Databricks wanted to raise $1 billion. Investors offered $15 billion. It closed $5 billion at a $190 billion valuation yesterday.
Funds are finishing Q2 reports right now, and anyone holding Databricks is about to tell their LPs what it was worth on June 30. That number was fine in June. It isn't now, and the LP won't know.
I lived this. My last company put $6 billion of financing through 20,000 portfolio companies, and I still made big decisions on stale numbers because the truth sat in documents nobody had time to read. I accepted it as the cost of doing business, and so does every operator I know.
The quarterly cadence itself is fine. LPs run their own books off point-in-time snapshots, boards compare managers on them, and audits depend on them. The problem is the lag between the day a round closes and the day the LP sees a number that reflects it. A round closes on a Wednesday, a charter and an SPA get executed, a new cap table lands in a folder, and then 45 to 90 days pass before that fact reaches the LP, because someone had to read the documents by hand and someone else had to type the number into a spreadsheet. The fund knows the LP is looking at a wrong picture for those weeks. It had a good reason: reading a stack of executed documents used to cost real money, and for a long time the lag was cheaper than the reading.
That's no longer the case. When a round closes, the paper shows up fast: an SPA, an amended charter, a new cap table. The SPA has the price per share, the charter has the liquidation preference, and the option pool and pro rata rights are in there too, so the number your LP wants exists in an executed document within days of the close, and the only wait was for somebody to read it.
For the last couple of years the pitch was to hand a folder of closing documents to a large language model and ask for the numbers, and everyone who tried it hit the same wall. The model would get the round size right and the price wrong, or make up a liquidation preference that wasn't in the charter, and after that happens once nobody at the fund trusts the rest of the table enough to put it in front of an LP.
Handing all of that to one big model never worked, so GoodStream doesn't. We run each agent pulling one field: the price per share off the SPA, the liquidation preference off the charter, the board changes off the voting agreement. Two models from different providers read every field, and when they disagree, a human decides with the source page right there. Every number in the result links back to the paragraph of the executed document it came from, and an auditor, an LP, or the investment committee can click it and read the same sentence the agent read.
That trace is what I demo first, because it is what makes a same-day mark defensible. A fund can move from the June 30 number to the new one the day the paper lands and show anyone who asks exactly where the new number came from, and the "where did this come from" thread that eats hours out of every quarter close never gets started.
So when the Databricks paper lands, a fund on GoodStream sees the June 30 number and the new one side by side, same day. The LP letter can still go out in October on the cadence LPs expect, and until now the habit was to hold the number until then and footnote it as a subsequent event. The fund can now update the mark the day the paper arrives and cite the paragraph, and the footnote becomes a line that says when the round closed.
LPs have started asking why they read about a round in the news before they saw it in the report. "The report closes 45 days after quarter end" was everyone's answer, and it was a fair one when reading the documents took a quarter. Quarterly reporting isn't going away. The letter that goes out in October can carry a number the fund has known, and been able to defend, since the week the SPA arrived.
