The most powerful AI resources — compute, data, and the talent to put them to work — are stacking up inside a small circle of companies. That concentration now looks like a systemic disruption risk, and it's driving calls for targeted regulation and a more diversified approach to investment. Martin Casado, a general partner at a16z, has weighed in on a key part of the debate: scaling laws, he says, are not breaking.
One Weak Link, Whole Chain
The worry here isn't just about competition. It's about the way the AI economy is built. A handful of firms control the infrastructure, the datasets, and the models that everyone else builds on. If one of those companies hits a technical wall, a supply constraint, or a regulatory shock, the impact doesn't stop at its own balance sheet. It spreads through every layer of the industry that depends on it.
That's the classic shape of a systemic risk — the same logic that makes regulators watch big banks closely. In this case, the failure of a single AI player could take down a wide network of products and services.
Scaling Laws Are Still Working
Martin Casado, a general partner at Andreessen Horowitz, addressed one of the most contested questions in AI right now: whether scaling laws are losing their power. His answer was clear — scaling laws continue to hold and are not breaking.
That matters because scaling laws are the engine driving the race toward larger and larger models. As long as they hold, the companies with the most compute, the most data, and the most specialized talent will keep pulling away from the rest. Which is exactly the concentration that creates the systemic risk in the first place.
Targeted Rules, Not a Blanket
The proposed response isn't a sweeping ban on AI. It's targeted regulation aimed at the specific spots where the concentration is most dangerous — the shared infrastructure, the concentrated compute power, the data moats. The idea is to reduce the systemic risk without putting the brakes on innovation.
Alongside regulation, the argument runs, there's a need for diversified investments. Spreading capital across the AI stack — not just the few big model builders, but the data providers, the infrastructure layers, the tools, the companies that haven't yet concentrated their power — would soften the blow if one pillar crumbles.
What Happens Next
The unresolved question is timing. If scaling laws keep holding, concentration will likely deepen. Regulators will face a choice: act before a disruption hits, or wait until the first real failure forces their hand. That first failure might not be a regulatory one at all.
