Amazon's AI advantage may include the terms it can negotiate
By @engtmk · Research as of 8 September 2026
Amazon’s AI advantage may extend beyond the infrastructure it builds to the commercial terms it can negotiate with the companies supplying it.
The new Qualcomm collaboration offers a useful example. Alongside a technology relationship, Amazon has received conditional rights to participate in its supplier’s equity upside. That adds a second potential source of value to a procurement decision.
My interpretation as an Amazon shareholder: this is evidence consistent with purchasing power. It is not yet evidence that Amazon will earn an attractive return on the equipment it buys, or that its shares offer attractive returns at any price.
The investment question is who retains the cash after the chips are delivered, the infrastructure is funded and customers receive their share of the efficiency gains.
Cover: conceptual AI-generated illustration of cloud infrastructure, semiconductors and energy exposure; not actual Amazon or Qualcomm hardware.
What the agreement actually says
Qualcomm’s company-issued announcement describes multiple generations of customized silicon for AI inference and optical connectivity, including solutions extending to 1.6T. Qualcomm also plans greater use of AWS infrastructure, including Amazon Bedrock, for chip-design workloads. These are development and efficiency objectives, not independently demonstrated performance gains. Qualcomm announcement, distributed by PR Newswire
The accompanying filing records the following terms:
| Term | Disclosed detail |
|---|---|
| Issued | 3 September 2026, to an Amazon affiliate |
| Maximum underlying shares | 25 million Qualcomm common shares |
| Exercise price | $161.26 per share |
| Initially vested | 3.75 million, based on initial purchase commitments |
| Further vesting | Tranches linked to commercial arrangements, binding orders and actual purchases |
| Payment scale | Up to $60 billion associated with the vesting framework |
| Expiry | 3 September 2036 |
| Settlement | Cashless exercise permitted |
The $60 billion is not a guaranteed order or recognized revenue. Vesting is also distinct from exercising a warrant and owning the underlying shares. Qualcomm 8-K; accessible reproduction of the filing.
The public terms do not establish product-level margins, an annual delivery schedule or realized savings for AWS. Dividing $60 billion by ten years would create an unsupported revenue forecast.
Why the warrant is not a $4 billion gift
Multiplying the maximum share count by the exercise price produces $4.0315 billion. That is the hypothetical outlay for exercising the entire warrant for cash. It is neither the value of the incentive nor cash Amazon has already received. Cashless exercise means that outlay is not obligatory.
The distinction becomes clearer with illustrative arithmetic:
| Hypothetical Qualcomm share price | Intrinsic value per vested warrant share | Aggregate intrinsic value if all 25 million were vested |
|---|---|---|
| $150 | $0 | $0 |
| $200 | $38.74 | $968.5 million |
| $250 | $88.74 | $2.2185 billion |
These are scenarios, not price targets or fair-value estimates. The calculation is the positive difference between the assumed share price and $161.26, multiplied by the assumed vested shares. It excludes time value, taxes, contractual adjustments and the probability or timing of vesting. A warrant can have time value even when its current intrinsic value is zero.
For investment analysis, the key point is conditionality. The eventual equity benefit depends on commercial progress and Qualcomm’s share price. A theoretical payoff should not be booked today as procurement savings.
Amazon: purchasing power must become customer economics
The strongest Amazon interpretation has three parts.
First, another credible technical option can improve procurement flexibility. If a supplier can meet demanding workload, reliability and delivery requirements, AWS has more room to optimize its infrastructure. The benefit could come from performance, energy efficiency, commercial terms or resilience of supply. The announcement alone does not establish which of those benefits will arrive.
Second, the equity incentive can return part of a successful supplier relationship to Amazon. A conventional buyer benefits from the product. Here, Amazon may also benefit if the supplier’s equity appreciates and the relevant rights vest. That creates potential upside without making the product purchase costless.
Third, infrastructure economics are determined at system level. A less expensive chip is not necessarily a less expensive service. Software integration, networking, memory, power, utilization, reliability and engineering effort all affect the cost of delivering a useful result.
The right comparison is therefore the cost of a completed workload at an acceptable service level—not a chip’s purchase price in isolation.
There is an important counterargument. Better hardware economics may be competed away through lower cloud prices. Customers could capture much of the benefit. That might stimulate demand, but Amazon still needs the resulting volume and margin dollars to justify the additional capital.
My purchasing-power thesis would weaken if the incentive encouraged Amazon to buy an inferior or more expensive overall solution merely to unlock equity rights. The warrant should improve the economics of an already sound procurement decision, not rescue an unsound one.
Qualcomm: a valuable customer still has to be profitable
For Qualcomm, the opportunity is to turn a major customer relationship into repeat business across product generations. Engineering work can become more valuable if it supports subsequent products and a broader customer base.
But order scale is a poor substitute for profitability.
The supplier must earn enough after development, production, support and capital requirements to compensate shareholders for the incentive and potential dilution. Contractual economics also matter: who absorbs a redesign, a delayed ramp, excess inventory or a change in customer requirements?
There are two very different outcomes:
- A strong customer accelerates adoption, spreads development costs and supports attractive incremental returns.
- A powerful customer captures most of the economics while the supplier carries costly customization and execution risk.
The disclosed warrant cannot tell us which outcome will prevail. Nor is 25 million the number of newly issued shares today; eventual issuance depends on vesting and exercise, including the settlement method.
This is not necessarily a contest with one winner. Both companies can earn attractive returns if the technology creates enough value. Both can disappoint if the infrastructure costs more than customers are ultimately willing to pay.
Amazon’s current numbers show why cash conversion matters
The latest reported quarter provides the starting point—not a measurement of this newly announced deal.
| Reported measure | Period | Result |
|---|---|---|
| AWS revenue | Q2 2026 | $42.2 billion; up 37% year over year |
| AWS operating income | Q2 2026 | $16.6 billion, versus $10.2 billion |
| Operating cash flow | Twelve months to 30 June 2026 | $161.4 billion |
| Net property-and-equipment purchases | Same twelve months | Approximately $169.0 billion |
| Company-defined free cash flow | Same twelve months | Outflow of $7.6 billion |
Amazon also reported $53.4 billion of non-operating pre-tax other income, primarily from Anthropic investments. For the distinction from operating earnings, see Why headline profit is often the wrong number. Source: Amazon’s Q2 results and cash-flow reconciliation.
The analytical tension is straightforward: a business can generate growing operating profit while investment consumes more cash than operations currently provide.
That is not, by itself, evidence of poor capital allocation. New capacity can require cash well before it reaches productive utilization. Conversely, rising accounting earnings do not establish that the full investment will earn an adequate return.
Depreciation spreads an asset’s cost over time; the cash may leave much earlier. Useful lives, replacement requirements and technological obsolescence therefore belong in the same discussion as revenue growth. Longer accounting lives would not make equipment last longer economically.
I would also avoid treating free cash flow as money automatically available for distributions. Lease obligations, financing needs, acquisitions and other commitments still matter. The useful question is whether recurring cash generation strengthens after the business funds the assets it actually needs.
Oil is a separate pressure on the same cash-flow statement
Brent reached an intraday high of $99.46 on 8 September following attacks on Saudi energy facilities, according to Reuters’ oil-market report. That is an intraday observation, not a closing price or forecast.
For Amazon, my concern is conditional transmission. Persistently higher delivered fuel costs could raise transportation expense. Higher household energy bills could also restrict discretionary spending. The effects depend on contracts, surcharges, delivery density, product mix and customer behavior.
Crude prices are not a direct measure of either Amazon’s fuel bill or data-center electricity prices. Regional power markets and procurement arrangements matter. I would not translate a one-dollar oil move into an Amazon earnings estimate without a defensible exposure model.
This builds on the earlier costs-and-rates analysis: operating pressure and valuation pressure can occur together, but neither follows mechanically from every oil-price increase.
What would confirm—or weaken—the investment case?
I will use subsequent company results and filings to test the following, with the next reporting cycle providing the first update rather than a final verdict on a multi-year collaboration.
| Test | Evidence that strengthens the case | Evidence that requires reassessment |
|---|---|---|
| AWS earnings | Sustained operating-profit growth as capacity enters service | Growth depends on price concessions while costs accelerate |
| Cash conversion | Operating cash generation catches up with investment over time | Repeated investment increases without corresponding cash improvement |
| Technology execution | Deployments and credible workload-level performance evidence | Delays, redesigns or integration costs overwhelm projected savings |
| Qualcomm economics | Commercial progress accompanied by profitable growth | Revenue expands but development costs and incentives absorb the benefit |
| Retail resilience | Efficiency and mix absorb persistent cost pressure | Fuel, fulfillment costs and weaker spending erode margins |
No single quarter settles those tests. The direction across several reporting periods matters, especially when capacity is still being built.
Good commercial news is not a valuation conclusion
This agreement improves the questions I can ask about Amazon’s competitive position. It does not supply a price target.
A defensible valuation would need an explicit path for AWS and retail earnings, reinvestment, depreciation, taxes, financing and eventual cash generation per share. It would then need to be compared with a dated market price. This article does not establish that Amazon or Qualcomm is mispriced.
There are three plausible paths worth monitoring:
| Path | Business outcome | Investor implication |
|---|---|---|
| Profitable scaling | Useful capacity, sufficient demand and disciplined investment | Better cash returns can support value for both parties |
| Customer-led gains | Hardware improves, but cloud competition passes savings through | Usage can grow faster than shareholder cash returns |
| Delayed payback | Deployment or demand lags while spending continues | Earnings and valuation assumptions may require downward revision if the delay exceeds what was priced in |
These are analytical scenarios without assigned probabilities. The public disclosures do not justify precise odds or an investment return estimate.
My shareholder conclusion
I own Amazon. The Qualcomm arrangement is a reason to examine the quality of Amazon’s spending more closely, not a sufficient reason on its own to increase a position.
My working view is that platforms with customer access and procurement scale may negotiate a meaningful share of AI’s economics. Suppliers with scarce, differentiated technology can retain substantial value too. The division depends on alternatives, performance and execution—not the platform or semiconductor label.
This extends the question raised in the AI concentration review: exposure to AI spending is not the same thing as ownership of its eventual profits.
Amazon has gained potential equity upside alongside a supplier relationship. The durable advantage will be demonstrated when those terms help produce better cash returns per share.
Disclosure: I hold AMZN. This article records my analysis and opinions for education; it is not personalized investment advice. All investments involve risk, including loss of capital. Company plans and hypothetical scenarios are not guaranteed outcomes.