Building an AI Stack the World Wants to Choose

Over the past several months, as the U.S. government has begun exploring how to structure “tech stacks” for AI exports, one theme has emerged in nearly every conversation I’ve had with innovators: deep anxiety that Washington’s approach might inadvertently advantage one segment of the tech industry over another.

You hear it in the voice of the chipmaker worried that cloud providers may dominate.
You hear it in the cloud provider who fears chip vendors or model developers will be crowned the “winners.”
You hear it from model companies concerned they will be boxed out by highly integrated platforms.

And in capitals around the world, you hear an even broader concern:

Will the U.S. design an AI architecture that empowers citizens — or one that centralizes control?
Will it make room for allied firms and local partners?
Will it be flexible enough to serve a wide range of markets, or so restrictive that it leaves whole regions to authoritarian models?

These questions sent me back to two very different chapters in my own life:
my years in acquisitions at the Pillsbury Company, and my years in Congress.

Surprisingly, both offer useful lessons.

What the Ice Cream Business Taught Me About Market Structure

While at Pillsbury — where I helped acquire a small Bronx ice cream company called Häagen-Dazs and expand it globally — I learned something few outside the industry appreciate: U.S. branded food companies enjoyed margins that were hard to match overseas.

Why?
Because the structure of the U.S. market favored brands.

Abroad, especially in the United Kingdom, the structure was reversed. The “Big Four” grocers — Tesco, Sainsbury’s, Asda, and Morrisons — wielded enormous power. They drove down margins for food companies, built powerful store brands, and concentrated influence at the retail layer.

Same products. Same consumers.
Wildly different outcomes because of the architecture of the market.

That lesson stayed with me:

Value, power, and innovation are shaped less by the product itself than by who controls the system around it.

We see similar dynamics in AI.

Some layers of the digital stack naturally concentrate — chips, cloud infrastructure, and large foundation models — because economics like scale, data gravity, switching costs, and network effects tend to create dominant players. Left unmanaged, these dynamics risk producing new chokepoints that undermine both innovation and sovereignty.

If the U.S. blesses a single, integrated stack dominated by one segment — hyperscalers, chip providers, or model companies — we risk recreating something like the UK Big Four: a concentrated gateway with enormous leverage.

If we design a rigid, one-size-fits-all system, we make U.S. technology less exportable, especially to countries with strong sovereignty expectations.

And that is not how the U.S. will win the global competition for trusted AI.

What Congress Taught Me About Intra-Industry Rivalries

In Congress, I learned that inter-segment rivalry is everywhere.

One of the clearest examples was the recurring tension between credit-card companies and retailers.
Both groups were indispensable.
Both claimed to be the “natural” leader in shaping payments policy.
Both feared the other gaining an advantage.
And both mobilized intensely when legislation was drafted.

(I saw similar dynamics between nurse anesthetists and anesthesiologists — a rivalry that could fill entire hearing rooms. The pattern was the same.)

These debates taught me something important:

The goal of public policy is not to referee turf wars.
It is to design systems that serve the public, ensure access, and promote innovation.

The same holds in tech.

Chip firms, cloud providers, model developers, telecom operators, cybersecurity companies, and integrators each believe themselves to be the “natural” backbone of the AI stack.

The U.S. government should listen — then rise above the rivalry.

What Should the U.S. Government Actually Optimize For?

Not which industry segment gets the upper hand.
Not which vendor commands the architecture.
Not which revenue model prevails.

Rather:

1. Empowering citizens, not controlling them.

AI exports must strengthen freedom, transparency, and opportunity — not enable digital authoritarianism.

2. Ensuring America remains the partner of choice.

That means technology must be exportable, adaptable, and sovereignty-respecting, not rigid or bundled.

3. Preserving U.S. innovation dynamism.

Our ecosystem thrives because no single layer dominates. Competition exists within and across layers. Breakthroughs can come from anywhere.

4. Preventing chokepoints.

We must avoid overdependence on any one hyperscaler, chipmaker, cloud model, or foundation-model provider.

5. Offering deployable, trusted alternatives to authoritarian systems.

Standards matter — but infrastructure wins.
The 5G experience proved that.
We need modular AI components, trusted telecom, sovereign cloud options, and competitive financing to lead with installed infrastructure governed by empowering standards.

What This Means for Designing an American “Tech Stack”

If I were offering advice to policymakers, it would be this:

A. Build a Modular Stack — Not a Mandated One

Modularity ensures:

• No hyperscaler becomes the “Big Four” of global AI.
• No chipmaker becomes a bottleneck.
• No model provider controls the entire pipeline.
• Innovation flourishes across the system.

Modularity is how we preserve dynamism while avoiding dependency.

B. Cloud Providers Are Essential — But Not the Only Possible Leads

Cloud is indispensable.
But the lead entity in a tech-stack consortium might instead be:

• a systems integrator,
• a model developer,
• a telecom provider,
• a cybersecurity firm, or
• a chip + model + security partnership.

Leadership should reflect governance capability, not simply who owns data centers.

C. Sovereignty Matters

Countries will not adopt U.S. AI systems if they fear loss of control.
Cloud zones — sovereign, trusted, hybrid — must be part of the architecture.

D. Financing Is Strategic Infrastructure

DFC, EXIM, USAID, and USTDA must be empowered to support deployment of trusted cloud and AI infrastructure, just as they have historically supported roads, power grids, and telecom networks.

Otherwise, as with 5G, competitors will create the facts on the ground.

E. Guardrails Must Support Empowerment — Not Control

Export rules should:

• protect model weights,
• enforce trusted compute,
• secure telecommunications,
• prevent exports to systems of repression, and
• encourage transparency and safety.

This is how the U.S. wins trust globally.

The North Star: Empowerment

Across Pillsbury’s acquisitions, across partisan debates in Congress, and across today’s AI policy world, one principle has remained constant:

Public policy should not be about choosing winners.
It should be about designing systems that empower people.

If we design the American AI tech stack with that principle in mind — modular, exportable, secure, sovereignty-respecting, and innovation-nourishing — we will advance U.S. interests while strengthening freedom globally.

That is how American leadership in AI can be not only technologically superior,
but morally compelling.

In line with my belief that embracing AI is essential to both personal and national success, this piece was developed with the support of AI tools, though all arguments and conclusions are my own.