In partnership with

Startups tend to raise money in discrete increments, which have historically involved investors getting the same economic terms, and differing mostly in how big an investment they could make and in what additional privileges—board seats, information rights, etc.—they got alongside their equity. This broadly made sense in a world where VCs either couldn't or didn't do all that much for companies other than providing capital. They might vary in how much extra help they provided, but the way they monetized that was through better returns and better future dealflow.

But this was a somewhat rickety arrangement, since it assumed that the main value proposition of venture was, and would remain, the cash they provided. Giving some investors extra rights provided a little wiggle room, but it's economically suboptimal to have standardized pricing for a variable bundle of services and then to try to equalize the exchange through a limited menu of tweaks. A more sensible, straightforward approach is that the no-name fund that can't make credible promises to help the company in any way beyond cash puts in money at one valuation, and Sequoia puts in money at a lower valuation, reflecting the fact that "Sequoia-backed" is a phrase that wrenches talent loose from big tech companies, gets cold emails replied to, ensures a TechCrunch writeup (or these days, a Donald Jewkes-style launch video on X), and makes the next round's investors willing to pay a little more, etc. Some funds' dollars are worth more than other funds' dollars, and the valuations they get should reflect this.

Brand values are debatable, and they don't last forever. Sequoia remains a great firm, but a decade ago it would have been much less debatable that they were the default example of the VC firm you'd really want on your cap table. Some venture firms provide services—introductions to potential corporate partners, recruiting, PR (increasingly, full-stack marketing, general advice—but these are also hard to put a dollar value on. It's hard for startups to evaluate whether or not these will be worthwhile. In fact, it's hard for VCs to do this, too; a16z recently laid off most of the team it had doing recruiting for portfolio companies. It's worth something, but there's a market for lemons, because promising to help is incredibly easy and collecting on that promise is hard. (In fact, one of the ways a16z made this promise credible is they put people on their payroll specifically to do it.)

But today, circumstances are different. "Venture" mostly means AI, and investing in AI companies mostly means paying for some mix of inference and training, with salary, rent, snacks for the office, etc. sometimes being closer to a rounding error. (That varies a lot by company and stage; Anthropic’s spending more on personnel than inference compute right now, for example. But early on, when headcount is disproportionately founders and the company’s still exploring, compute costs can easily dominate.) There is both a lot of compute and a serious shortage of it, and some VCs have positioned themselves well to either provide it directly, because they have a cluster specifically for the use of their portfolio companies (e.g, a16z’s Oxygen program), or to provide it indirectly, because they have established relationships with cloud computing companies, neoclouds, or the big labs.

This makes their promise to help a little more legible, both as a specific commitment and as something you can put a dollar value on. Even if all they do is speed up access to compute, that's valuable because time is unusually expensive for AI companies. First, they're funded by venture capital, which is a particularly expensive kind of capital that makes any delays in launch ultimately dilutive to founders, employees, and early investors. And second, AI itself moves fast enough that a one-month delay in launching something can mean the difference between defining a category and being the first fast-follower there. (Consider how much brand value Jev) accrued from being the first decision model to launch: in the two weeks after their product was publicly announced, there were six competitors, including one from OpenAI and another from Databricks, and that timeline implies that these models already existed internally. But they all get described as being similar to Jev, which gets to be the default.

That dynamic creates some complementarity between capital-and-access providers and pure capital providers; that access is still contingent on the ability to pay, and enough backup liquidity to pay more if the initial use case turns out to be popular.

That's not a problem. It happens all the time with companies that are financed with equity as well as debt, or with companies that cut a special deal for a strategic investor, like when Facebook raised a venture round from Microsoft that included an ad tie-up and a promise from Microsoft to stop marking so many emails from Facebook as spam.

It's fine until it becomes a source of markups, especially an addictive one. If a well-connected company can instantly print a 2x or 5x or 15x markup on an investment, the danger is that they'll do that any time their IRR is lagging. And if they had some early winners, their IRR is, by its own mechanics, constantly declining when they aren't printing new gains.

So the temptation for funds, and the danger for their LPs, is that a fund can manufacture short-term gains, use these to produce a track record, and then raise off that track record before their returns mean-revert. The access funds have to compute is valuable because it's scarce, and they burn some of that scarcity every time they get a favorable valuation dependent on compute. There are many cases where economic and accounting gains and losses get booked with different timing, which can create perverse incentives. This is much more common in debt and derivatives, but the venture market is in a state where this is a persistent problem.

Tranched rounds are a pretty new topic to The Diff, but the broader idea of investor access and price-discrimination has been a common theme.

• We explicitly talked about tranched rounds last week ($), and this piece is an extension of that.

• We've looked at how companies use their logos, and those of their customers and investors.

• Founder/funder fit ($) is an important part of early-stage investing.

• We first started looking at VCs offering portfolio companies compute two years ago, when it was just one more part of the bundle ($).

Speak naturally. Send without fixing.

Wispr Flow turns your voice into clean, professional text you can send the moment you stop talking. No rough transcription you have to clean up. Actual polished text, ready for email, Slack, or any app.

Speak the way you think. Go on tangents. Change your mind mid-sentence. Flow strips the filler, fixes the grammar, and gives you text that reads like you spent five minutes writing it.

Start flowing free

Join the discussion!