Ramtin Naimi on the State of Venture Investing - [Invest Like the Best, EP.499]
Ramtin NaimiPatrick O'Shaughnessy
- Naimi rates venture’s opportunity set 10/10 and froth 6.5–7/10, versus 2021’s 10/10. Faster adoption and product-market fit support the enthusiasm, but financing behavior is familiar: investors who miss the primary round may pay 3–5× its valuation in a simultaneous secondary tranche. “Markups just feel like you're keeping pace with the state of the market.”
- AI’s headline growth rates conceal businesses with very different revenue quality and margins. Growth of 10–20× can involve customer concentration, usage revenue, annualized data sales or negative margins. The upside is simultaneous margin expansion; the unresolved valuation question is whether vertical AI deserves software multiples or access to labor budgets. Naimi’s caveat: “I haven't seen them actually start to replace labor yet.”
- Accelerated fund deployment is starting to delay the cash behind signed company financings. Mega-funds’ deployment periods are about three times faster than expected, and they have raised 2–3× as much in the same 18–24-month window. LP liquidity depends heavily on a few holdings. Naimi has seen a $150 million commitment require three months to fund: investors are “committing to deals before they even have the money to finance them.”
- Compute demand’s financial constraint may emerge through debt rather than equity. Naimi estimates that 80–90 Neo Labs have raised amounts in the $1–5 billion range. One was raising $1 billion for a training run and thought it would probably need another $5 billion the following year. Infrastructure builders need debt exceeding 5–10× their equity, he thinks; the question becomes financing terms that preserve “business viability.”
- The strongest seed opportunity is exceptional founders whose markets have not become obvious. Naimi distinguishes reactive investing, which he thinks can deliver top-quartile results, from proactive investing with top-5% potential. Conduit’s BCI work and Etched’s chips illustrate “great founders working on non-obvious ideas”; meanwhile, trillion-dollar outcome assumptions are lifting entry prices across the market.
- Buying established winners can secure exposure without securing exceptional returns. Patrick proposes shifting toward businesses such as Databricks that investors can actually investigate. Naimi counters with Stripe’s move from $100 billion to $179 billion over five years—an estimated 11–12% IRR—and worries that access-driven growth investing has become consensus: “if everybody believes that's an easy way to make money, it's probably not a great way to make money.”
- Abstract’s larger fund reflects the cost of owning more of its winners. Its first fund was $100 million in 2018; its latest is $800 million in 2026. Naimi says the biggest return contributors were never allocation constrained: “We were capital-constrained.” His counterfactual is that each fund could have deployed approximately twice the capital with an identical return profile.
1. Financing speed is familiar; the underlying opportunity is different
Naimi describes preferred investors receiving the primary allocation while losing bidders pay 3–5× its valuation simultaneously, versus 2021’s 2× fast-follow rounds weeks later. Silicon Valley calls this the “cuck tranche.” He considers tranching a necessity for capital-intensive companies rather than, by itself, an unhealthy market signal.
The resemblance to 2021 includes rushed offers and renegotiations: he has seen more deals reworked in the preceding six to eight weeks than at any point since then. But adoption is different, including among his own family. In 2021, he says, companies with $8 million in SaaS revenue and constrained markets could raise at billion-dollar valuations, with a seven-year path to $100 million in ARR. Round structures, deal volume and pressure to keep up feel similar now, but he sees more compelling companies. His timing caveat matters: “we could be in 2019 and not 2021.”
Revenue growing 10–20× rarely resembles clean, contracted, 90%-plus-margin SaaS, Naimi argues. It can carry low margins, customer concentration, consumer or usage exposure, or last month’s data sales multiplied by 12. Data-labeling margins can be 15–20%. Founders should compare themselves with similar businesses rather than the largest headline growth figures. He also notes that some margins are improving rapidly.
He expects portfolio companies to grow 200%–700% this year while margins move from –15% to +35%. Vertical businesses fetching 100× revenue versus public SaaS at 13× require another argument: law firms spend far more on paralegals than on software, so capturing 20% of paralegal spending could make the math work. Yet “I haven't seen them actually start to replace labor yet.”
2. Faster deployment is exposing the limits of fund commitments
Funds normally plan initial deployment over 18–24 months. Naimi says mega-platform funds have raised 2–3× as much in that same window, with deployment periods about three times faster than expected. They underestimated the number of companies they wanted, their prices and the capital needed to get their desired ownership. Without more capital, they must accept less ownership and capped upside, or fewer “shots on goal”—which requires being a better picker.
Naimi says some LPs had 35–50% of their exposure in venture; on closer inspection, he describes 30–50% of their AUM as effectively concentrated in three or four companies, including SpaceX, Anthropic, OpenAI and Databricks. Liquidity from those positions matters disproportionately. LPs without exposure to funds holding those companies are receiving little cash back, pushing managers toward new capital sources.
His unnamed examples make the acceleration concrete: a $2.5 billion fund began deploying in July and expected to return to market in January; another received $3 billion nine months earlier and was seeking $5 billion. His assessment of those managers remains categorical: “They’ll get it.”
Company funding is now affected. One $150 million term sheet required waiting three months until mid-January for a new fund’s closing and facilities. Another investor committed $40 million of a $50 million round but could wire only $5 million at closing, with the remainder eight to ten weeks later. Naimi expects the money to arrive; the issue is deployment speed.
3. Compute demand needs financing that can survive its capital intensity
Patrick’s challenge starts with Anthropic’s S-1: he cites $140 billion raised, $10 billion spent, $132 billion on the balance sheet and 85% gross margins, while setting aside future commitments and liabilities. Does compute require new financial products? General Catalyst’s Customer Value Fund offers a precedent; Naimi describes financing growth if $10 million can demonstrably unlock $30 million of revenue.
Some investors already supply compute instead of cash, which founders tell Naimi can guarantee an allocation. But pricing, depreciation, reuse and whether access covers one training run or continues indefinitely complicate the exchange. Founders generally prefer clean equity: “we know what we gave away and what we got in return.” Naimi says new structures also face adoption hurdles: founders often rely on VCs, who historically discourage venture debt, and the Customer Value Fund required significant founder education in its first couple of years.
Large neoclouds often will not engage below $200 million, leaving smaller companies assembling capacity piecemeal. Hyperscalers and frontier labs dominate, but Naimi estimates there are 80–90 Neo Labs that have raised amounts in the $1–5 billion range. One was raising $1 billion for a single training run the following year and thought it would probably need another $5 billion that year.
Impressive launches may reflect “benchmark-maxing” rather than generalizable commercial advances; Naimi cannot tell whether all, half or only 10% of these labs will keep raising the capital they need. Asked whether capital could constrain compute, he answers “Potentially,” emphasizing debt. He thinks infrastructure builders need debt exceeding 5–10× the equity they raised, making rising borrowing costs a question of continued business viability.
4. Seed alpha comes before the market agrees on the category
Naimi tracks terminal-value goalposts from $10 billion in 2016 to $100 billion around 2021 and now $1 trillion. Prices move accordingly, although only a handful of companies were on track to reach $100 billion in the prior cycle. His current bet is that there “might be 2 or 3 more” trillion-dollar businesses, “not 40 more.”
His distinction is reactive investing for potentially top-quartile results versus proactive investing for top-5% results: “great founders working on non-obvious ideas.” Patrick asks whether two, three or four term sheets are preferable to 15. Naimi says fewer bidders can mean more reasonable pricing and greater leverage, while preserving the caveat that most opportunities attracting little investor interest simply are not compelling.
Naimi seeded Conduit nine months earlier, when BCI attracted little interest. Unlike Neuralink’s implanted-chip approach, Conduit extracts brain information through a device worn on the head. Its model showed 80% accuracy in understanding a person’s thoughts before they typed them; the wording differed, but the end result was the same. It now has 30,000 hours of human training data, and the category’s recent popularity brought 13 term sheets. If Conduit succeeds in doing a million-hour training-data run and its scaling laws hold, Naimi imagines an ear-worn device directing agents from actionable thoughts in five to ten years.
Etched supplies the parallel: investors considered Gavin Uberti brilliant but doubted a startup could challenge incumbent AI-chip companies. After Sequoia backed it at a $10 billion valuation, enthusiasm spread across the category. Naimi’s point: “the market was obvious to these founders before it was obvious to the general market.”
5. Proven winners still leave investors exposed to entry price
Pricing has become so erratic that Naimi’s plausible valuation range can span 4×. One bidder may pay triple the next-highest offer; that outlier becomes another founder’s comparable. An offer an investor considers generous can therefore offend its recipient.
Patrick’s pushback is to buy businesses such as Databricks, where investors can investigate actual operations, rather than play a “weird multiplayer game of pricing, totally untethered.” Naimi agrees this offers passive exposure to great companies with less competition. The main question is whether the pricing environment is more like 2019 or 2021.
Stripe illustrates the limitation: $100 billion in 2021 versus $179 billion today implies roughly 11–12% IRR over five years by Naimi’s estimate, potentially still top-quartile venture performance. But he has watched firms converge on access to obvious winners as a fundraising strategy over 18 months. “I can't really think of many times in history when an easy strategy also turned out to be a good strategy long term.”
6. Data businesses contain two different visions of durability
Patrick challenges the old characterization of data suppliers as a trade rather than a company. Naimi says model builders’ needs were dramatically underestimated. One robotics-data camp believes demand can support another 10, 15 or 20 years of selling data and “tens of billions of dollars in revenue.”
The other camp believes data sales have an eventual ceiling but can finance proprietary robotics-model research today. These companies know how to collect valuable data and sell it to other model builders while developing their own models. Naimi stresses that both camps understand the category deeply despite disagreeing on the endpoint.
7. Abstract is scaling around relationships and capital constraints
Abstract now leads Series Bs in Socket, Doctronic and Dust, alongside passive growth investments such as Crusoe. About 15% of its 2023 vintage went into xAI’s formation round; it was also a big investor in Polymarket’s Series B. Its $100 million first fund looked oversized to LPs in 2018; its $800 million fund earns praise for staying small in 2026.
In almost all the investments producing the lion’s share of returns, “we were never allocation constrained. We were capital constrained.” Naimi calls undersized funds his biggest mistake and believes every fund could have maintained an identical return profile with probably twice the capital, while leading more portfolio-company financings.
Trusted relationships allow Abstract to approach founders outside formal processes and offer a number they do not feel they need to shop. Taking enough of a round that founders need only ask other investors for their pro rata can help Abstract build its positions. Future fund size follows a reassessment roughly six months before fundraising: can the firm deploy more capital into equally good or better companies without compromising returns?
Asked about employees who may become wealthy through IPOs of the biggest companies, Naimi welcomes more $100,000–$1 million personal checks before institutional funding. He estimates Silicon Valley’s density of such investors is probably 100× or greater than elsewhere. Their willingness to lose that money helps explain the region’s startup advantage.
8. The next bellwethers extend beyond the obvious IPO candidates
Naimi scores the opportunity set 10/10 across nearly ten years in venture: Abstract has never found so many companies it wants to back. Froth is 6.5–7 versus 2021’s 10 because rapid growth, adoption and product-market fit provide substantial justification.
His biotech watchlist includes NewLimit, Chai Discovery and Isomorphic Labs: “can you build an AI-native, next-gen Genentech?” He sees $500 billion to $1 trillion outcomes, potentially larger if their ambitions work, supported by teams, investors and access to capital.
He also highlights Applied Compute’s idea of owning your data and knowledge intelligence, prediction markets as rapidly growing consumer businesses that are not AI companies, and Physical Intelligence as a potential multi-hundred-billion-dollar or trillion-dollar company. Talent retention is his recurring filter: exceptional employees can obtain richer offers elsewhere, so high talent density coupled with retention makes him “extra bullish on those companies.”
Full transcript
One of our mutual LPs describes you as the person he learns the most per minute from about what's going on in venture markets. We're going to try to recreate that here today and focus less on the big-picture stuff—the history of Abstract and so on, which we covered the first time—and more on this moment in time. How would you describe the state of things right now, zoomed all the way in to the deal level here around San Francisco?
It's very fast-paced, insanely fast-paced, in the sense that rounds are happening quicker than ever. Follow-on rounds are happening quicker than ever. There are certain companies that get a lot of attention very early in their life cycle, and they get a lot of demand for their financing, whether it's their seed, Series A, or Series B. There are a handful of firms those founders want to raise capital from, and those firms get preferential pricing, which is really tranche number one.
There are a bunch of VC firms that lose that deal. Historically, the VC firms that lost the deal would say, “Okay, we'll maintain a relationship. We'll put in a small check or something to maintain that relationship, and as the thing performs, we'll do the next round.” In 2021, you had the fast follow, where a couple of weeks later, they would raise at 2x the valuation from one of the people who lost that financing. But in this market, it's like, “Okay, we lost this round, but we don't want to wait for the next round. What do we need to pay to get into this financing today?” It's usually not 2x. It's usually 3x, 4x, or 5x, happening in tandem with the primary.
That's the secondary tranche, which has become known elegantly in Silicon Valley as the “cuck tranche.” The interesting thing is these companies are so capital intensive that they actually need that level of capital, and they don't want to raise all of it in the primary tranche. So I don't think tranching is an unhealthy sign of the market. I think it's actually a necessity for these types of companies and the environment we're in. Sometimes you get into the early tranche, other people get into the second tranche, and sometimes you're the one getting cucked. I think anybody who tells you they've never gotten cucked is completely full of shit. Everyone's gotten cucked once or twice, at least. Anyone who says otherwise is lying to you.
Things like that are happening. You're also having people running around like chickens with their heads cut off, trying to keep up with the market and throwing out offers very quickly. And then you're seeing some bad behavior, where people are changing those offers quickly after the fact or pulling them. A lot of things are somewhat reminiscent of 2021 in that sense. But the pace at which things are moving also makes everyone feel like they need to do something to stay ahead of the curve.
In this market, getting markups is the bare minimum requirement to keep up, to the point where markups don't even feel good anymore. They just feel like you're keeping pace with the state of the market. Once every 3 weeks, you have something happen in your portfolio that makes you feel pretty good, and you're like, “Okay, cool. I'm doing really, really well in this market.” Then, in the subsequent 3 weeks, you see things happen to companies you're not in, and you're like, “Holy shit, I'm falling behind once again.” You see the pace at which things are happening. You're seeing exits that are the largest in history, and then it's news for 2 days and people move on to the next thing.
I'd say the pace at which things are happening is not unprecedented, because I think a similar pace happened in 2021. I think the pace is healthier today because there's real value creation. These companies are doing truly amazing things. I do think the thing that's a little misleading about this market is the story behind the revenue at all these companies. You do have companies growing 10, 15, 20x year over year, but I think any company growing at that rate has something interesting about its revenue. It's not clear, clean, pure-play SaaS revenue, with 90%-plus gross margins and annualized recurring revenue contracts.
They all have something that's either ridiculously high growth but very low margins, or low or negative margins; ridiculously fast growth but really high customer concentration; or annualized revenue because they're selling data, and whatever their revenue was last month, they multiply by 12 and that's their annualized revenue. Or it's consumer-based revenue and not enterprise-based revenue, or it's usage-based revenue and not contractual revenue. There's a lot of that, and I think it could be somewhat discouraging to founders whose companies aren't growing at that pace when they see those numbers and think, “Oh my God, I'm completely left behind.” But if you peel back the layers and start looking at it, you need to compare your revenue growth to companies within your style of company, rather than the absolute top-line figures.
You have data-labeling companies growing at insane rates, but their margins are 15% to 20% because getting that data is expensive. You have other companies reliant on foundation model providers to power their products, and all of the margin goes to the foundation model companies. But the cool thing you're seeing is that these margins are improving at really rapid clips.
Anthropic was at 85% in the first half of this year, they said.
We have companies in our portfolio that are going to grow 200% to 700% this year, but their margin profiles are going from negative 15% to positive 35%. So you're getting massive revenue growth and margin growth simultaneously. Then you have other companies that are vertical SaaS businesses getting 100x revenue multiples. People look at that and say, “Well, how do you compare that to the public markets when the SaaS multiples publicly are at 13x revenue?” And you look at these and you're like, “Well, are these actually SaaS businesses, or are these labor-replacement businesses?”
Some people don't buy the labor-replacement thesis, but the way you actually have to believe in it is that these companies are still growing 5x, 7x, 10x year over year, even at the massive scale they're already at. I haven't seen them actually start to replace labor yet, but the pace at which they're growing... If you look at companies that sell to law firms, you look at the total amount of money law firms spend on software every year, and you're like, “Oh, these companies can't be that big, actually.” But you look at how much money law firms spend on paralegals every year, and you're like, “Oh, that's actually a really big number.” So if these companies can capture 20% of that, the math—
Works, the math works.
The math starts to really, really work.
If you were to compare this to 2021, which you lived through and were very active in, obviously two very euphoric, frothy-seeming markets, what are the most notable differences and similarities?
If you look back to 2021, the companies weren't that interesting when you looked at them. It's just that we didn't have anything else to compare them to. You were looking at SaaS companies growing 5x, 3x, 3x, 2x, and everyone's like, “That's best in class.” You would have companies with $8 million of SaaS revenue raising at billion-dollar valuations, and they had very constrained TAMs. You kind of had to look at these things on a 7-year timeline to get to $100 million of ARR. That was what venture was back then, and that was exciting.
Now you look at companies today and you're like, “I cannot believe we even got out of bed for those types of opportunities back then.” In terms of the round structures, the volume of deals, and people feeling like they need to keep up with the market, it's all very similar.
In 2021, there was a six-month window where, if you asked pretty much every VC and even LPs about the situation, VCs were just ripping checks. If you ask any of those VCs to look back at their portfolios for that six-month window in 2021, a lot of them are like, “Yeah, we scratch our heads and think, ‘What were we thinking, doing all of those deals?’”
I’m not saying we’re at that point in the market today. We could be in 2019 and not 2021, but some of the things I’m starting to see happen today that happened in 2021—like public market investors trying to aggressively come into venture—are very actively happening right now. You can see how frantically people were trying to do deals back then versus how they’re trying to do them now.
Some of the things we started seeing near the latter part of 2021 were companies getting term sheets very, very quickly, and then those term sheets getting pulled, reneged on, or renegotiated very, very quickly. In the last 6, 7, 8 weeks, I’ve probably seen a larger number of deals get renegotiated after being offered than I have since 2021.
The thing that’s most dissimilar is the enthusiasm around the companies. I don’t think people were investing in companies in 2021 thinking these companies were going to be category-defining or global-scale companies within 18 to 24 months. You’re seeing that play out in this market. Companies are finding product-market fit at a faster rate than ever. I’m looking at companies in my portfolio that I’m using very actively, and I can’t think of anything I invested in in 2021 that I was using personally. Now I’m starting to see how much value these things can drive.
Family members are using some of these products who would never be early adopters of AI products or tech products, period, and these products have become integrated into their daily lives. I think there’s a huge paradigm shift in the technological innovation in this market. People truly believe that if there’s a window of opportunity where you can’t afford to sit on the sidelines, it’s this moment in time.
Can you drill a level further down the stack, to the LP side and the funds that are being raised? This money comes from funds going into these companies at record paces and amounts. VCs also have to raise money themselves from a group of LPs that have historically been the reason venture capital exists: endowments, foundations, and things like this. At some point, it seems like we’re all spending so much money so quickly that the sources of capital themselves are starting to ask deeper questions.
1. Funds Deploy Faster Than Expected
When a venture capital fund goes out to raise capital, they’re raising for an initial deployment period of 18 to 24 months. That’s the amount of cash they want to deploy into initial checks, and then they reserve whatever they reserve for follow-ons to keep financing those companies and maintain their ownership.
I think in the last 18 to 24 months, the mega-platform funds you hear about that are raising these record-size funds have probably raised 2 to 3 times as much in that same 18-to-24-month window. Their deployment periods are about 3 times faster than they thought they would be.
What’s interesting is that when venture capital firms go out to raise, these are the guys who have their pulse on the market better than anybody. They always raise an amount of capital that supports their portfolio construction model, which means they want X number of bets, with X amount of ownership over X period of time, typically 18 to 24 months. Every one of the mega firms has dramatically underestimated how much capital they need to get that level of ownership in those companies over that window of time. That basically means there are more companies they want to invest in than they thought there would be. The price of those companies is higher than they thought it would be, and to get the ownership they want, they need to invest much more than they initially modeled.
That’s why you’re seeing these guys not only come back to market sooner than they initially anticipated, but also come back for much larger sums of capital. The most interesting thing you’re seeing in this market is that LPs are looking at these guys and realizing that one of two things is happening: they either maintain their ownership in these companies but end up with way fewer shots on goal, or they end up with way less ownership in these companies and keep the same number of shots on goal.
Shots on goal are basically your hedge against not being a great picker. You need enough shots on goal to have the number of winners you need for a good venture capital fund. But if you’re not getting your ownership, your upside will be capped in one way or another. The alternative is to say, “Hey, we actually need our ownership, so we’re going to invest in fewer companies this vintage because we have to invest more capital in the companies we want to invest in to get our own ownership.” In that case, you need to be a better picker, but you could still produce outsized returns because you’re getting core positions in the companies you want core exposure to.
From an LP perspective, I think those guys were expecting dramatically more liquidity at this point in the cycle than they’ve actually gotten. It’s funny, because you look at LPs that had 35% to 50% of their exposure in venture capital. But if you peel back a layer, it’s not 30% to 50% of their AUM in venture capital; it’s 30% to 50% of their AUM in 3 to 4 companies, like SpaceX, Anthropic, OpenAI, and Databricks. Those are the names that need to generate liquidity for these guys to get their exposure levels to the asset class down.
It doesn’t actually have to do with all of these companies you’re hearing about. There are just a few companies that really move the needle for all of these LPs. If you’re an LP that wasn’t in a fund with a lot of exposure to SpaceX, you haven’t gotten any liquidity. Basically, the LPs that are going to generate meaningful liquidity are the ones with exposure to funds that have meaningful exposure to those companies. Outside of exposure to those companies, LPs aren’t getting much liquidity back at all.
I think that puts the funds whose LPs don’t have exposure to those companies in a position where they’re trying to extract new sources of capital for their companies, because a lot of the traditional LPs—the ones you and I would think of as the people who fund the venture capital ecosystem—are limited on cash.
Could you give an example of a fund, without naming it, and some rough numbers—not specific—of its history and accelerated pace of deployment?
I know one fund that raised $2.5 billion, started deploying in July, and told me they’re going to be back in the market in January. I know of another fund where my LPs told me they gave them $3 billion 9 months ago, and now they’re back looking for $5 billion. They’ll get it. These guys are all great managers, and they’ll get the numbers they’re looking for, but that’s the intensity of the pace.
The other thing that’s really interesting about this market is that these guys not only need more money quicker than they thought they would; now it’s starting to have ripple effects on the companies getting their term sheets, in terms of when they actually get the capital they were committed to. I’m seeing companies in the last 8 to 12 weeks—and this is the first time I’ve started to see this in this cycle—where the VCs are giving them term sheets and saying, “Hey, we want to put $150 million in your company. Is it okay if we fund it in mid-January, which is 3 months from now?”
The founder’s like, “Wait a second. You need 3 months to fund?” And they’re like, “No, no, no, we’re good for it.” And they are. All the money in our current fund is already committed to existing companies, and we’re in the middle of closing a new fund. But by the time we get all those documents in and onboarded, and we get our capital—
Capital. Oh, yeah.
—facility set up, it’s going to be mid-January before we fund. We had another company that got a $50 million term sheet from one VC firm, and the VC was going to do $40 million of the $50 million. They said they could wire $5 million upon closing, and they needed another 8 to 10 weeks to fund the remainder of their commitment.
That’s kind of interesting: these guys are committing to deals before they even have the money to finance them. They’ll get the money, but it draws attention to the pace at which these things are really getting deployed.
2. Financing the Compute Buildout
To step all the way back, this seems furious and unsustainable, and maybe understandable given that it’s funding the most interesting technology buildout any of us will ever see—or in human history, or whatever. But the Anthropic S-1 came out, which makes it an interesting day to be having this discussion. The most glaring stat in that thing is that they’ve raised $140 billion, spent $10 billion of that, and have $132 billion on the balance sheet.
That’s pretty amazing—
Mm-hmm.
—that they’ve burned $10 billion of cash. Forget about all the future commitments, the liabilities, and all these things for just a moment, but $10 billion of cash to get to the revenue numbers they’re at with 85% gross margins is unbelievable. Does all this make you wonder if we need some innovation in financial products? It seems like when you boil down why everyone needs all this stuff, the answer is always just compute. It makes me wonder: are we trying to solve a very modern problem with what’s become an antiquated instrument or style, given this psychotic behavior?
Yes and no. There’s no shortage of new debt vehicles to finance growth in companies. I think General Catalyst has their—
Yeah, their Customer Value—
Customer Value Fund. Exactly. It’s like, if you could prove that for every $10 million it unlocks $30 million of revenue, we’ll give you the money to finance that.
I guess that makes my point, though. That thing’s been, from a product perspective—
Yeah.
—enormously successful.
Mm-hmm.
They’ve raised tons of money. If you just think of it as a product, it’s done really well. What I’m wondering is, do we need way more things like that for each type of spending that’s happening in these companies, lots of which is panic buying or reserving compute ahead of anticipated demand? Some of these personal agent companies are thinking about how much money they’re going to need to spend to grow to the daily active user counts they need.
I think there are vehicles that are emerging. There are some VC firms that are financing companies with compute as opposed to cash. For a lot of the founders we talk to, that’s actually more attractive. They’re like, “Hey, if you bring us compute instead of cash, you’ve guaranteed your allocation this round, because we’re going to take this money and spend it on compute regardless.”
I think some people are doing that, but then you have to go secure the compute and make sure you’re getting good pricing on it. It’s a little complicated: do you give them the compute in perpetuity? Do you give them the compute for one training run? How many times can you recycle that same amount of compute to make up for the amount of capital you’ve deployed? Can you recycle the amount of compute you have three times, two times, or one time? How much do these things actually get degraded over time? I think there are a lot of complexities in that.
I haven’t seen much else. Getting new capital structures widely adopted in Silicon Valley is complex and challenging, because VCs, by default, like their companies to have clean equity that doesn’t have unique covenants that can put the company in a tricky spot if they don’t think about them. By default, most founders are like, “We just want a clean equity deal where we know what we gave away and what we got in return.”
General Catalyst’s Customer Value Fund is doing great now, but I think in its first couple of years, there was a very significant education process around why this is a good product and why it should work for you. Founders usually rely on their VCs for opinions on anything capital-related, and VCs historically don’t like venture debt products. So they give you a million reasons why you shouldn’t do it. Then the other side needs to give you a million reasons why you should, because it does make sense. It’s a new product created for this day and age of venture capital. I think there’s always room for innovation in capital structures to fund a new paradigm of companies that need billions of dollars in equity capital to spend on compute. That might not be the best use of equity capital.
We were just talking about the weird nature of the market, where, on the one hand, you’ve got Elon saying there’s going to be all this stranded capacity in compute, and neoclouds are having quite a tough week in the markets. At the same time, literally nobody can seem to get allocations for the prior generation, let alone the current generation—Vera Rubin or something—from NVIDIA, and the price of GPUs just looks unidirectional. It seems to me like a very bizarre moment in the market.
The thing that we’re seeing in our portfolio companies is that we’re routinely helping them find compute, because finding compute for them is not easy. A lot of the largest neoclouds, and we have relationships with a number of them, aren’t really interested in engaging with compute deals that are less than $200 million in scale. So companies that don’t have that level of scale are trying to lease some capacity here, some capacity there. But those guys, on a relative basis, don’t actually make up for the scale of this market.
The people who are actually moving the needle on how much compute is necessary in this ecosystem are the hyperscalers, the frontier model labs, and the companies that are private Neo Labs raising $1 billion-plus seed valuations, effectively. I think the only thing that limits demand for compute over time is if these companies’ compute needs diminish, or if the companies that have these compute needs have a harder time raising capital. There’s a subset of companies that you and I know are going to continue raising because they’re the best assets, and they’ll have all the cash they need to buy all the compute they need.
I think the most interesting segment of the market, which on a relative basis is still small compared to the hyperscalers—but if you aggregate all of them together, I don’t think the scale is that small—is the Neo Lab segment. People started looking at Neo Labs thinking there were going to be two or three great ones, and now there are probably 80 or 90 Neo Labs that have raised somewhere between $1 billion and $5 billion. The funniest thing with them is that they all need to raise tons of cash.
We met a Neo Lab a couple of weeks ago that was raising $1 billion. I think this was their second round of financing, and they were going to spend the entire billion next year on compute for one training run. They said they would probably need to raise another $5 billion throughout next year. The tricky thing with these companies is it’s very hard to underwrite how well they’re doing.
The funniest thing with Neo Labs is that you can’t underwrite how well they’re doing or how much money they’re raising, but oftentimes they’re launching things that are very impressive. Oftentimes it’s hard to discern whether they’re launching something that’s truly impressive on a generalizable scale, or something specifically tuned to perform really well on benchmarks. That’s the whole concept of benchmark-maxing: are these things truly groundbreaking and innovative, or did they do something that shows really well on benchmarks but isn’t really commercializable yet, and won’t be commercializable in the near future? I think there are a lot of those companies with massive capital needs, and it’ll be interesting to see if all of them continue to raise, if 50% of them continue to raise the capital they need, or if it’s 10%.
Could it just be that capital is the actual constraint? If the capital markets dry up, is that what causes the compute story to change? That’s it?
Potentially, and I think it’s more on the debt side than the equity side. A lot of these guys building these things need debt in excess of 5–10× the equity they raised. Given what happened in public markets with some of the neoclouds in the past couple of weeks, I think their cost of debt is going to increase.
So I think it’s less about whether these companies are able to raise the equity they need to produce this compute for the demand. It’s going to be more about whether they could raise the debt they need at terms that make sense for business viability, for the show to go on.
What about seed as a concept? Some seed deals get done at crazy prices and don’t really feel like seed. They’re seed risk for Series D pricing or something like that.
Yeah.
Of course, the promise that these new markets are bigger than ever and everything can be a $100 billion company now—
Yeah.
—is part of every single pitch. How do you feel about seed, given that’s where you’ve spent most of your career?
Every few years in Silicon Valley, the goalpost moves for what terminal outcomes can actually be. When I got started in early-stage investing in 2016, people looked at every company through the lens of, “In a best-case scenario, this could be a $10 billion outcome,” and priced deals accordingly. Then you started getting toward 2021, when Stripe, Databricks, Airbnb, and DoorDash were approaching $100 billion valuations. Everybody started looking at companies through the lens of, “In a best-case scenario, this company can be worth $100 billion,” and started pricing deals accordingly.
It turns out the companies that were on track to be $100 billion companies in 2021, give or take 2 or 3 more, ended up being the only companies worth $100 billion. But every other company started getting priced as if that was the potential outcome for them. Now we’re at a new goalpost: “In a best-case scenario, this company could have a trillion-dollar outcome.” Between SpaceX, Anthropic, and OpenAI, you have 3 trillion-dollar companies. My bet is there might be 2 or 3 more, not 40 more. But now everyone’s looking at companies through the lens of a best-case scenario being a trillion-dollar outcome. So what is a $1 billion entry valuation in a trillion-dollar-outcome scenario? I think that’s what ends up shifting the dynamics of what these valuations are.
But if you look at every stage in venture, there are always opportunities to get good pricing on great companies. Generally speaking, the investors that don’t do great in bad vintages are the ones that deploy the majority of their capital into the most obvious thing. Venture capitalists who deployed the majority of their capital into the most obvious thing in 2021 were doing 100x revenue-multiple SaaS deals with pretty much their entire portfolio. They’re probably now in the bottom quartile, or bottom 75 percent, of venture capital funds for that vintage.
Generally speaking, if you’re at the early stages and backing great founders working on obviously good ideas, you’re going to pay sky-high market prices, especially in frothy markets. But if you back great founders working on non-obvious ideas, that tends to be the best way to generate alpha in markets like this. If you’re truly backing the best founders, they’re not working on bad markets; they’re usually just early to markets.
I think there are 2 ways to make returns in venture capital: being reactive and being proactive. Being reactive, if you pick correctly, allows you to be a top-quartile manager. Being proactive allows you to be a top 5 percent manager. Being proactive means trying to back great founders working in markets that aren’t yet obvious. If the ideas aren’t yet obvious, that means you won’t get a lot of competition, and you can price these deals accordingly.
A couple of examples: We seeded a company about 9 months ago called Conduit, which is working on BCI, or brain-computer interfaces.
Really interesting people.
Really interesting people. Anybody who met those founders would have told you they were brilliant. But 9 months ago, nobody in the Valley really cared about BCI as a category. What these guys are doing that’s super interesting is different from Neuralink, which implants chips into your brain and gets perfect readouts as a result. These guys are extracting brain information via a product you wear on top of your head.
When I met Conduit, they had a small amount of human training data. By the way, their model was showing scaling data: the more data it gets, the more accurately it understands your thoughts. It showed that if somebody typed their thoughts out, before they did, their device, software, and model could understand what they were thinking with 80 percent accuracy. If you read what the model thought they were thinking versus what they actually typed out, you got the same end result, but the words were different. Now they have 30,000 hours of human training data.
In the last 8 to 10 weeks in Silicon Valley, BCI became a very popular category. As you know, Conduit raised a very large, highly oversubscribed round with 13 term sheets, because now it’s a great team working on a market that everyone cares about. If these guys succeed and go do a million-hour training-data run, and their scaling laws hold, you could see a world where, 5 or 10 years from now, you and I have a little device that we wear over our ear. It understands our thoughts, communicates with our agent, and can discern between actionable and non-actionable thoughts. Everything you’re thinking about, your agent is just running for you on the back end. You just became the most efficient human being in the world.
Now everyone’s thinking, “Wait a second, somebody could actually build a trillion-dollar-outcome company in BCI.” Conduit’s latest round prices that in. It’s a perfect example of backing a brilliant founder in a market people might not be that interested in at the time. They were just early to a market, because smart people generally understand markets well before the market starts to pay attention to them.
You have a perfect example with Gavin Uberti from Etched. Gavin met every great investor in Silicon Valley, and everybody walked away from the meeting thinking he was brilliant. I don’t think there’s a single VC who met Gavin who didn’t think he was brilliant. But a lot of people just assumed a startup company couldn’t outcompete incumbents in AI chip fabrication. That was never going to happen. It wasn’t until Sequoia blessed him with a $10 billion valuation that everyone said, “Oh, wait a second, now we actually believe startups will run circles around the incumbents in this market.” Then you saw every other chip company in the market raise at an astronomical valuation, because now people thought, “This is an obvious market to place bets in.” But the market was obvious to these founders before it was obvious to the general market.
If I were to shrink all of those observations down to one metric, would it be that there’s an ideal number of term sheets? I know Gavin’s story. It’s not that he got no term sheets; he got a handful. He met every VC and got a small handful of—
Yeah.
—term sheets from great investors. Is that the right way to think about it? What you want is for the companies you’re backing to get 2 or 3 or 4, not 15?
If a company gets 15 term sheets, it’s going to get very, very expensive. If a company gets 2 or 3 term sheets, you’ll probably get reasonable pricing because you have less competition. As a great investor, you have a lot more of a competitive advantage in convincing that founder to take your money and getting a core position in that business.
If a lot of people take a meeting with a company, they’re generally taking it because the founder is very impressive or they’ve heard great things about that founder from people in their network. For a company to get a large number of meetings with great investors, a lot of people need to vouch for that founder as a high-quality founder. But if the company walks away from those meetings with a small number of conversions, it’s probably not because everybody thought the founder was bad or not that great. It’s because people walked away thinking it wasn’t that interesting of an opportunity.
Most of the time, the default in venture is that it’s not that interesting of an opportunity. But in a small subset of cases, it’s a very interesting opportunity, and most VCs, as I mentioned, are reactive and not proactive. Nobody wanted to be proactive about learning about an opportunity that this clearly very special, very intelligent person has decided it makes sense to dedicate at least the next decade of their career to working on.
If you think about the landscape today—seed versus early-stage A and B versus late stage—is there anything else that’s especially unusual about this period compared with your whole history of doing this?
The valuations in this market mean you really have to have a big imagination for what these companies can turn into to justify the prices you’re paying. It’s easy to have that imagination when you’re looking at companies like Anthropic and OpenAI raising at trillion-dollar valuations and going public. The market has allowed you to expand your imagination to that point.
When you look at what these deals are getting financed at, what I’m saying is probably reminiscent of 2021: it’s getting harder for people who actually know the market to understand what something will clear at, pricing-wise.
In a somewhat normalized market, I can tell a founder, “Hey, I think if you go out and we get a number of term sheets and we have some leverage, we could probably get this valuation ballpark from a high-quality partner.” I think in the current market, you can get a term sheet as low as here and as high as here, and that’s like a 4x difference in price. I’m not even on the money when it comes to where these things will price, because it’s so erratic that somebody might just come in and do something irrational to not lose the deal. The irrational thing might be to pay triple what the second-highest bidder is willing to pay.
Part of the reason that’s happening in this market is that every single founder you want to lead a financing round for has an expectation for what they want to raise at. Oftentimes, it’s pegged to the fact that they have a friend with a similar company and similar traction who raised at this valuation. The valuation that friend may have raised at may have been because one person paid 3x what the second-highest bidder was willing to pay. You have these situations in this market where you’ll give a founder an offer that you think is generous, and they’ll generally be offended by it because they can point to an example of someone they know at a comparable stage who raised at 3x or 4x the price.
3. Consensus Bets Feel Crowded
Do you think, then, in this market, the most rational thing to do is go all the way to the other end of the spectrum, to the businesses where we can really investigate them? Everything you’re describing feels so untethered from any underlying reality of a product or its traction, anything that the business is ultimately going to get built on. It’s literally just reputation and this weird multiplayer game of pricing, totally untethered. Whereas Databricks, where I think you’re a big shareholder, it’s so, so much simpler to understand or think about an investment in Databricks from here. There are fantastic companies growing really fast that are really big. Do you think perhaps the smarter thing to do is to go to the total other side of the spectrum, given this craziness?
I think that’s probably—
A lot less competition there.
It’s less competition. You could deploy large sums of capital into those companies and be a passive investor. I think it’s probably a safe way to know that you have exposure to great companies. The main question is, if you’re pricing those companies in today’s environment, you need to answer for yourself whether you believe it’s 2019 or 2021.
In 2021, you could have invested in Stripe at $100 billion, and Stripe was a great company. It still is a great company, arguably one of the best. But today Stripe is $179 billion, which is great, but that’s an 11% to 12% IRR over 5 years. By the way, that’s still better than the bottom 75% of ventures. So you can probably secure yourself a top-quartile position in the venture market by backing the best companies at the highest prices, because they are great companies, and great companies tend to continue to be great companies. They do tend to compound over time.
But the thing that gives me a little concern about that strategy is how consensus the approach you described has become recently. I’ve seen a record number of VC firms in the last 18 months hone in on the same idea: There are a handful of obviously great companies that can absorb a huge amount of capital, and as a VC, if you can demonstrate access to these names, you can raise large funds and become a passive investor with large sums of money in these companies. It’s almost become a consensus strategy, an easy way to deploy large sums of money and do well. You might have more data on this than I do, but I can’t really think of many times in history when an easy strategy also turned out to be a good strategy long term. I think that’s the interesting thing about this market: If everybody believes that’s an easy way to make money, it’s probably not a great way to make money. But I can’t think of any strategies that were considered easy money at the scale that people believe it’s easy money right now.
What about the data companies, where the narrative for years was, “This is more like a trade than a company”?
Yeah.
We need to arm all these models with tons of data, but once we have the data, we’re going to be good. It seems like maybe that narrative has shifted, and these things seem to be more persistent or bigger than one might have thought. What do you think of the data companies?
Data companies are kind of a conundrum to most investors because I think the historical belief was that these companies don’t have long-term sustainability. Eventually, the people they sell the data to will get all the data they need and won’t need their data anymore. But I think people have dramatically underestimated the data needs of these model builders. It seems their data needs are only increasing, and there are certain companies that are just ahead of the curve on knowing what data they need.
They can deliver the data at the quality the model builders need, and they’re going out and collecting that data. Some of the data is harder to collect than others, and those companies charge accordingly, and their margins are treated accordingly. I’ve seen two types of companies that sell data in robotics. I just believe data is a gigantic business, and we can continue selling data for the next, 10, 15, 20 years and collect tens of billions of dollars in revenue as a result. I don’t think there’s ever going to be an end to the data needs of these companies.
There’s a subset of VCs that believe that and just want to invest in a lot of data provider companies. Then there are other companies selling data that are like, “We believe that data is a means to an end, and that ultimately there is an end in sight to how much data we can sell and how much we can scale our business. But we believe selling the data is a really good way to finance our own research and train our own robotics model, because we know this market better than anybody. We know how to collect the data, and we can either go out and raise tons and tons of equity capital, which we’ll do as well, or we could also finance our research by selling the data to the other model providers.” So it’s two camps, and they’re both incredibly well-versed in the category, but they have two very different perspectives on the ultimate outcome of a data-selling business.
What do you think the impact is going to be of the newest influx of angel investors, all the people who are going to get incredibly wealthy from these big 3 companies and maybe a few others going public? Some of the employees’ balance sheets are the size of huge angel funds or huge seed funds, even.
I think it just continues making Silicon Valley the best place in the world to start a company. I think the reason no other ecosystem has ever taken off the way Silicon Valley has is because Silicon Valley has the largest density—by an order of magnitude, probably 100x or greater—of individual people running around writing $100K to $1M personal checks off their balance sheets to help put companies into business before they’re institutional. I think that’s just a unique thing you won’t have anywhere else: people who’ll write you a million-dollar check, and if they lose it, they don’t care. It’s the mentality you have here, and I think it’s probably good for innovation the more and more people you get doing that. But I do think a bunch of people making huge amounts of money off early angel investments is probably good for continuing startup innovation.
4. Abstract Scales Its Strategy
I’m sure you want Abstract to keep evolving. How are you thinking about that evolution? You’ve been kind of a pure-play seed investor for a long time and established yourself as one of the seed firms founders come see when they want to raise money. Do you want to stay that? Do you want to evolve into more than that?
We have evolved into more than that. In recent years, we’ve dabbled a little bit more at the venture stage, which I’d define as Series A through C. We’ve led some Series B financings in companies like Socket and Doctronic or Dust. We’ve also made passive growth investments in companies like Crusoe. I think we put about 15% of our 2023 vintage fund into the formation round of xAI. We were a big investor in the Series B round of Polymarket.
We’ve scaled. I think our first fund was $100 million. The latest one we just raised is $800 million. But to give you a sense of how much the market has shifted from when I went from a $100 million fund to an $800 million fund: when I was raising a $100 million seed fund in 2018, I got so much shit from LPs saying that it was too big of a seed fund for a fund one and that I should raise a smaller fund.
In this latest vintage, I went out for $800 million, and I got LPs commending me on my discipline for staying so small. It’s just funny that $100 million was seen as a large fund in 2018 and $800 million is seen as a small fund in 2026. But if we look at some of the best deals we’ve invested in over the years that have produced the lion’s share of the returns for our fund, in almost all of them, we were never allocation-constrained. We were capital-constrained.
We have a lot of data points that support that. We have very good relations with a lot of the companies in our portfolio. We have great conviction in those companies, but historically, our fund size hasn’t given us the option to be a lead investor in those financings. Having a relationship with a founder who trusts you, knowing them early, showing up outside of a traditional financing process, and giving them a number that they don’t feel like they need to shop is probably the easiest way for us to continue piling capital into our value-driving positions. Taking enough of the round that they don’t feel like they need to do much besides go ask their other investors for pro rata helps, too.
But most managers who are building a venture capital firm or franchise at the stage that we’re at are often overly fixated on what their fund looks like in the future, and not fixated enough on what their current fund could be. My general take is that if you deploy the capital you have as well as you possibly can, by the time you go out to raise your next fund, you’ll have more optionality than you would have otherwise.
Every time we go out to raise, we basically base it on where we are about 6 months before we go to market. What is our position in the market today? How does it compare with our position when we raised our last fund? Do we believe that we can deploy more capital into the same quality of companies, if not greater quality of companies than we were in the previous vintage? And if so, do we believe we can raise a larger fund without compromising on the return profile we’ve created for ourselves? If the answer is yes, then we raise a fund size that corresponds with what we believe we could actually deploy.
Somebody asks me, “What’s the biggest mistake I’ve made since building Abstract?” The answer is actually that we’ve always raised our funds too small. In every single fund we’ve had, I could have maintained an identical return profile with probably 2x the capital we had there. From an LP perspective, they would’ve gotten the same quality of return. From our founder perspective, we could have given them more capital.
We could have been a lead investor more frequently than we were a co-investor, and that would have improved our position within those companies and our name in the ecosystem as a firm that not only leads seed rounds but also leads later-stage financings, if we’d had the ability to do so. So I think, if anything, the model is going to continue to be: do the best you can with the money you have, then evaluate your position and raise accordingly at the next vintage.
I thought it’d be fun to ask the same 3 questions at the end of each of these. The first 2 are ratings from 1 through 10. The first rating is how exciting the opportunity set of companies feels to you at the current moment, relative to your own history. You and your firm are among the people with the most reps looking at companies. So, the opportunity set versus history: 1 through 10, with 10 being the highest.
Frothiness, 1 through 10, self-explanatory. 10 is the most frothy. Maybe early to mid-2022 was like a 2 or something. The third question is what you think the bellwether companies are today. You can name 1, 2, 3, whatever, and don’t just say Anthropic. What other companies do you think would teach you a lot about what’s going to happen, or what is happening, in the marketplace, if you watch them? You can answer them however you want.
Sure. In terms of enthusiasm around the opportunity, I’d say it’s a 10 relative to my time in venture, which is coming up on 10 years now. You see it with a subset of the market. It’s not that people are funding all the best companies with huge amounts of capital. People want to fund so many companies. It’s the largest number of companies we’ve found that we want to be investors in, relative to any other vintage.
In terms of opportunity set, it’s a 10. In terms of frothiness, I’d call 2021 a 10, and I’d probably call this market a 6.5 to 7. While the frothiness is there, there’s also a lot to justify it, which I think is really interesting. The rate at which these companies are growing, getting adoption, and achieving product-market fit is kind of unprecedented, and that would warrant frothiness.
If you leave out the obvious—the companies that are IPO candidates in the near term—and look at the next generation of companies we’d probably be looking at 5 years from now, those are the companies you definitely want to have exposure to in this vintage. I’d put NewLimit and Chai Discovery really high up on there, and Isomorphic Labs really high up there in terms of bio and whether you can build an AI-native, next-gen Genentech. The market opportunity for those companies is $500 billion to $1 trillion in outcomes. It’s even bigger if they execute on their ambitions, and they have the best teams, the best investors, and access to the capital they need to actually pull those things off.
I think those companies are super interesting. I think the idea of owning your own data, owning your own knowledge intelligence, is a really unique idea. I think the company that probably pioneered that, as the leader in the category, is Applied Compute. It’ll be really interesting to watch as they continue to execute. I think it’ll be very interesting to see how the prediction markets play out in a few years. The rate at which those companies are growing is insane, and they’re not even AI companies. They’re just true consumer businesses. I think Physical Intelligence is—
Sure.
—super interesting. I think that could be a multi-hundred-billion-dollar, trillion-dollar company. It has an exceptional team, a huge magnet for talent, and a huge ability to retain exceptional talent. Those are the kinds of things I always look for. In the market we have in venture in Silicon Valley, there’s no shortage of talent. But for a company, retaining the highest-quality talent is probably the single hardest thing you have to do, because the best people at your company can get a job anywhere they want, with a higher comp and equity package than you’re paying them. So whenever I see companies with a high density of that talent and the ability to retain it, I just get extra bullish on those companies.
Fun 3 questions to end with. It’ll be fun to compare in future episodes. Ramtin, thanks so much for doing this.
Of course, man.