Ramtin Naimi - The State of Venture - [Invest Like the Best, EP.499]
What this covers
Patrick talks with Ramtin Naimi, founder of Abstract Ventures, in a second conversation built around what his firm is seeing in live deals. The question he keeps returning to is whether this is 2019 or 2021, early in a run or near the top. He answers with deal mechanics. Hot rounds now carry a second tranche for investors who lost the deal, priced at 3 to 5 times the valuation, against roughly 2x in 2021. Some term sheets commit $150 million but fund about three months later. Bids on the same company can differ 4x.
The range then widens. On the fund side, Naimi says mega platform funds underestimated the capital needed to hold their target ownership, which leaves LPs choosing between shots on goal and ownership. On the AI side, he covers neolabs, whose compute buildout is limited by debt of five to ten times equity rather than by equity. He also discusses investors paying startups in compute instead of cash, and data companies split between selling data as a lasting business and selling it to fund their own robotics work. He expects only 10 to 50 percent of neolabs to keep raising.
The more contentious turns concern pricing and strategy. Naimi traces terminal-outcome anchoring, where the best-case outcome moved from $10 billion to a trillion dollars. He says that anchoring justifies 100x-revenue vertical SaaS only if growth stays extraordinary. He contrasts reactive and proactive returns, and discusses Silicon Valley's angel density. He closes on his own biggest mistake, raising every Abstract fund too small.
Naimi argues the current venture market looks like 2021 in pace, round structure, and deal chaos, but differs in real product-market fit, so prices are erratic and driven by shifting 'terminal outcome' goalposts, and the best returns come from proactively backing great founders early in non-obvious markets.
- Round pace, tranching, and pulled term sheets mirror 2021, but companies now show real product adoption and enthusiasm is more justified
- Mega funds underestimate capital needs for ownership, so they raise more and sooner, while LP liquidity is concentrated in a few names
- Valuations are set by irrational bidding and anchoring to peers' goalpost outcomes, so the best edge is proactive early conviction
Backing great founders in non-obvious markets, not consensus strategies, is the best source of venture alpha.
- The best source of alpha in frothy markets is backing great founders working on non-obvious ideas, because truly great founders are usually early to markets rather than working in bad ones, so competition and prices stay lower.
“if you back great founders working on non-obvious ideas, that generally tends to be the best way to generate alpha”
- When a brilliant founder meets many investors but converts few, it usually means investors judged the opportunity uninteresting rather than the founder bad, since most VCs are reactive.
“it's probably not because everybody thought it was a bad founder”
- Investors believed Gavin Uberti was brilliant, but many assumed a startup could not beat incumbents in AI chips until Sequoia's $10 billion valuation made the market believe, after which other chip companies raised at astronomical valuations.
“it wasn't until Sequoia blessed them with a $10 billion valuation that everyone was like, oh wait a second”
- There are two ways to make venture returns: reactive investing, which if done correctly gives top quartile performance, and proactive investing in markets not yet obvious, which can produce top 5 percent performance.
“I think reactive, if you pick correctly, allows you to be a top quartile manager. And I think being proactive allows you to be a top 5% manager.”
- Stripe bought at $100 billion in 2021 and now valued at $179 billion implies about an 11 to 12 percent IRR over five years, which is still better than the bottom 75 percent of venture funds.
“today Stripe is $179 billion, which is great, but that's what, an 11, 12% IRR over five years”
- When many VCs converge on the easy strategy of passive large checks into a handful of obviously great companies, that consensus signals the strategy is probably not a good long-term return strategy, since easy money at this scale rarely lasts.
“if everybody believes that's an easy way to make money, it's probably not a great way to make money.”
- The default outcome in venture is that most companies are not an interesting opportunity, and only a small subset are truly special, so the rare outcomes require investors to be proactive learners rather than reactive followers.
“The default of venture — most of the time it's not that interesting of an opportunity”
Current valuations and deal activity mirror 2021's frothy conditions, priced on best-case outcomes.
- Because deals move so fast, markups no longer feel like good news; they merely keep pace with the market, and missing one cycle makes investors feel they are falling behind.
“In this market, getting markups is the bare minimum requirement to be keeping up.”
- In 2021 there was a roughly six-month window when VCs were ripping checks, and many now regret those deals when they review their 2021 portfolios.
“there was like a six-month window in 2021 where VCs were just ripping checks”
- The main difference from 2021 is enthusiasm: investors now believe companies can become category-defining within 18 to 24 months and that products reach product-market fit faster than ever, including with non-early-adopter family members.
“I don't think people were investing in companies in 2021 thinking that these companies were going to be category-defining or be global-scale companies within 18 to 24 months.”
- In 2021 investors paid $1 billion valuations for companies with only about $8 million of SaaS revenue and constrained TAMs, which in hindsight looks unremarkable compared with today's opportunities.
“you would have companies with $8 million of SaaS revenue raising $1 billion valuations, and they had very constrained TAMs”
- In the last six to eight weeks, Naimi has seen more deals renegotiated after being offered than at any time since 2021, a sign of 2021-like instability.
“in the last six, seven, eight weeks, I've probably seen a larger amount of deals get renegotiated after being offered than I have since 2021”
- Investors who put most of their 2021 capital into the most obvious thing, 100x revenue SaaS deals, are probably now in the bottom 25 to 75 percent of funds for that vintage.
“they're probably now in the bottom quartile or bottom 75% of venture capital funds for that vintage”
- Pricing anchors on the best-case terminal outcome, which moved from about $10 billion in 2016 to about $100 billion around 2021 and now to a trillion dollars, so ordinary deals are priced as if they could be trillion-dollar outcomes.
“Every few years in Silicon Valley, the goalpost moves for what terminal outcomes can actually be.”
Venture returns are concentrated in a few trillion-dollar companies, with capital-hungry neolabs driving spending.
- LPs with 30 to 50 percent of exposure to venture are really concentrated in three to four companies such as SpaceX, Anthropic, OpenAI and Databricks, and only LPs exposed to those names are getting meaningful liquidity.
“That's 30 to 50% of their AUM in like three to four companies, and that's SpaceX, Anthropic, and OpenAI, and Databricks”
- One neolab raising $1 billion planned to spend all of it on compute for a single training run and expects to need another $5 billion next year.
“they were going to spend the entire billion next year on compute for one training run. And they said they would probably need to raise another $5 billion”
- SpaceX, Anthropic and OpenAI are already trillion-dollar companies, and Naimi bets there may be two to three more rather than forty.
“My bet is there might be two to three more and not 40 more.”
Founders should judge growth and valuation against companies with the same business model, not headline revenue.
- Data labeling companies are growing at extreme rates but earn only 15 to 20 percent margins because acquiring quality data is expensive, while companies built on foundation models give most margin to the model providers.
“You have data labeling companies growing at insane rates, but their margins are 15 to 20% because getting that data is expensive.”
- Vertical SaaS companies priced at 100x revenue can only be justified as labor-replacement businesses if they keep growing 5 to 10x year over year at massive scale, versus public SaaS trading near 13x revenue.
“the way you actually have to believe in it is these companies are still growing 5, 7, 10x year over year”
- Founders should compare their revenue growth to companies with the same business model rather than to headline top-line figures from very different businesses, since data labelers at 15-20% margins and model-dependent companies are not comparable.
“you need to actually compare your revenue growth to companies within your style of company, versus the absolute top-line figures”
- Some Abstract portfolio companies are expected to grow 200 to 700 percent this year while their margins move from negative 15 percent to positive 35 percent, meaning revenue growth and margin expansion are happening simultaneously.
“their margin profiles are going from negative 15% to positive 35%”
- Companies growing 10-20x year over year almost always have revenue that is not clean SaaS: it has very low or negative margins, high customer concentration, annualized one-month revenue, consumer or usage-based revenue instead of enterprise contracts.
“any company that's growing at that rate has something interesting about their revenue”
- The true addressable market for AI companies selling to law firms is the paralegal payroll rather than software budgets, so capturing 20 percent of that spend makes the valuation math work.
“So if these companies can capture 20% of that, the math starts to really, really work.”
Debt, not equity, is the binding constraint on compute buildout and its financing.
- Most VCs prefer clean equity without unique covenants and historically dislike venture debt, so new capital structures like General Catalyst's value creation fund needed years of education to gain adoption.
“VCs by default like their companies having clean equity that don't have unique covenants”
- The binding constraint on compute buildout is debt rather than equity, because neolabs need debt of five to ten times their equity, and debt costs are rising after neocloud public-market weakness.
“A lot of these guys that are building these things need debt in excess of 5 to 10x of the equity that they raised.”
- Some VC firms now finance startups with compute rather than cash, which founders prefer, but the structure is complicated by questions of how long compute is provided, how many times it can be recycled, and how fast it degrades.
“There are some VC firms that are financing companies with compute as opposed to cash.”
- Neoclouds are unwilling to take compute deals under $200 million, so smaller companies must lease scattered capacity, while hyperscalers, frontier labs and well-funded neolabs actually move the compute market.
“A lot of the largest neoclouds ... aren't really interested in engaging with compute deals that are sub-$200 million in scale.”
- Compute demand is only likely to decline if the compute needs of major players diminish or their capital access weakens, and the neolab segment is the most interesting part of demand even though it is small relative to hyperscalers in aggregate.
“the only thing that limits demand for compute over time is if the compute needs from these companies diminish or the companies that have these compute needs have a harder time raising capital”
Mega funds must deploy faster and larger, forcing a painful choice between ownership and breadth.
- One fund raised $2.5 billion, began deploying in July, and plans to return to market in January, while another raised $3 billion nine months ago and is now seeking $5 billion.
“I know one fund that raised $2.5 billion and started deploying in July and told me they're going to be back in market in January.”
- Mega platform funds raised two to three times as much capital within the same 18 to 24 month window as before, so their deployment periods are about 3x faster than they planned.
“So the deployment periods are about 3x faster than they thought they would be.”
- Every mega platform fund underestimated how much capital it needs to hold its target ownership, because there are more companies to back and higher prices, so they must invest far more than modeled and return to market sooner and for larger sums.
“Every single one of the mega firms has dramatically underestimated how much capital they need to actually get that level of ownership”
- Because funds lack capacity to keep both ownership and breadth, LPs face a choice: keep ownership and get far fewer shots on goal, or keep the number of companies and accept lower ownership, which caps upside.
“They either maintain their ownership in these companies, but they end up with way fewer shots on goal, or they end up with way less ownership”
Multi-tranche structures and bidding wars make term sheets expensive and their pricing unpredictable.
- In the current market, VCs who lost a hot round pay 3 to 5 times the valuation to join a secondary tranche alongside the primary, versus roughly 2x in 2021's fast follow, a practice Silicon Valley calls the 'cuck tranche'.
“it's usually not 2x, it's usually like 3 to 4 to 5x happening in tandem with the primary”
- Multi-tranche rounds are not an unhealthy market sign but a necessity, because capital-intensive companies need far more capital than they want to raise in one primary tranche.
“I think it's actually a necessity of these types of companies in the environment that we're in”
- VCs are offering $150 million term sheets that fund in mid-January, about three months later, and one $50 million term sheet wires only $5 million at close, meaning money is committed before it is in hand.
“we could wire $5 million upon closing, and we need another eight to ten weeks to fund the remainder of our commitment”
- Term sheets for the same company can range 4x in price, and any single bidder may pay triple the second-highest bid to avoid losing a deal, so prices are unpredictable even for insiders.
“you can get a term sheet as low as here and as high as here, and that's a 4x difference in price.”
- When a company gets around 15 term sheets it will be very expensive, while two or three term sheets usually mean reasonable pricing and a better chance for an investor to win a core position.
“If a company gets 15 term sheets, it's going to get very, very expensive.”
Small funds blocked leading rounds, so managers should size funds to the opportunity available.
- In most of Abstract's best deals it was capital-constrained rather than allocation-constrained, so its small fund size prevented leading rounds and capturing more ownership in core positions.
“in almost all of them, we were never allocation constrained, we were capital constrained”
- Managers should size each new fund by assessing their position in market about six months before raising, asking whether they can deploy more capital into companies of equal or better quality without hurting the return profile.
“Every time we go out to raise, we basically base it off of about six months before we go to market.”
- A $100 million seed fund was criticized as too large in 2018, while an $800 million fund in 2026 earned LP praise for discipline, showing how much the market has shifted.
“So it's funny that $100 million was seen as a large fund in 2018 and $800 million is seen as a small fund in 2026.”
- Abstract's biggest mistake has been raising every fund too small; with about twice the capital it could have kept the same return profile while leading more rounds.
“In every single fund that we've had, I could have maintained an identical return profile with probably 2x the capital we had there.”
- Abstract has moved beyond pure seed, leading Series B rounds in companies such as Socket and Doctronic, making passive growth investments in Crusoe, putting about 15 percent of its 2023 vintage into xAI's formation round, and investing in Polymarket's Series B.
“We put I think 15% of our 2023 vintage fund into the formation round of xAI.”
Neolab launches are uncertain, with only a fraction likely to keep raising capital.
- Abstract seeded Conduit nine months earlier, when brain-computer interfaces were not a hot category; Conduit's model reached about 80 percent accuracy decoding typed thoughts, has 30,000 hours of training data, and drew 13 term sheets after the category became popular.
“We seeded a company nine months ago called Conduit, which is working on BCI — brain-computer interfaces.”
- Where investors once expected two or three great neolabs, there are now 80 to 90 neolabs that have each raised between $1 billion and $5 billion.
“now there's probably 80 or 90 neolabs that have raised somewhere between $1 to $5 billion”
- It is hard to tell whether neolab launches are generalizable breakthroughs or benchmark-tuned demos that will not be commercializable soon, and Naimi expects only a fraction, perhaps 10 to 50 percent, to keep raising capital.
“whether it's like 10% of them continue to raise”
- Relative to his nearly ten years in venture, Naimi rates the current opportunity set a 10 for the number of companies he wants to back, and rates current frothiness 6.5 to 7 versus 10 for 2021, because growth justifies part of the froth.
“In terms of frothiness, I would call 2021 a ten, and I would probably call this market a 6.5 to 7”
- The next-generation companies Naimi expects to matter in five years include New Limit, Chai Discovery, Isomorphic, Applied Compute and Physical Intelligence, which he sees as potential hundreds-of-billions to trillion-dollar outcomes.
“I would put New Limit and Chai Discovery really high up on there. Isomorphic really high up there in terms of bio”
Talent retention and peer-set pricing shape the cost structure of venture-backed companies.
- Silicon Valley's advantage is its density, by roughly 100 times other ecosystems, of individuals writing 100K to 1 million personal checks before institutional funding, and recent angel wealth from big IPOs will likely reinforce that.
“Silicon Valley has the largest density, by an order of magnitude of probably 100x or greater, of individual people”
- Founders set expectations using peers' rounds, so when a friend's company raised at 3 to 4 times the normal price, founders are offended by offers Naimi considers generous.
“they have a friend with a similar company with similar traction that raised at this valuation”
- For companies building in venture-backed markets, retaining top talent is probably the single hardest task, because the best people can get higher compensation elsewhere, so high talent density is a strong bullish signal.
“retaining the highest quality talent is probably the single hardest thing you have to do as a company building in venture”
- Companies must pay top talent competitively against any employer offering higher compensation and equity, which makes retention a structural cost issue for every venture-backed company, not only a talent-market quirk.
“the best people you have at your company can get a job anywhere they want at a higher comp and equity package”
Data selling and consumer prediction markets may be durable businesses worth watching.
- Investors underestimated how much data frontier model builders need, which keeps increasing, so data companies may be durable businesses, though some sell data mainly to fund their own robotics research.
“I think people have dramatically underestimated the data needs of these model builders. It seems that their data needs are only increasing”
- Data selling could remain a business generating tens of billions of dollars of revenue over the next 10 to 20 years, because there will never be an end to model builders' data needs.
“we can continue selling data for the next ten, 15, 20 years and collect tens of billions of dollars of revenue”
- Data companies split into two camps: those who see data selling as a permanent business and those who treat it as a means to finance their own robotics research and training, with the second camp expecting an end to how much data they can sell.
“But we believe selling the data is a really good way to finance our own research, to train our own robotics model”
- Prediction market companies such as Polymarket are growing at an insane rate despite not being AI companies, because they are true consumer businesses, which makes them worth watching as a bellwether.
“the rate at which those companies are growing is insane, and they're not even AI companies, they're just true consumer businesses.”