Bucky Moore @ Lightspeed Venture Partners: Why You Cannot Do VC If You Do Not Do Pre-Seed

Summary

• Bucky Moore announces he’s joining Lightseed as a partner after 7.5 years at Kleiner Perkins and ~11 in venture. His case for mega platforms: outcomes have jumped from the $30–50bn Databricks/Snowflake era to multi-trillion (SpaceX, OpenAI, Anthropic), enabling a conversation venture has never had — “what if we invest a billion dollars in one of these companies… turning a billion dollars into 10, 20 billion.” But the chip stack only wins if the firm stays devoted to formation-stage investing, because Altman- and Dario-class founders pick partners on startup DNA.
• His biggest change of mind in 12 months: model APIs are “a death by a thousand cuts” — 100x year-over-year price-per-token deflation, near-zero switching costs, heavy competition, ever-rising capex. Even a flagship franchise like Anthropic’s codegen dominance holds only until a rival ships a better model: “the revenue is going to yo-yo.” Durable value sits in products built on top of models — “even inside of the model labs” — so every frontier lab will build products.
• Harry’s data pushback: Anthropic’s $4bn-round investors made only ~3.5–4x after dilution despite it now being a 12x model-provider company — not obviously venture-grade. Moore’s defense: “the story there is still very much being written” — billions in run rate growing over 100% year-over-year is unprecedented (ServiceNow was amazing at 20–30%), and the app layer atop these models is just starting.
• Venture’s structure goes barbell: LPs commit for 3–4 funds at a time and have already committed, leaving ~5–6 firms with a captive opportunity to bankroll trillion-dollar assets; small specialists rise too, while the “uncanny valley” of $500m–$2bn funds becomes “an increasingly challenging place to be.” On model-lab conflicts: “you either have to be in all of them or you have to pick one of them. There’s no middle ground.”
• Sizing a fundamentally new market “borders on being a fool’s errand” — the best founders “set their own rules” (Rippling’s market “keeps getting bigger every quarter”) — and per his ex-partner Mamoon, “the best companies always feel expensive.” With agents moving from software spend to replacing labor budgets, his answer is to build conviction and go all-in. Quickfire stack-rank: team, traction, market — “traction is so hard to manufacture.”
• The seed fight: Harry charges mega platforms are “destroying seed” — “we do three on 15, you drop 10 on 50” — and Moore is “completely wide-eyed about the trade-off”: deterministic markets can absorb jumbo seeds; new-market founders need optionality, since a low post-money preserves M&A exits and extra runway on a drifting company “can actually be really punitive” for founder time.
• The scarce input in AI apps is AI engineering talent, not domain expertise — Harvey’s lawyer-plus-deep-AI-CTO pairing is the template — and the buy signal is CIO zeitgeist: “their CEO and their board told them you’re going to get fired” without AI, which is why enterprises are adopting faster than in any prior cycle and why he’s bullish on enterprise apps in finance, software engineering, and cybersecurity.
• On AGI: stay open to a plateau — “AGI is already here in certain pockets” via test-time compute — and even a halt would leave “an incredible wave” for society and investors; the real question is “when do we run out of new scaling dimensions,” which looks unlikely near-term. Asked if he did SSI: “I can’t comment on that at this point, but I suspect SSI will announce their fundraising and their investors.”

Deep dive

1. Bucky Moore joins Lightseed — the chip-stack case for mega platforms

• The news: after 7.5 years at Kleiner Perkins and almost 11 in venture, Moore joins Lightseed as a partner to drive the early-stage enterprise investing “that really set the tone for Lightseed’s success” — “this new chapter of my career… I think it could be the last.”
• His argument for the walls-of-money firms: a new shape of company “puts to shame” the Databricks/Snowflake/DoorDash-era hallmarks — multi-trillion-dollar outcomes in SpaceX, OpenAI, Anthropic. Scale platforms get to ask “what if we invest a billion dollars in one of these companies that could be worth trillions… turning a billion dollars into 10, 20 billion — that’s just not a conversation that’s ever been had in the venture industry,” or, he adds, in any asset class he knows.
• The catch: “none of that matters if you don’t maintain a lot of dedication to early-stage investing.” If you’re Sam Altman or Dario choosing whose billions to take, you’re buying brand DNA — these labs are still startups in how they think, so the platform must keep “working with raw formation-stage companies” or lose exactly what the labs select for.

2. Model-provider economics: brutal dilution today, “the story is still being written”

• Harry’s data-backed pushback: outside OpenAI, model providers haven’t been venture investments — those who did Anthropic at $4bn hold roughly 3.5–4x after dilution and employee stock payouts, considering it is now a 12x model-provider company. Moore concedes the margin compression and capex-driven burn but insists “the story there is still very much being written.”
• His counter: “billions of dollars of run rate growing in excess of 100% year-over-year — I don’t think we’ve ever seen that before.” ServiceNow at billions of ARR growing 20–30% was considered an amazing company. Harry’s corroboration: he challenged Thrive’s Vince on OpenAI’s $30bn round ~15 months ago; today revenues sit at roughly half that valuation.
• Moore’s biggest reversal in 12 months is the revenue mix. At that $30bn round, “the question in the room was, should we even value this ChatGPT thing as anything?” while APIs were pitched as the next Stripe or Twilio. Reality inverted: APIs face a 100x year-over-year decline in price per token, near-zero switching costs, heavy competition, and ever-increasing capex — “a death by a thousand cuts.”
• The consequence is instability even at the top: Anthropic’s codegen dominance looks healthy, “but the moment another player releases a better model in that use case, the revenue is going to yo-yo.” The compelling long-term businesses are products built on top of the models — “even inside of the model labs, not just by third parties.”

3. Where apps survive the labs: the long tail is “so long and so fat”

• His mental model is the hyperscaler era: as Amazon, Microsoft Azure and Google went multi-product, categories demanding unique customer insight and focus still produced independents. Likewise now — the long tail of AI applications is “so long and so fat” that “it’s hard for me to see how the model providers are able to play in each of them.”
• The danger zones are the core: he’d be “a little nervous” about codegen — “far and away the killer use case for LLMs” and the obvious next lean-in for the labs — plus consumer. Exhibit A on intent: OpenAI “famously saying, hey, if you’re an investor in Glean you don’t get to be an investor in OpenAI,” with the possible Windsurf acquisition as further foreshadowing.
• Outside that core, big companies get built that “monopolize all the customer insights that come from being first and going really deep” — and those either get acquired by the model providers or force the labs to struggle against them.

4. LP physics caps the platform count — and hollows out venture’s middle

• Will more mega platforms emerge? “Harder and harder for me to see”: sophisticated LPs commit for at least three to four funds, and the big pools have already committed — “the physics of how big they are” leaves no room. You could argue that 5–6 firms have a captive opportunity to bankroll trillion-dollar assets, which “may not have to go public for a really long time” given private-market demand.
• How big can they get? “The honest to god answer is I don’t know” — it’s indexed to how AGI, space and robotics actually play out. His quickfire prediction: a barbell — small dedicated specialists and large platforms both rise, while “that uncanny valley in between,” which Harry pins at $500m–$2bn funds, “is going to be an increasingly challenging place to be.” On liquidity, Moore says it is not structural: managers will devise solutions; Harry mentions a continuation fund as one example.
• On conflicts: at Kleiner, backing competitors “was kind of a red line”; Thrive and Founders Fund concentrate in OpenAI; a16z tries to invest in every lab — “a rational strategy so long as you have buy-in from the entrepreneurs.” His own line: “You either have to be in all of them or you have to pick one of them. There’s no middle ground.”
• On concentration of value: polled investors would say only a few winners per mega-category, “but we’re still really early in the super cycle” — there’s talk of likely Helsing in defense, OpenAI, Anthropic, maybe xAI, “I just don’t think we know.” On robotics, “I’ve convinced myself that this time is truly different” industrially — “we could be surprised to the upside.”

5. “The best companies always feel expensive” — and market sizing is mostly a fool’s errand

• On paying up for the 0.01% assets, the line from his former partner Mamoon that sticks: “the best companies always feel expensive — and I think that’s proven true in every example I can think of.” With agent products “quite literally going from going after existing software spend to starting to replace human labor” and its budgets, outcomes get bigger — so once a company shows a path to market leadership, “you really have to build conviction and go all-in in such transient times.”
• Market sizing “depends”: for a better cybersecurity product chasing known spend, sizing keeps you sober about how to capitalize the company. For anything fundamentally new, “the act of sizing a market is just so imprecise that it borders on being a fool’s errand” — the best founders “set their own rules,” deciding what market they’re playing in and convincing customers it exists. Rippling as told: started as payroll, but layers synergistic products so effectively that “the market size just keeps getting bigger every quarter.”
• Spreadsheet investors aren’t dead, but their window “is just a lot narrower”: blue-chip AI companies raise enormous sums “before there is a spreadsheet,” so those investors “are just going to keep getting pushed later and later stage” — and “the book is still being written on whether that’s a good thing.”

6. Harry’s charge: “you mega platforms are destroying seed”

• The spiciest exchange: Harry — “we do three on 15, you guys come in and drop 10 on 50… I don’t think that is good for the companies and rarely do I see that play out well.” Moore doesn’t deny it: he’s seen multiple companies conclude “we raised too much money at too high a price,” losing flexibility “not just in downside scenarios but in some cases base scenarios.” “I’m completely wide-eyed about the trade-off.”
• His delineation maps back to market sizing: founders with a deterministic market — cybersecurity replacement, known headroom — can underwrite jumbo rounds; founders fleshing out a new market that “could be really large, could be non-existent” should stay lean, because “optionality is without a doubt your friend.”
• Optionality means two things, and he confirms he meant price first: a low post-money makes “finding a good home” through corporate acquirers far easier (“I was once on a corp dev team, so I know that world very well” — Cisco’s). Second, founder time: an extra two or three years of runway on something not resonating “can actually be really punitive given the opportunity cost of amazing founders’ time” — raise too much and “in some sense they’re kind of stuck with it.”

7. Timelines are lengthening while the revenue bar rises

• The low-hanging software fruit is picked; the interesting companies now solve “something deeply technical… that’s never been solved before,” which takes years — and that slowness is defensible and hard to replicate. His example as told: Clay, now growing very fast in sales tech, had “five or six years of very little to no growth before it took off.” Harry’s pushback on the Figma parallel — Dylan was always obviously brilliant — draws the concession that many special founders’ companies still don’t work, but the canonical “either works in the first 12 months or not” is no longer sufficient.
• Meanwhile Harry worries a series B doubling from 7 to 14 gets stack-ranked against likely-Mercor/Lovable/Bolt trajectories and dismissed. Moore confirms he’s right: “the bar is going up for what best idea looks like” — while flagging that the quality of that AI revenue “is very mixed” and its durability unproven.
• The signal he now hunts, learned from Glean and Windsurf (led by former partners whose names are unclear in the captions): finding the zeitgeist at the senior-most decision maker. CIOs are “furiously seeking out ways to apply AI… because their CEO and their board told them you’re going to get fired” otherwise; once one CIO’s magic moment lands, mimesis spreads it CIO-to-CIO. Next up: finance, software engineering, cybersecurity — the root of his enterprise-app bullishness.
• This breaks the conventional wisdom that enterprises adopt slowly: unlike cloud’s “gradually then suddenly,” there’s broad-based consensus that failing to embrace AI “is just existential” to a company’s existence — “unprecedented appetite and urgency” is why growth rates look “like nothing we’ve ever seen.”

8. Picking beats winning — the parlor tricks don’t matter

• What wins competitive rounds isn’t the meme — “famous VC picking you up in his helicopter… the courtside seats at the Warriors or Knicks game.” Those parlor tricks exist at every firm and “none of that stuff really matters.” In every contested process he’s lost, the winner “had been doing that work for the better course of a year.” The founder quote he keeps: “the person that I went with, they just had insights about the business that not even my existing investors had.”
• Hence picking: you must choose whom to spend that year with — and at a magnet platform the top-of-funnel is so large that “picking actually can be the failure mode of a lot of GPs at these funds.” Harry’s disagreement, worth keeping: generational founders “are quite obvious” — domain specialization “doesn’t help you pick, it helps you win,” because the founder chooses whoever made them smarter. The hypothetical founder rationale is: “Harry was super nice, but he didn’t make me smarter and he didn’t know the market like Bucky did.”
• Moore’s biggest self-correction in 11 years: he arrived from Cisco corp dev evaluating “lucrative markets and the right technology — notice I did not mention founder quality.” Now thesis work is just “a flashlight” showing where to look for amazing founders, and the hardest discipline is “the humility and the restraint to conclude that there aren’t great founders in that space.” His bad investments came from downweighting the team’s execution capacity “at the expense of my intrigue for the product and the technology.”
• Quickfire stack-rank: team, then traction, then market — but he’s learned to overweight traction: “traction is so hard to manufacture,” and it signals a founder good enough “to overcome any market size limitations that exist.” On Mamoon: “known as someone who is very metrics-driven, and I think that could not be more false — he has this incredible taste in people.”

9. “You cannot do venture unless you’re doing pre-seed”

• Competitive sets have exploded from “one or two players” to four to six per category — “the single biggest challenge” of seed and series A investing now. In a recent anonymized AI-app deal, his team waited to see whose product was best before choosing between domain-heavy and AI-heavy founding teams; they won the deal, but “the price was a lot higher than it would have been had we leaned in earlier.”
• The hindsight instinct: “you think it’s the domain expertise that matters. It is without a doubt the AI expertise that is more scarce and therefore matters more.” Harvey is the template — a former lawyer plus a deep-AI CTO who can recruit AI-native engineers; domain-expert CTOs without that background “are just not able to get the talent in the building.”
• Why platforms must do pre-seed: stop and “you lose the instincts for just how fast these companies can change” — Harvey’s series A product was raw, and one investor mistake in this wave of AI investing is “they’ve failed to imagine what they can be as the models get better.” Harry amends: it’s also access — series-B-and-beyond firms now do 50 customer calls or 28-page market analyses for a first meeting; “the cost of first-meeting entry has gone through the roof.” Moore agrees only partly: a small early check works as a wedge with real legwork behind it, but “the check itself does not” buy access.
• On signaling when a multi-stage seed lead doesn’t do the A: rival seed funds “will go very very out of their way to instill this wild fear,” but in practice the A gets done on milestones “rather than the whimsical taste of the multi-stage firm that led the seed.” Harry’s counter he accepts as fair: the killer is the middle ground — companies doing “okay” that get orphaned, increasingly common these past two years. Still, seed funds and platforms need each other: “to alienate those seed funds… would just be crazy.”

10. AGI may plateau — and that’s fine

• His contrarian read on consensus: most assume “this path to AGI is just this ever-increasing upward slope”; he thinks “we should be much much more open-minded to the possibility that we could arrive at some form of plateau and still have an incredible outcome.” In fact, “AGI is already here in certain pockets”: test-time compute — “the same thing that brought us AlphaGo” — lets models try thousands of paths “whereas we kind of have to think about the best one… in a sense, humans are still predicting the next token.” Even if progress halts, “we are so early in bringing these capabilities to bear.”
• On plateau odds: pre-training looked less lucrative, then test-time compute, post-training and reinforcement learning opened new axes. The right frame: “when do we run out of new scaling dimensions” — “very unlikely at least in the near term,” and with the best and brightest concentrated inside the labs, “I’ve learned to never bet against human ingenuity.”
• SSI (likely Ilya Sutskever and Daniel Gross) is the strategic mirror-image of the app-building labs: refusing products means no short-term optimizations trickling into research — which he’s “seen with my own eyes” talking to lab employees — buying “less distraction and more ambition.” Did Lightseed invest? “I can’t comment on that at this point, but I suspect SSI will announce their fundraising and their investors.” Harry’s own 12-month reversal, stated flat: he was dubious of OpenAI’s sustainability; now a $2 trillion outcome is “unwavering.”

Verification Notes

• The raw captions do not clearly identify the former partners credited for Glean and Windsurf; their names are omitted above.