The complete stack powering plentiful AI capability

Artificial intelligence is becoming more capable and more affordable at the same time, and there's a specific reason why. OpenAI's CFO, Sarah Friar, recently broke down how progress in four key areas—chips, compute, models, and products—builds on itself to deliver intelligence that is more useful, available at greater scale, and less expensive for everyone. This article is for anyone curious about the business and engineering logic behind modern AI, explained in plain language. By the end, you'll understand why improvements at the bottom of the technology stack ripple all the way up to the tools you use every day.

The Four Layers Working Together

Friar's central point is that AI progress isn't driven by any single breakthrough. Instead, it comes from four layers that reinforce one another:

1. Chips — the physical hardware that does the actual number-crunching
2. Compute — the data centers and infrastructure that run thousands of chips together
3. Models — the software "brains" trained on that hardware
4. Products — the applications people actually use, like ChatGPT

Each layer improves independently, but the real magic happens when those gains stack. A better chip makes compute more efficient. More efficient compute makes training larger models affordable. Better models make products more valuable. And successful products generate the revenue and feedback needed to invest in the next generation of chips and compute.

Why Compounding Matters

The word "compound" is doing a lot of work here. Think of it like compound interest in a savings account: small gains in multiple places multiply rather than simply add up. If chips get a modest efficiency boost, and software gets better at using those chips, and models get better at delivering useful answers with fewer steps, the combined effect can be dramatic.

This compounding is what allows OpenAI to serve more people at a lower cost per interaction. What required enormous computing resources a couple of years ago can now be delivered far more cheaply, which means intelligence becomes accessible to more users, in more languages, across more use cases.

Scale and Cost Move in the Right Direction

Two of Friar's key themes are scale and cost—and they're closely linked. As infrastructure grows and models become more efficient, the cost of serving each user request drops. That falling cost is what makes it possible to offer powerful AI to hundreds of millions of people, not just enterprise customers with big budgets.

It also opens the door to new kinds of products. When a single AI interaction is cheap enough, developers can build tools that call on AI many times behind the scenes—summarizing, translating, planning, and checking work—all in one seamless experience. Expensive intelligence limits creativity; affordable intelligence expands it.

What This Means for Everyday Users

You don't need to understand semiconductor design to benefit from this trend. The practical takeaway is simple: the AI tools available today are likely to keep getting better and cheaper, because the entire stack underneath them is improving at once.

For consumers, that means more capable assistants. For developers and small businesses, it means the cost of building with AI keeps falling. And for the industry as a whole, it suggests this cycle of investment and improvement is self-sustaining—as long as each layer continues to advance.

The Bottom Line

Sarah Friar's framing offers a useful mental model: AI progress is not one leap but a staircase where each step—chips, compute, models, products—lifts the others. That's how OpenAI aims to deliver more useful intelligence at greater scale and lower cost, and it's why the pace of improvement shows no sign of slowing down.

Note: This article summarizes publicly reported remarks. For exact figures and full context, refer to OpenAI's original statements.