Do ChatGPT (OpenAI) and Claude (Anthropic) have any moats? There’s currently a lot of talk in the media and on X that frontier labs have no real moats, and that as they move toward public markets, they’re increasingly trying to secure them through regulation. Independent of what’s true there, let’s assess the moats they actually have today and the ones they could develop, using Hamilton Helmer’s 7 Powers. 🏗️ 1. Scale Economies: 5/5 Very strong for both, with a slight edge to OpenAI. Their sheer compute scale should create better unit economics through utilization, purchasing power, volume discounts and infrastructure optimization. But compute scale also matters on the product side. More compute means more resources to train models, experiment, run evals and improve products. Scale helps both margins AND model development. 🌐 2. Network Economies: Developing, potentially huge Does every additional user make the product more valuable to other users? Today: somewhat. Long term: potentially massively. Think agents communicating with other agents, businesses connecting their systems into the same ecosystem, users interacting through AI, and developers building apps for a massive installed base. More users → more developers/integrations → better ecosystem → more users. Anthropic has particularly strong potential here in B2B. OpenAI has that PLUS potentially enormous consumer network effects. This might ultimately become their strongest moat. ⚔️ 3. Counter-Positioning: Basically gone OpenAI initially attacked incumbents from a completely different position. Now OpenAI and Anthropic ARE the incumbents. Counter-positioning is increasingly being used against them: Open weights. Cheaper inference. Customizable models. Different business models. I wouldn’t consider this a meaningful moat for either anymore. 🔒 4. Switching Costs: Developing rapidly This could become enormous. Memory is the obvious early example. But take it much further: Your AI knows your entire history. Your contacts. Your company. Your files. Your preferences. Your workflows. Your agents. Your personal context. Everything is customized around you. At that point, switching AI systems could become incredibly painful. Today: moderate. Potential: massive. 🏷️ 5. Branding: OpenAI very strong, Anthropic moderate For most consumers, ChatGPT basically IS AI. That’s an enormous brand asset for OpenAI. I’m less convinced branding matters nearly as much in B2B. If an enterprise is spending millions and AI materially affects its productivity, it will eventually care much more about performance, reliability and economics than which logo is on the product. So: Consumer branding: very strong OpenAI moat. Enterprise branding: much weaker moat. 🧠 6. Cornered Resource: Weak today Compute overlaps heavily with Scale Economies. Could preferential access to NVIDIA GPUs or infrastructure become a cornered resource? Maybe, but probably not sustainably enough. Talent is another candidate. But if you’re paying extraordinary compensation to retain that talent, is it really cornered? A lot of talent will likely follow money and opportunity. One interesting potential cornered resource would be privileged regulatory or government access. If regulation eventually creates barriers that only gigantic incumbent labs can satisfy, that could become a very real moat against smaller competitors. That is one reason the regulatory capture debate matters so much. ⚙️ 7. Process Power: Unproven, potentially enormous Today, I don’t think this is proven. But self-improving AI could completely change that. Imagine: Better model → better AI researchers → faster research → better internal tools → better model → repeat. If frontier labs develop deeply embedded, proprietary AI-driven R&D processes that outsiders cannot easily replicate, Process Power could become enormous.