Analysis
The clearest evidence that AI mega-round pricing has broken from normal venture mechanics is showing up in the paperwork itself: rounds that carry two valuations instead of one. Newcomer reports this dual-valuation structure -- splitting a single financing into tranches priced at different marks -- has gone from a rare workaround to a standard tool this year.
Baseten's $1.5 billion Series F is the clearest example, closing across two tranches rather than one blended number:
- Tranche 1 — priced at $13 billion for earlier-committing investors
- Tranche 2 — priced at $11 billion for later capital in the same round
- Prior round — $5 billion, set just five months earlier
The mechanism exists because at the speed AI infrastructure rounds are moving, getting every investor to agree on a single price within the negotiating window has become harder than just letting early commitments lock a lower number while later capital pays a premium for the same allocation.
The pattern shows up elsewhere without the formal dual-tranche label:
- Databricks — closed its latest round at $190 billion, a six-month jump from its prior $134 billion mark, after Coatue-led investor demand reportedly pushed as high as $15 billion in fresh capital against a company that only wanted to raise $1 billion
- Cognition — the AI coding startup behind Devin is reportedly in talks to jump from a $26 billion valuation in May to $40 billion now, a roughly 54% increase in about three months
What this tells founders and GPs: valuation is becoming less of a single negotiated number and more of a range investors are willing to accept just to secure allocation in a scarce set of companies. That's good for the specific founders raising these rounds -- less time spent negotiating a single price, more capital certainty -- but it's a warning sign for anyone marking portfolio companies to the most recent private round, since the 'real' price of these deals is now genuinely ambiguous even to the people writing the checks.
