. It’s that the organization stopped solving its own problem the instant it accepted someone else’s answer as the goal.
Which raises the only question an executive actually has about “align the substrate first”: which substrate issues matter? You cannot align what you have not found — and the step the sequence has long assumed without naming is diagnosis. The 1998 result didn’t come from substrate alignment in the abstract; it came from a diagnosis that surfaced a hidden dependency, which then told the team what to align. A diagnosis belongs to one organization’s real conditions, which is precisely why no other organization’s solution can stand in for it. You can buy a platform. You cannot import a diagnosis.
What I’m claiming — and what I’m not
Two claims, and they are not equally strong, so I’ll keep them apart.
The first is a pattern: the firms cited as AI-era winners have track records of absorbing earlier technology eras, so their AI-era success is not, in the main, an AI-era event. Checked against the dated record — and with the younger firms’ lines honestly begun only when they existed — that is a strong, documented finding and a direct answer to the idea that AI-era success is an AI-era phenomenon.
The second is the cause: that organizational discipline, not sovereignty, is what produced the success. This is the more tentative claim, and I’ll state it as tentative. A chart drawn only from winners can’t escape two ceilings — correlation isn’t causation, and a list of survivors can’t show you the disciplined firms that lost or the undisciplined ones that won. So I hold the causal claim as the best-supported reading available rather than a proven one. It rests not on the company chart but on the controlled pairs, where the technology is held constant, and on the foresight record, where the sequence was specified before the outcomes were known. More controlled pairs would strengthen it. More lines on the chart would not.
I’ll add one thing, carefully. As far as I’m aware, no one else has made this specific connection — turning a sovereignty argument’s own named winners against it, and showing the discipline predates the strategy by decades from a contemporaneous record. The adjacent research exists and supports the pattern; MIT Sloan found in early 2026 that the firms seeing the strongest AI gains were the ones already digitally mature before adopting it. But the inversion itself — sovereignty as the manifestation, discipline as the cause — I haven’t seen elsewhere. I state that as “not that I’m aware of,” not as “no one has,” because the same discipline that caps the causal claim should govern the originality claim too.
The bottom line
Gale is right that AI isn’t failing, and right that the winners prove something. He’s wrong about what. They don’t prove that sovereignty is the choice that turns the flywheel. They prove that organizations which had already done the substrate work — silos down, processes aligned, decision rights settled — could absorb one more technology era, and that AI sovereignty is what that pre-existing discipline looks like in 2025. The choice did not create the capability. The capability made the choice possible.
So the 87% are not one purchase away from the flywheel. They are one diagnosis away from learning whether they can absorb anything at all. That’s not a verdict. It’s a starting point — and a more generous, more accurate, and more useful thing to tell them than buy what the winners bought. The winners were already winning. The honest question isn’t how to copy them. It’s whether you’re ready to win your own.
The full argument — the method, the confidence levels stated explicitly, the founding-date sourcing, and the complete record of the multimodel verification exchange that stress-tested this piece — is available as a companion working paper (63 pages). Anyone named here, Michael Gale included, is invited to challenge any attribution; corrections are made in public.
Jon Hansen is the founder of Hansen Models™ and publisher of Procurement Insights, an independent research practice operating on a nineteen-year archive with zero vendor sponsorships.
-30-
* For the musically precise: McCartney’s signature instruments are basses — the Höfner 500/1 “violin bass” and the Rickenbacker 4001S. He did play Fenders over the years (the Esquire and Telecaster among them), but Fender is the brand a general audience pictures, which is why it anchors the metaphor.
Share this:
Like this:
Related