Two Signals. One Conclusion.

This week, Anthropic accidentally revealed their next model to the world. They deleted it within hours. And hardware costs are still climbing. These two things are happening at the same time for a reason.

Signal One: They Leaked a Superintelligence Model by Accident

Anthropic did not mean to tell anyone about Mythos. It slipped out through what Fortune described as an “unsecured data trove” — internal details about their next model release, sitting exposed long enough for the press to find it before anyone hit delete.

The specific name does not matter. What matters is the pattern. A company at the absolute frontier of AI development, working on the most consequential technology in human history, could not keep a lid on it for more than a few hours. Not because they are careless. Because the pace is so fast that operational security is losing the race to the release cadence.

Think about that for a second. We are moving so quickly that the companies building these systems cannot manage the information flow around them. Every few months there is a new capability jump, a new benchmark broken, a new model that makes the last one look like a prototype.

The frontier is not stable. It is not plateauing. The companies closest to it cannot even keep their own roadmaps from leaking. If you are waiting for things to settle down before you start building, you are waiting for something that is not coming.

Signal Two: Hardware Costs Are Still Going Up

At the same time that models are accelerating, the cost of the compute underneath them is not dropping. It is rising. Data center buildouts are at historic scale. GPU lead times are long. Power infrastructure is a genuine constraint. The big players are signing hundred-billion-dollar deals for chips and electricity.

If you have been waiting for AI to get cheap enough to justify getting in, you have been misreading the market. The cost curve for running frontier models is not going down the way consumer software costs did. Demand is outrunning supply at every level of the stack.

Both signals point the same direction. The convergence is not a coincidence.

What Two Upward Curves Actually Mean

Here is what most people get wrong. They see rising compute costs and think, “This is getting harder to afford.” They see accelerating models and think, “Maybe I should wait for the next version.” Both instincts lead to the same place: standing still while the world moves.

The real meaning of two upward curves is this: the gap between the people who started and the people who have not is widening every single day. The people who have been building with AI for two years are not just two years ahead. They are operating in a different category. They have the workflows, the instincts, the context, and the compounding advantages that come from actually doing the work.

Rising capability means the tools available to early movers keep getting better. Rising costs mean the barrier to entry keeps going up for late ones. Both trends benefit the person who is already in the game. Neither one benefits the person on the sideline.

There is no coming correction that will make this easier to start later. The window has been open for a while. It is still open. But two upward curves do not stay patient forever.

The Question Is Not Whether to Start

The Mythos leak is not interesting because of what the model does. It is interesting because of what it says about the speed of the people building it. They are moving too fast to manage their own information. That is not a warning. That is a signal.

If the companies at the frontier are sprinting, and the cost of the infrastructure they run on is climbing, the only rational response is to start building now with what exists today. Not perfect tools. Not settled technology. The tools that are available right now, which are already better than anything that existed 18 months ago.

Two signals. One conclusion. The time to start was yesterday. The second-best option is today.