Leonid Zadorozhnykh @ProDoser: "Anthropic just gave me five..."

Finland

Anthropic just gave me five more days on its best model. Nice for me. The more interesting question for investors: why is it rationed at all? Fable 5 was supposed to leave subscription plans on July 7. Anthropic pushed that to July 12, still capped to 50% of the limit, still limited. I burned through my allowance every single day this past week, so I am not complaining about the extra time. But honestly, my usage limit is not the story. The bottleneck behind it is. One of the most in-demand AI models on the market got put on a leash. Anthropic capped it to 50% of weekly usage and is moving it to metered pricing, and they were blunt about why: not enough compute to serve demand. And this is after adding over 220,000 GPUs and more than 300 megawatts through a new data center deal with $SPCX (Space Exploration Technologies Corp) Still not enough. Two stories are getting mixed together here. One is government review and export control, which pulled the model offline back in June. The other is pure capacity. As an investor, I care far more about the second. Now look across the field. OpenAI limited its newest model to a small group of approved partners after a government request over cyber concerns. Google’s flagship slipped from June into July and still is not broadly out. The leading labs are gated, supply constrained, or both. Nobody is sitting on spare compute right now. My own read, from my own usage and from people actually building with these tools: one engineer burning 500$ to 1500$ a month on tokens is becoming normal, and companies sign off on it without blinking. And when the squeeze hit, guess who kept priority. Enterprise and API customers. The cheap consumer plans got throttled first. That is the whole thesis in one line: One engineer on a 1000$ budget beats fifty retail users on a 20$ plan, and providers serve the budget that pays. While the market keeps arguing about a bubble, the bottleneck looks physical and frankly boring: chips, data centers, power, grid. This is not demand from a slide deck. It is real companies that cannot serve the customers they already have. So I still prefer exposure to the picks and shovels, not just the models. $NVDA (NVIDIA Corporation) $ASML (ASML Holding NV) and $MU (Micron Technology, Inc.) on silicon. $MSFT (Microsoft) on hyperscalers who might benefit from their capex spending once infrastructure is ready! $SPCX becomes interesting, given whose compute it now runs. A few open questions I keep chewing on: Is rationing a bubble signal, or real demand running ahead of supply? Does a government gate make a model look dangerous, or does “too good to export” just make it the most wanted product in the room? If the real money is enterprise tokens and not consumer subscriptions, is the market pricing the wrong layer of the stack? I am not calling a top or a bottom. I am watching a supply story that keeps getting sold as a bubble story. Where do you land on this pullback in $NSDQ100 and $SPX500 buying opportunity, or the start of a repricing? ℹ️ Disclaimer: This post reflects my personal opinions and market observations and is not financial advice. 🤝 You can start copying my trades from just 300$ - check pinned post for more details. ⚠️ Past skyrocketing gains are not a guarantee of future results.

Not investment advice. The author may have financial interests in the mentioned instruments.