OpenAI's CFO just laid out the clearest explanation yet of why AI is getting dramatically cheaper and more powerful at the same time — and it's not luck, it's a compounding stack.
The Breakthrough: Compounding Gains Across Every Layer
Sarah Friar's piece isn't a press release — it's a rare inside look at how OpenAI thinks about the architecture of progress. The core argument: improvements in chips, compute infrastructure, model efficiency, and product design don't just add up — they multiply.
Think of it like compound interest. A 20% gain at the chip level, combined with a 20% gain in model efficiency, doesn't give you 40% better AI. It gives you something closer to 44% — and that math keeps stacking across every layer of the system.
This is why the cost of running a capable AI model has fallen by orders of magnitude in just a few years, while the quality has shot upward. It's not one big invention — it's dozens of smaller wins compounding across the full stack.
What the "Full Stack" Actually Means for AI capability
Friar breaks the stack into four layers: chips (custom silicon getting faster and more efficient), compute (data centres optimised for AI workloads), models (training and inference becoming leaner), and products (smarter interfaces that extract more value from each model call).
The implication is significant: even without a single dramatic new model launch, AI gets meaningfully better every quarter just from incremental gains compounding across all four layers simultaneously.
For context, this is precisely why the future of AI inference is one of the most consequential — and underrated — topics in the field right now. Inference efficiency is where a huge chunk of these cost reductions are actually happening.
What This Means for Learners
If AI is getting cheaper and more capable on a compounding curve, the skills gap between people who understand AI and those who don't is widening at the same rate. The tools available to a skilled AI user today are dramatically more powerful than those available 18 months ago — and cost a fraction of what they once did.
Understanding how models are built, optimised, and deployed isn't just for engineers anymore. If you want to use AI effectively at work, knowing what's happening under the hood helps you anticipate what's coming next. Our course on how neural networks really work is a solid starting point for building that foundation.
The practical takeaway: don't wait for a single "breakthrough moment" to level up your AI skills. The breakthroughs are already happening — quietly, continuously, across every layer of the stack.
