Suppliers are giving customers stock that only vests if the customer keeps buying. History suggests the schedule mostly follows demand the customer already had, and when the arrangement disappoints, the supplier is often the one that gives something up.
In August and September 2026, Marvell gave Google a warrant on 58,970,907 shares and Qualcomm gave Amazon a warrant on 25,000,000 shares. In both cases the shares only vest as the customer buys. Marvell's vest in 240 tranches, one for every $500 million of revenue. Qualcomm's vest against purchases running up to $60 billion.
At first glance that looks like a supplier buying a customer's commitment. The historical record points somewhere narrower.
The structure is not new. Plug Power, Symbotic, ARRIS, Marqeta, Credo and Astera Labs have all run a version of it since 2016, and one of them ran the same instrument twice with two different buyers in the same year. Watching what those buyers actually did, rather than what the announcements said, gives a fairly clear picture of what this instrument buys and what it does not.
In 2017 Plug Power issued warrants to Amazon in April and to Walmart in July. The instruments were essentially identical. Both covered up to 55,286,696 shares. Both vested against payments of up to $600 million. Both moved in $50 million increments.
By the third quarter of 2020 Amazon had nearly completed the payments required for the second tranche.
Walmart's second tranche did not reach final vesting until 30 October 2023.
The incentive was held constant. The supplier was the same, the schedule was the same, the increments were the same. Only the buyer changed, and the buyers differed by a factor of two in how fast they got there.
What differed was each buyer's own programme. Amazon was electrifying warehouse fleets at pace. Walmart was moving more slowly through the same category.
Same incentive, radically different purchasing speeds. Whatever the equity did, the buyer's own demand mattered more.
Walmart's remaining warrant units vested in May 2022 and it exercised for $104.0 million. Symbotic's revenue then grew 26 percent in fiscal 2025 and its backlog reached roughly $22.5 billion, three years after the last unit vested. What holds that relationship is 42 integrated distribution centre deployments, a Walmart-funded development programme, and a business line Symbotic bought from Walmart in January 2025. Not the warrant.
Symbotic annual filingsCredo issued Amazon a warrant against payments of up to $201.0 million in December 2021. Fourteen months later, 80,000 of 4,080,000 shares had vested. Amazon then became a very large share of Credo's revenue during fiscal 2025. The timing is more consistent with the later demand cycle than with the warrant itself.
Credo quarterly filings. The fiscal 2025 share is analyst reporting, not a filed figureTwo warrants against tranches of payments, with 1,165,513 shares vested by December 2025 and none exercised. Vesting is still running, so this one cannot yet be tested for what happens after the equity stops being earnable.
Astera Labs quarterly filingsMarqeta had been running Block's Cash App and Square Debit Card programmes since April 2016. Block was already the dominant customer five years before the warrant existed. Roughly 1.3 million of the roughly 1.9 million issued shares had vested when both windows closed in 2025, so about a third of the programme lapsed.
Marqeta annual and quarterly filingsARRIS granted Comcast a warrant in June 2016 for up to 8 million shares and Charter a warrant in September 2016 for up to 6 million, both vesting in annual tranches tied to purchases. Vesting above a floor in each year required increased purchases and a set share of those purchases in one ARRIS segment.
Roughly 2.2 million of 14 million shares vested, all of them on 2016 purchases. Nothing vested in 2017 or 2018. Over the same stretch Comcast's purchases from ARRIS fell 32.0 percent from the 2016 level and Charter's fell 12.8 percent. The warrants were terminated and cashed out in April 2019 when CommScope acquired the company.
Two things to hold in mind when reading those numbers. ARRIS bought Pace in 2016 and its total sales jumped 42 percent that year, so the rise in 2016 buying is not evidence the warrant worked. And ARRIS never said how the 2.2 million shares split between the two buyers, so we know the total earned but not who earned it.
In the ARRIS case, an incentive around one percent of associated purchases was not enough to overcome the buyers' declining purchasing cycle.
Across the historical examples, the vesting schedule mostly recorded what the buyer's own programme was already doing.
The equity vested and the arrangement looked effective. Amazon was electrifying warehouses. Walmart was automating distribution. Block was issuing cards at scale.
We cannot tell whether the equity added some extra purchasing on top of that.
The equity went unearned regardless of how much was on the table. Comcast and Charter walked away from more than four fifths of the shares available to them.
Credo makes the same point from the other direction. The same warrant sat almost entirely unvested for three years, then vested quickly when an unrelated demand shock arrived.
The two clearest priced cases suggest size affects whether the incentive is worth taking. ARRIS's incentive was about 1 percent of purchases and mostly went unearned. Plug's was roughly 210 percent of associated purchases and was fully earned after Plug waived the remaining conditions. Neither establishes that the incentive created purchasing that would not otherwise have happened.
The evidence suggests the buyer's own programme matters more than the equity incentive. We cannot tell whether the equity added some extra purchasing. We can tell that it did not override the buyer's underlying demand cycle.
Its board authorized management to negotiate acceleration with Amazon, recording that it would cause a substantial charge. On 31 December 2020 Plug waived all remaining vesting conditions. The warrant charge for the year was $420.0 million and Plug reported negative consolidated revenue of $93.2 million. Twenty months later it issued Amazon a second and larger warrant, on 16,000,000 shares against $2.1 billion of spend.
It renewed with Block on terms that allowed reduced pricing and made Block responsible for managing the Cash App programme with respect to the primary card network. Marqeta's net revenue fell 25 percent the following year, primarily driven by that amendment.
It bought a business line from its own customer and signed a new master automation agreement.
It terminated the warrants inside its own sale.
Across these historical examples, when the economics disappointed, the supplier was the side that changed the deal. In none of them did the buyer give anything back.
That does not mean the equity caused the supplier's weaker position. In several cases the supplier was already highly dependent on the customer before the equity was issued.
The customer equity does not appear to create demand on its own. But there is another problem with the historical record: many of these suppliers were already dependent on the customer before they offered the equity.
Symbotic was already heavily tied to Walmart. Marqeta had been running Cash App for years before Block received a warrant. Plug was already concentrated in a small number of customers.
That means the equity may be less a cause of supplier dependence than a sign that the supplier needed the customer badly enough to offer it in the first place.
The useful conclusion is narrower.
Equity appears better at pricing demand a customer already controls than creating demand that was not already there.
If this customer's own programme paused tomorrow, would the incentive change their spending?
When this arrangement disappoints, who is expected to move?
In none of these examples could we establish that the incentive changed what the buyer otherwise would have purchased. On the second question, it was the supplier every time.
The historical examples mostly involved very large buyers dealing with smaller or highly concentrated suppliers. Qualcomm and Marvell are different. They are much larger suppliers with more bargaining power.
That gives us a useful test.
If the old pattern was mostly about weak suppliers rather than the equity itself, Qualcomm and Marvell should stay much less dependent on Amazon and Google.
One difference is worth noting now. Amazon and Google also have to pay real money to exercise these warrants, at $161.26 and $206.58 a share. Most of the historical examples were priced at a token amount or a deep discount. That makes these two less generous than what came before.
The historical record does not show equity creating demand on its own. It shows suppliers attaching equity to demand the customer already controls. The next test is whether that changes when the supplier has more bargaining power.
This section explains how the work was tested for readers who want to go deeper. Nothing in it is needed to use the findings above.
Empirical verdict PLAUSIBLE BUT UNPROVEN, on the registered primary mechanism of reverse dependence. Portfolio contribution NEW EVIDENCE, downgraded after completion from NEW MECHANISM because attribution succeeded in only one independent case of four.
The study examined four independent adoptions of supplier equity vesting on a buyer's own commercial activity: Amazon, Walmart, the ARRIS cable programme covering Comcast and Charter, and the Marqeta customer programme. Independence was defined as an independent commercial adoption of the mechanism, so buyers receiving substantially the same instrument from one supplier during one contracting programme count as one case with internal replications.
The cohort was closed before the fieldwork ran, and ARRIS was designated the counterexample in advance. Instruments vesting on time, on deployment milestones, or on conditions that a supplier share price could independently block were excluded. Primary sources were preferred throughout: filings and exhibits, warrant and commercial agreements, investor relations material, and court records.
Verdict thresholds were fixed before coding and applied mechanically to the coded evidence before any narrative was written.
Registered before the fieldwork ran, and not forced to be mutually exclusive where the evidence supported more than one.
| Mechanism | Amazon | Walmart | ARRIS | Marqeta |
|---|---|---|---|---|
| Incremental demand | UNRESOLVED | UNRESOLVED | NO | UNRESOLVED |
| Expensive rebate | YES | UNRESOLVED | YES | UNRESOLVED |
| Temporary lock | YES | NO | NO | UNRESOLVED |
| Operational lock-in | NO | YES | NO | YES |
| Reverse dependence | YES | YES on measures, attribution UNRESOLVED | NO | YES on measures, attribution UNRESOLVED |
Scroll the table sideways to see all four cases.
The attribution rule required that timing and contractual structure connect a dependence shift to the instrument. Concentration on its own was ruled out as evidence before the work began, which is what moved Walmart and Marqeta to unresolved.
| Amazon | Walmart | ARRIS cable | Marqeta | |
|---|---|---|---|---|
| Buyer importance before issuance | Large. Two customers at 66.8 percent combined in 2018 | Symbotic already 66.9 percent Walmart | Comcast 21 percent, Charter 20 percent | Block dominant since 2016 |
| Incentive cost | About 210 percent of associated purchases | Not comparable. Booked as share-based expense | 1.12 percent of associated purchases | Under 1 percent of associated revenue |
| Activity during vesting | Rose. $200m in 3.4 years | Rose. $200m in 6.3 years on identical terms | Fell in both threshold years | Unresolved at buyer level |
| Percentage vested | 100 percent, by waiver | 100 percent Symbotic. About 68 percent Plug by 2024 | 15.7 percent | About 68 percent of the programme |
| Post-incentive persistence | Fell, roughly half | Rose | Not applicable | Unresolved |
| Concentration, before to after | 40.8 percent peak, then 13.0 percent | 66.9 percent, peak 94.4, then 84 plus | 24.0 and 15.6, then 16.5 and 13.8 | 68 percent, then 45 percent |
| Customer-specific investment | Hydrogen supply commitment to 2029 | 42 distribution centre deployments, funded development programme | None disclosed | Cash App programme |
| Concessions | Waiver, then a second larger warrant | Acquired a business line from the buyer | None | Reduced pricing, network control transferred |
| Dependence direction | Toward the buyer | Toward the buyer | Away from the buyer | Toward the buyer |
| Attribution | Connects | UNRESOLVED | Not applicable | UNRESOLVED |
Scroll the table sideways to see all four cases.
Several arrangements commonly described as customer warrants do not qualify under the inclusion rules and are not used as evidence anywhere in this study. Atlas Air's main warrant to Amazon vests on the commencement of aircraft operations, which is a deployment milestone rather than a purchase. Air Transport Services Group's later warrants were fully vested when issued, which is time vesting, and ATSG booked them as a lease incentive rather than as consideration payable to a customer. Clean Energy Fuels and SpartanNash could not be re-verified from primary sources inside the bounded scan and are coded UNRESOLVED.
Purchase-linked instruments were coded in three subtypes: 1A vesting on dollars purchased or revenue generated, 1B vesting on transaction or unit volume, and 1C vesting on the share of the buyer's business directed to the supplier. Subtype 1C can evidence allocation but not aggregate incremental demand, and is never used that way.
ARRIS was designated the counterexample before the fieldwork ran, so that a negative result could not be treated as a footnote.
Incremental purchasing could not be established in any case. No counterfactual exists for any of these relationships, so growth during vesting is never called incremental.
In three of the four cases the supplier was already concentrated on the buyer before any equity was issued. That makes it impossible to separate the instrument from bargaining asymmetry that already existed, and it is why the attribution rule rejected those cases.
Threshold behavior could not be measured anywhere. No case discloses dated purchasing against dated hurdles, so no claim is made in either direction about buyers bunching purchases at vesting thresholds.
Marqeta's later disclosures aggregate its two warrant programmes and round to the nearest 0.1 million, so buyer-level vesting outcomes inside that case are unresolved and are never inferred from programme-level numbers.
Realized value to the buyer could not be priced consistently. Exercise dates and prices are disclosed in one case and not the others, so the accounting cost of the incentive and the buyer's eventual gain are reported separately and never combined.
The disclosed cohort begins around 2016. Nothing usable survives from earlier, so this is not an estimate of how often the instrument has been used historically.
Every mature case pairs a dominant buyer with a small or acutely concentrated supplier, so nothing here can be generalized to relationships with flatter bargaining power.
One instrument in the cohort vests on the share of a buyer's business directed to the supplier rather than on volume. That can evidence allocation. It cannot evidence aggregate incremental demand, and is not used that way.