To The Moon

For most of corporate history, the relationship between a customer and a supplier was relatively simple.

One company made something.

Another company bought it.

Money went one way.

Products went the other.

Artificial intelligence may be changing that relationship.

Because some of the largest technology companies on Earth are no longer simply buying AI infrastructure.

They are increasingly taking economic stakes in the companies building it.

Amazon has just entered a multi-generation AI infrastructure partnership with Qualcomm.

The headline number is enormous.

Amazon could purchase up to US$60 billion of Qualcomm AI data-centre chips and related products under the agreement.

But the part that interests me more is buried underneath.

Qualcomm has also granted Amazon warrants worth approximately US$4 billion, allowing Amazon to acquire Qualcomm shares at US$161.26 each.

So Amazon is not merely becoming a massive Qualcomm customer.

It could also become a Qualcomm shareholder.

And this is not an isolated experiment.

Only weeks ago, Google struck an expanded custom-silicon agreement with Marvell.

Google received warrants to purchase almost 59 million Marvell shares at US$206.58 each, representing a potential stake worth around US$12.2 billion if fully exercised.

Even more interestingly, most of those warrants do not simply vest with time.

They vest as Google buys products.

Marvell’s filing shows the purchase-linked portion divided into 240 separate tranches, with one tranche vesting for every US$500 million of qualifying custom-product revenue through fiscal 2033.

Think about the structure.

Google buys more chips.

More warrants vest.

Google potentially owns more of the supplier.

If Marvell becomes more valuable partly because Google keeps buying its products, Google participates in some of that upside.

That is a very different relationship from:

“Send us 100,000 chips and invoice Accounts Payable.”

🧠 PROCUREMENT IS STARTING TO LOOK LIKE STRATEGIC CAPITAL

I think we may be watching a new stage of the AI infrastructure boom.

The first stage was straightforward.

Hyperscalers bought enormous quantities of NVIDIA GPUs.

The second stage involved designing more silicon internally.

Google developed TPUs.

Amazon built Trainium and Inferentia.

Microsoft created Maia.

Meta pursued custom accelerators.

The motivation was obvious.

Reduce dependence on a single supplier.

Optimise chips for specific workloads.

Improve power efficiency.

Control cost.

Own more of the technology stack.

But designing a chip internally does not mean manufacturing every piece of it yourself.

Hyperscalers still need semiconductor expertise.

Networking.

Memory interfaces.

Optical connectivity.

Storage controllers.

Advanced packaging.

Foundry capacity.

Chip design expertise.

And that has created another possible strategy.

Instead of buying the semiconductor company outright, align its economic future with yours.

That is what purchase-linked warrants can accomplish.

They sit somewhere between an ordinary supplier contract and an acquisition.

I would call it:

SOFT VERTICAL INTEGRATION.

Big Tech does not technically own the supplier.

But it gains influence, alignment and potential financial upside as the relationship expands.

🔥 AMAZON’S QUALCOMM DEAL SHOWS HOW BIG THIS COULD BECOME

Qualcomm’s partnership with Amazon is not simply about one accelerator.

The companies plan to collaborate across multiple generations of customised silicon for AI inference.

They are also working together on optical connectivity reaching 1.6 terabits per second, with future generations planned beyond that.

Qualcomm also intends to use AWS infrastructure more deeply in its own chip-design workflows.

That’s fascinating because the relationship operates in both directions.

Amazon becomes a huge Qualcomm customer.

Qualcomm becomes a larger AWS customer.

Amazon helps shape Qualcomm’s AI silicon roadmap.

Qualcomm helps Amazon diversify its AI infrastructure.

And Amazon potentially becomes a Qualcomm shareholder.

This is no longer just a supply contract.

It is increasingly an ecosystem.

💰 WHY WOULD AMAZON WANT EQUITY UPSIDE?

Imagine Amazon buys tens of billions of dollars of Qualcomm technology.

That demand helps Qualcomm grow its data-centre business.

The market begins valuing Qualcomm differently.

Qualcomm’s share price rises.

If Amazon owns warrants, Amazon potentially participates in that increase in value.

In other words:

Amazon’s own purchasing behaviour can help strengthen the supplier while simultaneously giving Amazon exposure to the supplier’s upside.

That does not make the chips free.

Warrants have exercise prices, conditions and risks.

But economically, it creates something much more interesting than a normal procurement relationship.

The customer can potentially benefit from helping create the supplier’s growth.

Google’s Marvell agreement makes this especially visible because warrant vesting is explicitly connected to product purchases.

The larger the commercial relationship becomes, the stronger the potential ownership relationship becomes too.

That is a powerful incentive alignment.

🏭 THERE IS ANOTHER REASON BIG TECH MAY LIKE THIS MODEL

AI infrastructure requires an extraordinary amount of capital.

Suppliers need to hire engineers.

Secure manufacturing capacity.

Develop advanced packaging.

Build networking technology.

Acquire intellectual property.

Design multiple generations of chips years before customers deploy them.

That creates risk.

If a semiconductor company spends billions developing products for one customer and that customer later changes direction, shareholders can be left holding the bill.

Purchase commitments change that equation.

A large customer effectively says:

“Build this capacity. We intend to use it.”

And equity-linked structures go further.

They say:

“If this partnership creates enormous value, we want to participate in some of it too.”

That could encourage suppliers to take bigger technological risks because demand visibility improves.

For an industry facing enormous upfront development costs, that matters.

⚔️ THIS IS ALSO ABOUT NVIDIA

I do not think Amazon needs Qualcomm to defeat NVIDIA.

I do not think Google needs Marvell to destroy Broadcom.

That framing is too simplistic.

The hyperscalers have a different objective.

They need options.

If NVIDIA remains the best choice for certain workloads, they can buy NVIDIA.

If an internal accelerator delivers better economics for inference, they can use that.

If Qualcomm provides power-efficient customised silicon, Amazon can deploy it.

If Marvell helps Google build additional TPU-related components, Google can use those too.

The hyperscaler does not need one winner.

It benefits from creating competition among suppliers.

That could ultimately become one of the most powerful negotiating tools in the AI industry.

The threat is no longer:

“Lower your price or we will buy from your competitor.”

It increasingly becomes:

“Lower your price or we will design around you, fund an alternative supplier and potentially own part of the upside.”

That is a very different power balance.

🚨 THE MARGINAL AI WORKLOAD COULD MATTER MORE THAN MARKET SHARE

NVIDIA can continue growing enormously while custom silicon also expands.

Those ideas are not contradictory.

The important question is where the next dollar of AI compute gets deployed.

Training the largest frontier models may continue favouring high-end general-purpose accelerators.

Inference can be different.

Inference involves repeatedly running models after they are trained.

At sufficient scale, optimising silicon for a specific workload can produce huge economic benefits.

That is exactly where Amazon and Qualcomm say they are collaborating.

So hyperscalers do not necessarily need custom chips to replace GPUs everywhere.

They need custom chips to become economically superior somewhere.

Then somewhere becomes larger.

📊 QUALCOMM’S STRATEGY IS CHANGING TOO

For decades, investors primarily associated Qualcomm with smartphones.

Snapdragon.

Modems.

Licensing.

Wireless technology.

That remains enormously important.

But the company has spent years diversifying.

Automotive.

Internet of Things.

Edge computing.

And increasingly:

AI data centres.

Qualcomm now targets roughly US$15 billion of annual data-centre chip revenue by 2029, and Amazon joins Microsoft and Meta among major cloud companies supporting that push.

Amazon’s agreement therefore matters for Qualcomm beyond the headline revenue opportunity.

It provides validation.

One of the world’s largest cloud operators has effectively said:

We believe your technology belongs inside our future infrastructure.

That can change how investors perceive the entire company.

Qualcomm may eventually become less of a smartphone semiconductor company with some diversification…

and more of a diversified computing platform spanning the phone in your pocket all the way to the AI data centre.

🔌 AND LOOK WHAT KEEPS APPEARING AGAIN

Optics.

The Amazon agreement includes advanced optical connectivity.

Qualcomm acquired Alphawave for approximately US$2.4 billion, gaining high-speed connectivity technology that now forms part of its data-centre strategy.

We’ve seen this theme repeatedly across AI infrastructure.

Compute gets faster.

Clusters become larger.

Then moving the data becomes harder.

So companies entering AI data centres increasingly need to offer more than processing.

They need:

Compute.

Memory interfaces.

Networking.

SerDes.

Optics.

Power efficiency.

Systems expertise.

AI infrastructure is slowly becoming a battle over the whole system, not just the chip.

🐂 THE BULL CASE FOR THESE DEALS

For suppliers, these relationships can be extremely powerful.

They provide:

Demand visibility.

Customer validation.

Long product roadmaps.

Potentially enormous revenue.

Better justification for R&D spending.

Closer access to the customer’s future workloads.

And when the customer holds equity upside, incentives become more aligned.

A hyperscaler wants its supplier to remain financially healthy, technologically competitive and capable of investing.

That can strengthen the partnership.

🐻 BUT THERE IS A DARKER SIDE

Every strategic partnership creates dependence.

If a huge percentage of a supplier’s future growth depends on one hyperscaler, investors need to ask uncomfortable questions.

What happens if that customer changes architecture?

What happens if another supplier wins the next generation?

What happens if internal silicon improves?

What happens if AI capex slows?

What happens if pricing becomes less attractive?

Marvell provides a useful example.

Its Google partnership carries enormous long-term potential, but investors recently punished the stock partly because meaningful revenue from the expanded arrangement may arrive later than the market initially hoped.

A giant contract headline does not automatically equal giant earnings tomorrow.

The timeline matters.

Margins matter.

Concentration matters.

Execution matters.

⚠️ CUSTOMER AND SHAREHOLDER INTERESTS ARE NOT ALWAYS IDENTICAL

This may become another fascinating tension.

A customer wants:

Lower prices.

Better performance.

More supply.

Faster development.

A shareholder generally wants:

Higher margins.

Strong returns on capital.

Pricing power.

Profitable growth.

What happens when the same company occupies both seats?

Amazon could eventually want Qualcomm products as cheaply as possible while simultaneously benefiting if Qualcomm creates enormous shareholder value.

Google may want excellent economics from Marvell while also owning warrants tied to the relationship.

Usually customers and shareholders sit on opposite sides of the table.

AI may increasingly place them on both sides simultaneously.

That could produce very interesting negotiations.

🕸️ BIG TECH MAY BE BUILDING SUPPLY NETWORKS WITHOUT ACQUIRING THEM

This is the larger thesis I keep coming back to.

Traditional vertical integration looks like this:

Company A buys Company B.

Full ownership.

Full control.

Full regulatory scrutiny.

Full capital commitment.

But Big Tech may not need to do that.

Imagine instead:

Amazon owns strategic rights in several suppliers.

Google owns strategic rights in several suppliers.

Meta supports specialised chip companies.

Microsoft works closely with custom-silicon partners.

Each hyperscaler builds a network of companies economically aligned with its infrastructure roadmap.

Nobody technically owns the whole supply chain.

But everyone becomes deeply connected.

You could end up with something resembling an AI industrial network, where purchase commitments, warrants, joint development agreements and long-term capacity reservations link companies together.

That structure could become one of the defining features of the AI economy.

🤔 THEN THERE IS THE CIRCULARITY QUESTION

This is where investors should remain sceptical.

Customer buys products.

Supplier’s revenue rises.

Supplier’s share price rises.

Customer receives or exercises equity rights.

Customer benefits from supplier valuation.

Supplier uses stronger financial position to expand.

Customer buys more.

That can be entirely legitimate and economically rational.

But investors still need to understand how much growth reflects organic broad-based demand versus highly concentrated strategic relationships.

Not all revenue dollars deserve identical valuations.

US$10 billion coming from thousands of customers is economically different from US$10 billion heavily dependent on one hyperscaler.

The revenue is real in both cases.

The risk profile is not.

🎯 MY TAKE

I started looking at the Qualcomm announcement as another AI chip deal.

Then I saw the Amazon warrants.

Then I remembered Google’s Marvell warrants.

And suddenly the story looked much bigger.

I don’t think Big Tech is simply diversifying its semiconductor suppliers.

I think it may be creating a new form of AI vertical integration.

Not through traditional acquisitions.

Through:

Purchase commitments.

Joint product development.

Equity warrants.

Cloud relationships.

Multi-generation roadmaps.

Strategic capacity alignment.

The result could be an AI supply chain where the distinction between:

customer

partner

supplier

and

shareholder

becomes increasingly blurry.

And if that continues, the next phase of the AI arms race may not simply be about who designs the best processor.

It may be about:

WHO BUILDS THE STRONGEST NETWORK OF COMPANIES AROUND IT.

That is why Amazon’s Qualcomm deal matters.

Not because Qualcomm suddenly replaced NVIDIA.

It didn’t.

Not because US$60 billion of purchases are guaranteed tomorrow.

They aren’t.

It matters because the structure itself tells us something.

Big Tech is no longer satisfied simply buying the AI supply chain.

It increasingly wants economic participation in the supply chain it helps create.

And if Google + Marvell and Amazon + Qualcomm become a template rather than exceptions…

we could be watching the beginning of a completely different AI industrial structure.

👇 COMMUNITY QUESTION

Which model wins the next phase of AI infrastructure?

A) Buy the best chips available

B) Design everything internally

C) Build custom chips with strategic suppliers

D) Build supplier ecosystems using contracts + equity alignment

And here’s the bigger question:

Do purchase-linked equity warrants create better long-term alignment, or do they make the AI supply chain dangerously concentrated around a handful of hyperscalers?

$QCOM $AMZN $MRVL $GOOGL $NVDA

Personal market analysis only. Not financial advice. Always do your own research.