Mirendil's $1 billion Series A for agentic AI in R&D marks a watershed moment—not because it's impressive (it is), but because it signals we've entered dangerous territory. When a Series A round hits $1 billion, led by Kleiner Perkins and Clearlake Capital, we're no longer evaluating business models or market readiness. We're making a bet that the entire category justifies valuations that dwarf companies with actual revenue and proven product-market fit.

The data from this week's top 10 funding rounds confirms the pattern. Mercor raised $750 million to connect AI developers with human experts—a 4-year-old company now valued at $20 billion. LangChain closed $300 million at a $3 billion valuation. Mavria, which literally just launched publicly, raised $130 million at a $1 billion valuation on the strength of its founders' pedigree from the Department of Energy. These aren't exceptions; they're the new normal. AI infrastructure is getting funded as if the infrastructure is the product, when in reality, AI infrastructure only matters if it solves a problem better and cheaper than alternatives.

What concerns me most is that we're funding the means before we've proven the ends. Yes, agentic AI for R&D and platforms to govern AI agents sound transformative. But how many of these companies will still be valued this way in three years if they haven't demonstrated that their technology actually accelerates R&D timelines or reduces costs? How many will survive if the capital influx stops and the market demands profitability over narrative?

The broader ecosystem is enabling this. When investors like Paradigm back Mavria with $130 million in pre-seed and seed funding combined, or when Magnetar Capital leads a $500 million growth round for Torus, we're not watching disciplined venture capital allocation. We're watching a consensus that AI infrastructure is too important to let market signals dictate its fate. That consensus could be right—or it could be building the next bubble.

Founders and investors in this space need to demand accountability from themselves. Before the next $1 billion Series A, the company should have proven not just that the technology works, but that customers actually need it, that they'll pay for it, and that it's defensible against competition. Mirendil might be all of those things. But the market is now priced as if every AI infrastructure play is. That's not confidence in the category. That's faith. And faith, not data, is what precedes corrections.