By Jon W. Hansen, FCIPS | Procurement Insights · November 2026
Last week, a reader named Daniel Oliveira left a comment on my post about the AI bubble that made my cash strand concrete.
Amazon’s quarterly filing shows $113.8 billion of operating cash flow for the first nine months of 2026, against $172.2 billion of net capital spending. That is a gap of $58.4 billion. In the third quarter alone, Amazon spent $75.9 billion on property and equipment while its operations brought in $42.4 billion. On the same call, Andy Jassy said Amazon still expects about $220 billion of capital spending this year, and “meaningfully more” in 2027. AWS grew 39%.
Demand is real. So is the cash it takes to serve it. They are separate strands, moving at different speeds.
But when I read the full filing, I found something that took me back twelve years.
The question I asked in 2014
In September 2014, I asked whether Amazon was the new Oracle. At the time, it was as much whimsy as reasoned argument. Buyers were asking why buying at work couldn’t be as easy as buying at home, and I suggested that for standardized, high-volume purchases, e-procurement was becoming e-commerce.
When I republished that post in October 2023, the question had answered itself in two ways. Amazon Business had become a searchable marketplace inside eProcurement platforms. And AWS had overtaken Oracle in the cloud. Amazon was no longer just a place to buy things. It had become infrastructure.
In June 2025, I noticed something else. A coffee order placed on a Saturday morning arrived the same day. Amazon’s network works on the same logic I used in 1998 for the Department of National Defence: inventory placed at strategic stocking locations, with a real-time polling mechanism choosing the location best suited to each delivery. When our multimodel panel compared the two, all four models found strong conceptual alignment, and none found evidence that Amazon drew on my work. Amazon built its own version, at a scale I never could. That is how it should be stated.
So by 2025, Amazon was three strands in many organizations: a buying channel, a cloud, and a logistics network.
The 2026 filing adds a fourth.
The strand nobody is tracing
Amazon is now financing part of the demand that fills its own cloud.
Unless noted, the figures below are from Amazon’s third-quarter 10-Q, filed in October 2026:
- OpenAI. AWS’s existing $38 billion commitment with OpenAI was expanded by $100 billion over eight years. In 2026, Amazon also completed a $50 billion investment in OpenAI’s preferred stock.
- Anthropic. AWS’s commitment with Anthropic was expanded by more than $100 billion over ten years. Amazon invested $10 billion in Anthropic this year, and up to $15 billion more is available under a financing facility that becomes drawable as AWS reaches milestones in delivering compute capacity.
- The backlog. AWS’s contracted but not-yet-recognized revenue reached about $578 billion at 30 September, up from about $200 billion a year earlier (third-quarter 2025 filing). It is contracted, not guaranteed: when and whether it becomes revenue depends on customers’ usage and their ability to pay.
To pay for the build-out, Amazon’s long-term debt rose from $65.6 billion at the end of 2025 to $173.1 billion at 30 September. It also arranged a new secured facility of up to $80 billion, available from early 2027, borrowed against the technology infrastructure it buys.
Put plainly: Amazon is borrowing to build capacity, investing in the labs that commit to use that capacity, and borrowing against the equipment that delivers it.
These are different instruments: equity stakes in the labs, the labs’ contractual commitments to AWS, a financing facility tied to compute delivery, and Amazon’s own borrowing. None is a simple loan to a customer, as Nortel’s were. But their combined effect is similar: part of the demand filling Amazon’s cloud depends on capital Amazon itself supplied.
None of this is illegal or hidden. It is in the filings. And none of it means Amazon is in trouble. Its operating income rose 61% in the third quarter, its core businesses are strong, and it holds $114.2 billion in cash and marketable securities.
What it means is that two strands that are supposed to be independent, demand and financing, are now partly the same strand.
There is a third strand moving with them. Amazon cut about 30,000 corporate jobs between October 2025 and January 2026.
It is not alone. In October, Microsoft offered voluntary exits to about 60,000 employees, roughly a quarter of its workforce, while protecting its data-center and sales teams. Microsoft expects fewer than 5% to accept.
These companies are not cutting people because demand fell. In effect, they are shifting spending from people to equipment. That is a different pattern from 2001, when layoffs followed collapsing revenue. But it means some of the people leaving may carry knowledge no system records.
We have seen this pattern before
In 2000 and 2001, the telecom equipment makers Nortel and Lucent did not collapse because people stopped using networks. Usage kept growing. They collapsed in part because they had lent money to the carriers buying their equipment. Vendor financing made demand look stronger than it was. When the carriers could no longer raise capital, the orders stopped, the loans went bad, and the equipment makers were hit from both sides at once.
I remember that period well. On 18 January 2001, I was interviewed on Rob Snow’s program on 580 CFRA in Ottawa. I had just sold my company for $12 million. Nortel’s chief financial officer was on the same show that day, the day Nortel reported its year-end results. He told listeners the future was very bright. In its results statement that day, the company said it saw “continuing strong market demand.” Four weeks later, Nortel warned of a loss and 10,000 job cuts, and its stock fell by a third.
My recording of Rob Snow’s program on 580 CFRA, Ottawa, 18 January 2001: the same broadcast on which Nortel’s chief financial officer appeared. Source The Procurement Insights Physical Archives.
The analysts who saw it coming did not rely on what executives said. They asked Nortel’s customers what they were buying. An executive’s reassurance is not a strand. What customers buy, what cash does, and what a company does with its people are.
The comparison has limits, and they matter. In 2000, the buyers were young, debt-funded carriers. Today, the main buyers of AI capacity include some of the strongest companies in the world, and Amazon is not Nortel. But the AI labs at the center of these commitments are a different matter. Their ability to pay over eight or ten years depends on their own growth and on continued access to capital. Their revenue does not appear in Amazon’s filings.
That is the crossing to watch: not Amazon’s cash gap on its own, but the point where the labs’ ability to pay meets the financing that made their commitments possible.
Why this matters to procurement
Most organizations manage Amazon as a supplier, or as several separate suppliers managed by different teams. Procurement manages Amazon Business. IT manages AWS. Logistics manages delivery. The AI team manages the models it runs on AWS.
That is the map. Here is the trace.
When one company is your buying channel, your cloud, your logistics network and the host of your AI models, and is also financing the labs whose models you use, a change at that company does not reach you through one strand. It reaches you through several at the same time.
I am not saying something bad is coming. I am saying that most organizations cannot currently see how much of their operation runs through a single company, because each piece of that exposure sits in a different department’s contract. The decisions that matter are already arriving: committed-spend agreements, cloud renewals, AI model choices, and contract terms written for a market where the supplier is under growing pressure to show a return on more than $200 billion a year.
In practice, several strands moving at once might look like this: a cloud price increase, a committed-spend minimum you can no longer use, and a change in an AI model’s terms, all arriving in the same renewal cycle.
For anyone tracing these strands, this is a grey swan. The strands are visible. The timing is not.
What to do now
Before your next AWS renewal or committed-spend agreement, map your Amazon exposure across every strand: buying, cloud, logistics and AI. Ask who owns each piece, what each contract commits you to, and what you would do if pricing, capacity or terms changed in two or three of them at once.
That is Phase 0™, the pre-commitment mapping discipline I have argued for since 1998, applied to a single supplier. It is the same discipline I have argued for since 1998: understand how your operation actually runs before you commit capital to it.
In 2014, I asked whether Amazon was the new Oracle. In 2026, the better question is whether you know how many of your strands now run through it.
Truth Is Believing. Accuracy Is Knowing. Outcome Is Proof.™
-30-
References
- Is Amazon The New Oracle or How The Emergence of The B2C World Marks The Official End of The ERP Era — September 12, 2014
- AMAZON: Is this the future of procurement? — July 31, 2023
- Is Amazon The New Oracle (Originally Published Sept. 2014) — October 20, 2023
- It is time to send Jeff Bezos an invoice! — June 7, 2025
- Bezos? Jeff Bezos? What does he have to do with anything related to procurement and supply chain? — June 7, 2025
- The AI Bubble Has Not Burst in One Explosion. The Strands Show How It Is Unfolding. — October 30, 2026
- Amazon.com, Inc. Form 10-Q for the quarter ended September 30, 2026 (SEC filing)
- Amazon.com, Inc. Form 10-Q for the quarter ended June 30, 2026 (SEC filing)
- 580 CFRA, Rob Snow program, Ottawa, January 18, 2001. Cassette recording, The Procurement Insights Physical Archives.
- Nortel sees 1Q loss, to cut 10,000 jobs — CNN Money, February 15, 2001
- The story behind Nortel’s fall — The Globe and Mail
- Amazon.com Announces Third Quarter Results
Related
Amazon Is No Longer One Supplier. It Is Several of Your Strands, And Why You Should Start Paying Attention.
Posted on November 1, 2026
0
By Jon W. Hansen, FCIPS | Procurement Insights · November 2026
Last week, a reader named Daniel Oliveira left a comment on my post about the AI bubble that made my cash strand concrete.
Amazon’s quarterly filing shows $113.8 billion of operating cash flow for the first nine months of 2026, against $172.2 billion of net capital spending. That is a gap of $58.4 billion. In the third quarter alone, Amazon spent $75.9 billion on property and equipment while its operations brought in $42.4 billion. On the same call, Andy Jassy said Amazon still expects about $220 billion of capital spending this year, and “meaningfully more” in 2027. AWS grew 39%.
Demand is real. So is the cash it takes to serve it. They are separate strands, moving at different speeds.
But when I read the full filing, I found something that took me back twelve years.
The question I asked in 2014
In September 2014, I asked whether Amazon was the new Oracle. At the time, it was as much whimsy as reasoned argument. Buyers were asking why buying at work couldn’t be as easy as buying at home, and I suggested that for standardized, high-volume purchases, e-procurement was becoming e-commerce.
When I republished that post in October 2023, the question had answered itself in two ways. Amazon Business had become a searchable marketplace inside eProcurement platforms. And AWS had overtaken Oracle in the cloud. Amazon was no longer just a place to buy things. It had become infrastructure.
In June 2025, I noticed something else. A coffee order placed on a Saturday morning arrived the same day. Amazon’s network works on the same logic I used in 1998 for the Department of National Defence: inventory placed at strategic stocking locations, with a real-time polling mechanism choosing the location best suited to each delivery. When our multimodel panel compared the two, all four models found strong conceptual alignment, and none found evidence that Amazon drew on my work. Amazon built its own version, at a scale I never could. That is how it should be stated.
So by 2025, Amazon was three strands in many organizations: a buying channel, a cloud, and a logistics network.
The 2026 filing adds a fourth.
The strand nobody is tracing
Amazon is now financing part of the demand that fills its own cloud.
Unless noted, the figures below are from Amazon’s third-quarter 10-Q, filed in October 2026:
To pay for the build-out, Amazon’s long-term debt rose from $65.6 billion at the end of 2025 to $173.1 billion at 30 September. It also arranged a new secured facility of up to $80 billion, available from early 2027, borrowed against the technology infrastructure it buys.
Put plainly: Amazon is borrowing to build capacity, investing in the labs that commit to use that capacity, and borrowing against the equipment that delivers it.
These are different instruments: equity stakes in the labs, the labs’ contractual commitments to AWS, a financing facility tied to compute delivery, and Amazon’s own borrowing. None is a simple loan to a customer, as Nortel’s were. But their combined effect is similar: part of the demand filling Amazon’s cloud depends on capital Amazon itself supplied.
None of this is illegal or hidden. It is in the filings. And none of it means Amazon is in trouble. Its operating income rose 61% in the third quarter, its core businesses are strong, and it holds $114.2 billion in cash and marketable securities.
What it means is that two strands that are supposed to be independent, demand and financing, are now partly the same strand.
There is a third strand moving with them. Amazon cut about 30,000 corporate jobs between October 2025 and January 2026.
It is not alone. In October, Microsoft offered voluntary exits to about 60,000 employees, roughly a quarter of its workforce, while protecting its data-center and sales teams. Microsoft expects fewer than 5% to accept.
These companies are not cutting people because demand fell. In effect, they are shifting spending from people to equipment. That is a different pattern from 2001, when layoffs followed collapsing revenue. But it means some of the people leaving may carry knowledge no system records.
We have seen this pattern before
In 2000 and 2001, the telecom equipment makers Nortel and Lucent did not collapse because people stopped using networks. Usage kept growing. They collapsed in part because they had lent money to the carriers buying their equipment. Vendor financing made demand look stronger than it was. When the carriers could no longer raise capital, the orders stopped, the loans went bad, and the equipment makers were hit from both sides at once.
I remember that period well. On 18 January 2001, I was interviewed on Rob Snow’s program on 580 CFRA in Ottawa. I had just sold my company for $12 million. Nortel’s chief financial officer was on the same show that day, the day Nortel reported its year-end results. He told listeners the future was very bright. In its results statement that day, the company said it saw “continuing strong market demand.” Four weeks later, Nortel warned of a loss and 10,000 job cuts, and its stock fell by a third.
My recording of Rob Snow’s program on 580 CFRA, Ottawa, 18 January 2001: the same broadcast on which Nortel’s chief financial officer appeared. Source The Procurement Insights Physical Archives.
The analysts who saw it coming did not rely on what executives said. They asked Nortel’s customers what they were buying. An executive’s reassurance is not a strand. What customers buy, what cash does, and what a company does with its people are.
The comparison has limits, and they matter. In 2000, the buyers were young, debt-funded carriers. Today, the main buyers of AI capacity include some of the strongest companies in the world, and Amazon is not Nortel. But the AI labs at the center of these commitments are a different matter. Their ability to pay over eight or ten years depends on their own growth and on continued access to capital. Their revenue does not appear in Amazon’s filings.
That is the crossing to watch: not Amazon’s cash gap on its own, but the point where the labs’ ability to pay meets the financing that made their commitments possible.
Why this matters to procurement
Most organizations manage Amazon as a supplier, or as several separate suppliers managed by different teams. Procurement manages Amazon Business. IT manages AWS. Logistics manages delivery. The AI team manages the models it runs on AWS.
That is the map. Here is the trace.
When one company is your buying channel, your cloud, your logistics network and the host of your AI models, and is also financing the labs whose models you use, a change at that company does not reach you through one strand. It reaches you through several at the same time.
I am not saying something bad is coming. I am saying that most organizations cannot currently see how much of their operation runs through a single company, because each piece of that exposure sits in a different department’s contract. The decisions that matter are already arriving: committed-spend agreements, cloud renewals, AI model choices, and contract terms written for a market where the supplier is under growing pressure to show a return on more than $200 billion a year.
In practice, several strands moving at once might look like this: a cloud price increase, a committed-spend minimum you can no longer use, and a change in an AI model’s terms, all arriving in the same renewal cycle.
For anyone tracing these strands, this is a grey swan. The strands are visible. The timing is not.
What to do now
Before your next AWS renewal or committed-spend agreement, map your Amazon exposure across every strand: buying, cloud, logistics and AI. Ask who owns each piece, what each contract commits you to, and what you would do if pricing, capacity or terms changed in two or three of them at once.
That is Phase 0™, the pre-commitment mapping discipline I have argued for since 1998, applied to a single supplier. It is the same discipline I have argued for since 1998: understand how your operation actually runs before you commit capital to it.
In 2014, I asked whether Amazon was the new Oracle. In 2026, the better question is whether you know how many of your strands now run through it.
Truth Is Believing. Accuracy Is Knowing. Outcome Is Proof.™
-30-
References
Share this:
Like this:
Related