By Jon W. Hansen, FCIPS | Procurement Insights · October 2026
In August 2024, Ulf Venne of Everstream Analytics suggested that AI probably would not crash and burn as sensationally as the dot-com bubble. I agreed, and added that this time the impact would be concentrated on procurement and supply chains, which touch every area of our lives.
It would likely be death by a thousand cuts: not the single burst of a balloon, but a much slower leak with a more significant hangover period.
A month earlier, I had listed the conditions that felt eerily similar to those before the dot-com bubble burst:
Too many solution providers in the market.
Venture capital money being thrown, somewhat blindly, at anything AI.
A resulting shift in focus from serving end-client needs to achieving quarterly revenue targets.
The proliferation of “professional managers” with little experience or expertise in the industries their solutions were designed to support, heading up companies.
And the big one: Gartner’s report that 80% of all AI initiatives were failing.
On July 9, 2024, I asked why Goldman Sachs’ head of global equities appeared to share my concern about an overhyped AI market.
On August 23, 2024, I went further and asked whether the eventual correction would be a big bang or a thousand cuts.
Today, financial results from Microsoft, Caterpillar, ServiceNow, Alphabet and Amazon have brought that question back into focus.
The evidence does not yet establish a broad collapse in AI demand. It shows something more significant for the longer term: the strands supporting continued AI expansion are beginning to move at different speeds.
That is how a thousand cuts begins.
The signals are contradictory only when viewed separately
Consider what the companies are reporting.
A conventional analysis may try to decide which interpretation is correct.
Is AI demand growing, or is the bubble bursting?
The answer can be both because these are different strands.
Demand can grow while capital economics deteriorate. AI consumption can increase while enterprise buyers struggle to demonstrate operating returns. Infrastructure providers can report record backlogs while inventory accumulates. Hyperscalers can produce rapidly growing cloud revenue while consuming more cash than the market is willing to tolerate indefinitely.
The outcome will be decided where those strands cross.
What Strand Commonality™ sees
Strand Commonality™ holds that seemingly unrelated strands have unique operating attributes that collectively contribute to a single outcome.
The methodology does not take one negative number and declare a collapse. It also does not take one positive number and declare the investment justified.
It follows the relationships among:
capital deployment,
infrastructure delivery,
physical capacity,
energy and water availability,
enterprise adoption,
actual usage,
monetization,
client outcomes,
cash conversion,
financing conditions,
provider viability,
and client exposure.
Each strand can appear healthy when assessed inside its own boundary.
The problem, or the opportunity, emerges at the crossings.
Microsoft’s Azure growth does not cancel its infrastructure constraints. Caterpillar’s inventory does not cancel its record backlog. ServiceNow’s growing AI consumption does not prove that every enterprise adopting AI is producing an adequate return. Negative free cash flow does not mean demand has disappeared.
The meaningful question is whether adoption is producing enough measurable value, quickly enough, to support the capital, infrastructure and provider ecosystem built around it.
That relationship can be expressed as a continuing operating loop:
Capital investment → infrastructure capacity → enterprise adoption → operating outcomes → monetization → cash conversion → renewed capital support
The loop remains viable when each stage supports the next.
It begins to weaken when:
capital investment grows faster than usable capacity;
capacity grows faster than enterprise adoption;
adoption grows faster than measurable operating value;
revenue grows without adequate cash conversion;
financing requirements increase as capital becomes more expensive;
or provider survival depends on new funding arriving before client outcomes can be demonstrated.
Those are not isolated financial ratios. They are operating strands.
A banker once told me that when a small business increases its sales but its profit percentage declines to achieve that growth, it is a red flag. The same test now applies to the largest companies in the world, one step further down: not only whether margins hold as revenue grows, but whether the cash does.
ProcureTech has already demonstrated the pattern
We do not have to wait for the entire AI market to correct before understanding what the correction may look like.
ProcureTech has already supplied a measurable case.
In February 2025, James Meads challenged my figure that 75% of ProcureTech solution-provider logos would disappear. The debate at the time framed it as a prediction. What it reflected were strands already visible in the provider market, the same ones I had listed in July 2024. James expected net consolidation of no more than 20% to 25%. Even if half of the existing companies disappeared, he argued, new entrants would replace many of them.
It was a fair challenge because the word disappear was open to interpretation.
We therefore replaced the disputed word with four observable outcome states and applied them to the baseline solution map:
Defunct: shut down or insolvent, with no continuity. 22%
Absorbed: acquired, brand discontinued, roadmap subordinated to the acquirer. 31%
Merged or renamed: no longer sold independently; product identity uncertain. 19%
Still operating independently: same product, same roadmap, same risk profile. 28%
The measured effective-disappearance rate was 72%.
The figure was 75%.
As the February post noted, the many solution maps overlap only partly, so a different baseline map could move the 72% up or down. The point was never the exact number. It was the volatility of the provider market, and its direct impact on practitioner outcomes.
The important finding was not that every provider failed or that the technology category disappeared.
The category survived.
Many of the logos did not.
Some shut down. Some were acquired, and their brands discontinued, with their roadmaps subordinated to the acquirer. Some merged or changed identity. Only 28% continued operating independently with the same product, roadmap and risk profile.
That is what a thousand cuts looks like.
There was no single morning when 72% of the market vanished. The correction accumulated through individual transactions, funding failures, acquisitions, mergers, discontinued brands and lost independence.
A separate review of the “10 for 2010” ProcureTech providers produced an even more severe result. Fifteen years later, only Ivalua remained a thriving independent provider. That represented a 90% effective-disappearance rate across a complete technology cycle.
This matters because the AI-provider correction does not require AI itself to fail.
It requires the capital, adoption, monetization and outcome strands to become sufficiently misaligned that many individual providers can no longer sustain an independent business.
From the strands identified in 2024 to the 2026 evidence
The record now forms a trackable sequence.
2024: The developing conditions
The initial posts identified too many providers, venture capital thrown at anything AI, a shift from client needs to quarterly revenue targets, inexperienced leadership and high AI failure rates, alongside Bloomberg’s warning that AI would cost a lot to build out and might not pay off.
Those strands pointed to a correction that would be gradual and distributed, without the single turning point people remember from the dot-com peak.
2025: The provider consequence comes into view
The discussion moved from the general AI bubble to the likely survival rate of AI-agent companies. In November 2025, I endorsed Max Henry’s prediction that more than 90% of AI agent providers in logistics would disappear by 2026, and argued that the survivors would be those that assessed readiness before deployment. Its stated horizon, “by 2026,” has not yet been measured. It did not mean that intelligent agents would disappear.
It meant that a market containing too many providers, weak differentiation, uncertain client outcomes and dependence on continuing capital would eventually consolidate.
February 2026: The historical mechanism was measured
The ProcureTech Vendor Survival Outcomes study showed that 72% of the providers in the baseline market map no longer existed as independent entities.
That measurement demonstrated how a technology category can continue growing while most of its original provider identities disappear.
October 2026: The pressure moves upstream
Microsoft, Alphabet, Amazon, Caterpillar and ServiceNow now expose different parts of the AI investment system:
the hyperscalers financing capacity;
the physical suppliers supporting construction;
the software platforms monetizing enterprise adoption;
the clients expected to convert AI usage into business value;
and the capital markets deciding how long deferred returns remain acceptable.
No single result proves that the AI bubble has burst.
Collectively, they show that the burden of proof has changed.
The market is beginning to ask whether growing usage can become measurable value, whether measurable value can become cash, and whether that cash can justify the capital already committed.
Those are the crossings that matter.
Invariant Physics™ is why this applies beyond ProcureTech
Strand Commonality™ is not confined to procurement, AI or technology markets.
The industries change. The technologies change. The participants change. The invariant relationship does not.
Every operating outcome continues to depend on the interaction among agents and strands with different:
objectives,
incentives,
authority,
timing,
resources,
constraints,
information,
and definitions of success.
I first encountered this relationship in 1983 when changing one line of dBase II code unexpectedly broke another line that had been working correctly.
It became measurable in the 1998 Department of National Defence MRO operation. The visible problem appeared to be slow procurement. The decisive strands included technician incentives, four-o’clock order batching, suppliers with no cross-border customs experience, courier cutoff times, customs clearance and next-day delivery requirements.
The same underlying physics appears in the ProcureTech provider market and in today’s AI infrastructure economy.
That is Invariant Physics™: the technology and operating environment can change while the underlying determinants of the outcome remain constant.
AI does not replace Strand Commonality™. It accelerates the identification, integration and recalculation of strands in real time.
The foundational theory remains the same.
Strand Commonality™ does not predict events. It discovers the strands already in motion. A black swan is rarely a true surprise. More often, it is the crossing of strands nobody had traced. The same crossing is a black swan to someone who never saw the strands, a grey swan to someone who saw them but not the timing, and a white swan to someone who traced them. Discovery does not always reveal when a crossing will occur, but it removes the surprise of what is coming.
The strands to watch next
The question is no longer whether AI will continue growing. It will.
The question is whether the strands supporting that growth will remain aligned closely enough to produce sustainable outcomes.
Here are the signals that should now be tracked together.
1. Capital commitment versus available capacity
Watch announced capital expenditure against actual data center completion, third-party delivery delays, equipment availability, grid connections and operational start dates.
Money committed is not capacity available.
2. Capacity growth versus paid usage
Track cloud and model capacity against actual enterprise workloads, token consumption and utilization rates.
Infrastructure can be fully constructed without being economically productive.
3. Usage versus measurable client outcomes
Growing consumption should be connected to operating measures such as revenue, cost, cycle time, quality, service delivery, risk reduction and working capital.
Usage is not an outcome.
4. Revenue growth versus cash conversion
Subscription revenue, AI contract value and cloud growth should be tracked against capital intensity, free cash flow, debt issuance and continuing financing requirements.
Revenue can grow while the underlying economics weaken.
5. Provider funding versus provider independence
Monitor shutdowns, distressed acquisitions, mergers, renamed products, discontinued roadmaps and the subordination of acquired technology.
The technology may remain available while the original provider and its commitments disappear.
6. Model capability versus operating readiness
More capable models can increase the speed and reach of an implementation. They cannot establish that the organization understands the operation into which those models are being deployed.
Greater agency accelerates both success and error.
7. Automation savings versus displaced operating knowledge
When people are removed, organizations should determine which undocumented relationships, exceptions, interventions and workarounds disappear with them.
A lower labor cost can produce a larger operating loss when the human agent was carrying an unrecognized strand.
8. Provider survival versus client exposure
Provider consolidation should be connected to implementation continuity, contractual obligations, data portability, integration maintenance, product support and roadmap control.
Vendor survivability is not simply a selection issue. It is an enterprise-readiness variable.
What the evidence now supports
The AI bubble did not suddenly burst because several companies published one quarter of results.
The more consequential development is that the strands identified in 2024 can now be followed through a measurable historical provider outcome and into the current AI capital cycle.
ProcureTech demonstrated the mechanism:
the category continued;
the technology advanced;
client demand remained;
and 72% of the original provider logos ceased to exist independently.
That is the pattern to watch in AI.
The strongest providers and technologies will survive. Some will be acquired. Some will merge. Some will be renamed. Some will disappear. Many clients will discover that choosing a capable product did not protect them from changes in ownership, roadmap, funding or commercial viability.
The correction does not have to destroy AI to prove that the bubble existed.
It only has to expose the distance between the capital invested and the operating outcomes produced.
In 2024, I called that prospect a thousand cuts.
In 2026, the individual cuts are becoming trackable strands.
Strand Commonality™ tells us how to follow them. Invariant Physics™ explains why the same method works across technologies, businesses and industries. Phase 0™ establishes whether an organization can absorb and sustain value before it commits significant capital.
That is the work: identifying the hidden strands and aligning them with a business’s operating reality.
The technology will survive.
Most of today’s logos may not.
Truth Is Believing. Accuracy Is Knowing. Outcome Is Proof.™
The AI Bubble Has Not Burst in One Explosion. The Strands Show How It Is Unfolding.
Posted on October 30, 2026
0
By Jon W. Hansen, FCIPS | Procurement Insights · October 2026
In August 2024, Ulf Venne of Everstream Analytics suggested that AI probably would not crash and burn as sensationally as the dot-com bubble. I agreed, and added that this time the impact would be concentrated on procurement and supply chains, which touch every area of our lives.
It would likely be death by a thousand cuts: not the single burst of a balloon, but a much slower leak with a more significant hangover period.
A month earlier, I had listed the conditions that felt eerily similar to those before the dot-com bubble burst:
On July 9, 2024, I asked why Goldman Sachs’ head of global equities appeared to share my concern about an overhyped AI market.
On August 23, 2024, I went further and asked whether the eventual correction would be a big bang or a thousand cuts.
Today, financial results from Microsoft, Caterpillar, ServiceNow, Alphabet and Amazon have brought that question back into focus.
The evidence does not yet establish a broad collapse in AI demand. It shows something more significant for the longer term: the strands supporting continued AI expansion are beginning to move at different speeds.
That is how a thousand cuts begins.
The signals are contradictory only when viewed separately
Consider what the companies are reporting.
A conventional analysis may try to decide which interpretation is correct.
Is AI demand growing, or is the bubble bursting?
The answer can be both because these are different strands.
Demand can grow while capital economics deteriorate. AI consumption can increase while enterprise buyers struggle to demonstrate operating returns. Infrastructure providers can report record backlogs while inventory accumulates. Hyperscalers can produce rapidly growing cloud revenue while consuming more cash than the market is willing to tolerate indefinitely.
The outcome will be decided where those strands cross.
What Strand Commonality™ sees
Strand Commonality™ holds that seemingly unrelated strands have unique operating attributes that collectively contribute to a single outcome.
The methodology does not take one negative number and declare a collapse. It also does not take one positive number and declare the investment justified.
It follows the relationships among:
Each strand can appear healthy when assessed inside its own boundary.
The problem, or the opportunity, emerges at the crossings.
Microsoft’s Azure growth does not cancel its infrastructure constraints. Caterpillar’s inventory does not cancel its record backlog. ServiceNow’s growing AI consumption does not prove that every enterprise adopting AI is producing an adequate return. Negative free cash flow does not mean demand has disappeared.
The meaningful question is whether adoption is producing enough measurable value, quickly enough, to support the capital, infrastructure and provider ecosystem built around it.
That relationship can be expressed as a continuing operating loop:
Capital investment → infrastructure capacity → enterprise adoption → operating outcomes → monetization → cash conversion → renewed capital support
The loop remains viable when each stage supports the next.
It begins to weaken when:
Those are not isolated financial ratios. They are operating strands.
A banker once told me that when a small business increases its sales but its profit percentage declines to achieve that growth, it is a red flag. The same test now applies to the largest companies in the world, one step further down: not only whether margins hold as revenue grows, but whether the cash does.
ProcureTech has already demonstrated the pattern
We do not have to wait for the entire AI market to correct before understanding what the correction may look like.
ProcureTech has already supplied a measurable case.
In February 2025, James Meads challenged my figure that 75% of ProcureTech solution-provider logos would disappear. The debate at the time framed it as a prediction. What it reflected were strands already visible in the provider market, the same ones I had listed in July 2024. James expected net consolidation of no more than 20% to 25%. Even if half of the existing companies disappeared, he argued, new entrants would replace many of them.
It was a fair challenge because the word disappear was open to interpretation.
We therefore replaced the disputed word with four observable outcome states and applied them to the baseline solution map:
The measured effective-disappearance rate was 72%.
The figure was 75%.
As the February post noted, the many solution maps overlap only partly, so a different baseline map could move the 72% up or down. The point was never the exact number. It was the volatility of the provider market, and its direct impact on practitioner outcomes.
The important finding was not that every provider failed or that the technology category disappeared.
The category survived.
Many of the logos did not.
Some shut down. Some were acquired, and their brands discontinued, with their roadmaps subordinated to the acquirer. Some merged or changed identity. Only 28% continued operating independently with the same product, roadmap and risk profile.
That is what a thousand cuts looks like.
There was no single morning when 72% of the market vanished. The correction accumulated through individual transactions, funding failures, acquisitions, mergers, discontinued brands and lost independence.
A separate review of the “10 for 2010” ProcureTech providers produced an even more severe result. Fifteen years later, only Ivalua remained a thriving independent provider. That represented a 90% effective-disappearance rate across a complete technology cycle.
This matters because the AI-provider correction does not require AI itself to fail.
It requires the capital, adoption, monetization and outcome strands to become sufficiently misaligned that many individual providers can no longer sustain an independent business.
From the strands identified in 2024 to the 2026 evidence
The record now forms a trackable sequence.
2024: The developing conditions
The initial posts identified too many providers, venture capital thrown at anything AI, a shift from client needs to quarterly revenue targets, inexperienced leadership and high AI failure rates, alongside Bloomberg’s warning that AI would cost a lot to build out and might not pay off.
Those strands pointed to a correction that would be gradual and distributed, without the single turning point people remember from the dot-com peak.
2025: The provider consequence comes into view
The discussion moved from the general AI bubble to the likely survival rate of AI-agent companies. In November 2025, I endorsed Max Henry’s prediction that more than 90% of AI agent providers in logistics would disappear by 2026, and argued that the survivors would be those that assessed readiness before deployment. Its stated horizon, “by 2026,” has not yet been measured. It did not mean that intelligent agents would disappear.
It meant that a market containing too many providers, weak differentiation, uncertain client outcomes and dependence on continuing capital would eventually consolidate.
February 2026: The historical mechanism was measured
The ProcureTech Vendor Survival Outcomes study showed that 72% of the providers in the baseline market map no longer existed as independent entities.
That measurement demonstrated how a technology category can continue growing while most of its original provider identities disappear.
October 2026: The pressure moves upstream
Microsoft, Alphabet, Amazon, Caterpillar and ServiceNow now expose different parts of the AI investment system:
No single result proves that the AI bubble has burst.
Collectively, they show that the burden of proof has changed.
The market is beginning to ask whether growing usage can become measurable value, whether measurable value can become cash, and whether that cash can justify the capital already committed.
Those are the crossings that matter.
Invariant Physics™ is why this applies beyond ProcureTech
Strand Commonality™ is not confined to procurement, AI or technology markets.
The industries change. The technologies change. The participants change. The invariant relationship does not.
Every operating outcome continues to depend on the interaction among agents and strands with different:
I first encountered this relationship in 1983 when changing one line of dBase II code unexpectedly broke another line that had been working correctly.
It became measurable in the 1998 Department of National Defence MRO operation. The visible problem appeared to be slow procurement. The decisive strands included technician incentives, four-o’clock order batching, suppliers with no cross-border customs experience, courier cutoff times, customs clearance and next-day delivery requirements.
The same underlying physics appears in the ProcureTech provider market and in today’s AI infrastructure economy.
That is Invariant Physics™: the technology and operating environment can change while the underlying determinants of the outcome remain constant.
AI does not replace Strand Commonality™. It accelerates the identification, integration and recalculation of strands in real time.
The foundational theory remains the same.
Strand Commonality™ does not predict events. It discovers the strands already in motion. A black swan is rarely a true surprise. More often, it is the crossing of strands nobody had traced. The same crossing is a black swan to someone who never saw the strands, a grey swan to someone who saw them but not the timing, and a white swan to someone who traced them. Discovery does not always reveal when a crossing will occur, but it removes the surprise of what is coming.
The strands to watch next
The question is no longer whether AI will continue growing. It will.
The question is whether the strands supporting that growth will remain aligned closely enough to produce sustainable outcomes.
Here are the signals that should now be tracked together.
1. Capital commitment versus available capacity
Watch announced capital expenditure against actual data center completion, third-party delivery delays, equipment availability, grid connections and operational start dates.
Money committed is not capacity available.
2. Capacity growth versus paid usage
Track cloud and model capacity against actual enterprise workloads, token consumption and utilization rates.
Infrastructure can be fully constructed without being economically productive.
3. Usage versus measurable client outcomes
Growing consumption should be connected to operating measures such as revenue, cost, cycle time, quality, service delivery, risk reduction and working capital.
Usage is not an outcome.
4. Revenue growth versus cash conversion
Subscription revenue, AI contract value and cloud growth should be tracked against capital intensity, free cash flow, debt issuance and continuing financing requirements.
Revenue can grow while the underlying economics weaken.
5. Provider funding versus provider independence
Monitor shutdowns, distressed acquisitions, mergers, renamed products, discontinued roadmaps and the subordination of acquired technology.
The technology may remain available while the original provider and its commitments disappear.
6. Model capability versus operating readiness
More capable models can increase the speed and reach of an implementation. They cannot establish that the organization understands the operation into which those models are being deployed.
Greater agency accelerates both success and error.
7. Automation savings versus displaced operating knowledge
When people are removed, organizations should determine which undocumented relationships, exceptions, interventions and workarounds disappear with them.
A lower labor cost can produce a larger operating loss when the human agent was carrying an unrecognized strand.
8. Provider survival versus client exposure
Provider consolidation should be connected to implementation continuity, contractual obligations, data portability, integration maintenance, product support and roadmap control.
Vendor survivability is not simply a selection issue. It is an enterprise-readiness variable.
What the evidence now supports
The AI bubble did not suddenly burst because several companies published one quarter of results.
The more consequential development is that the strands identified in 2024 can now be followed through a measurable historical provider outcome and into the current AI capital cycle.
ProcureTech demonstrated the mechanism:
That is the pattern to watch in AI.
The strongest providers and technologies will survive. Some will be acquired. Some will merge. Some will be renamed. Some will disappear. Many clients will discover that choosing a capable product did not protect them from changes in ownership, roadmap, funding or commercial viability.
The correction does not have to destroy AI to prove that the bubble existed.
It only has to expose the distance between the capital invested and the operating outcomes produced.
In 2024, I called that prospect a thousand cuts.
In 2026, the individual cuts are becoming trackable strands.
Strand Commonality™ tells us how to follow them. Invariant Physics™ explains why the same method works across technologies, businesses and industries. Phase 0™ establishes whether an organization can absorb and sustain value before it commits significant capital.
That is the work: identifying the hidden strands and aligning them with a business’s operating reality.
The technology will survive.
Most of today’s logos may not.
Truth Is Believing. Accuracy Is Knowing. Outcome Is Proof.™
-30-
References
Share this:
Like this:
Related