Eclipse Ventures realized $2.5 billion on Cerebras. That's a real outcome for the firm and its LPs, and I expect to see it on the second slide of their fundraise deck for the next two years. What I keep thinking about is what sat underneath that number when the LPs started asking follow-up questions: the transaction-level ledger, the round-by-round cap table history, every mark the position carried over its life, and a trail from each of those back to a signed document. Eclipse could answer them. LPs always have follow-ups once a headline number lands.
A decade ago the Cerebras number would have been the whole deck. You put the exit up front and the rest of the meeting was pleasant. Now the exit gets you into the room, and once you are in it the LP wants to know whether the other 40 positions in the portfolio get run with the same discipline as the one that worked, whether the marks on the long tail get defended the same way, and whether you can produce the Cerebras story for any other name in the book, on request, with the source documents attached.
LPs got burned on the 2021 vintage and they haven't forgotten. Comps were stretched and reserve strategies were optimistic, and the discount-to-public assumptions didn't make it through 2022. When LPs went back and checked what they had been told against what they could verify, the gap was uncomfortable. Some funds explained it. Others didn't, and they are still paying for it in their current raise.
So the LPs writing checks to VC funds in 2026 are evaluating the system that produced the deck, and they have gotten good at it. If your answer to "show me DPI by position, by sector, by quarter" is a spreadsheet your associate updated the night before the call, you are losing to a GP who can pull it up live. Both of you have the same data. The LP cannot see which of you has a working portfolio system, but they can tell which fund gives the same answer in March that it gave in December.
A few things follow from that, and none of them are about Cerebras.
A cap table you rebuild from the documents every time a round closes beats a mark you revisit at quarter end. A firm that keeps its ledger current answers the March question from the same picture it used in December, and a firm on a quarterly cycle rebuilds the picture from scratch each time, and the LP can feel that lag on the call even when nobody names it.
The mark history is now part of the track record. LPs have started diligencing the data layer the way they diligence the IC process, because the committee decides the marks and the data feeds the committee. If the data is a spreadsheet, the LP has a fair question about what the committee was looking at when it voted, and the answer to that question is going to be in the deck for the next fund whether the GP puts it there or not.
And with $106 billion of US direct secondaries in 2025, a fund has to be able to produce a mark whenever a bid comes in, on the bidder's calendar rather than the fund's, and the quarter close has nothing to do with it. A fund that can put a defensible mark together on five days' notice, with the SPA and the latest charter behind it, gets to set the price in that flow, and a fund that takes the price it is given will see the difference in returns within two vintages.
Eclipse will get the credit for Cerebras, and they earned it. The firms that produce their own version of it will be the ones that had the ledger, the mark history, and the document trail in place before any headline needed them, which is the plumbing we spend our days on at GoodStream.
What I would take from the Cerebras story, if I were raising this year, is that the next LP who sees the $2.5 billion is going to ask about the rest of the book, and the answer has to come out of the documents, position by position, the day they ask.
