Summary

Venky Ganesan’s answer to an overheated AI market is to keep playing, but change position size rather than pretend anyone can time the turn. Firms that exited dot-com investing in 1996–97 missed 1997–99 and returned in 2000; today, each seed check is “an option to see if it’s an outlier,” with concentration added only after revenue and quantifiable evidence arrive. “A dot is not a line.”
Seed still exists, but core AI “neo-labs” and $10–20 million application rounds make traditional small-fund construction brutally difficult. Menlo uses some seed checks to buy a seat at the table and is therefore “somewhat indifferent” to the initial valuation, expecting the real capital deployment to come after winners reveal themselves. For $30–100 million funds that need $1–3 million allocations, the high-stack poker table is a punishing place—though exceptional managers can still win.
AI revenue must be underwritten as a business reality, not accepted as a fundraising artifact. Harry flags contracted “annual” revenue that is neither live nor annual and run rates extrapolated from the best day of sales; Venky’s rule is that “if any metric is measured by an investor and they put a lot of weight on it, it’s gonna be gamed.” The decisive distinction is whether founders optimize for terminal value or the next markup.
Price and ownership matter, but neither can be judged without outcome size, evidence and access. Venky would rather own 2% of a trillion-dollar company than 20% of a $100 million company; Menlo owns less than 2% of Anthropic yet has laddered its investment until Anthropic reached 20% of one Menlo fund. Once an outlier is obvious, “there’s no alpha there”—the game becomes access and position sizing—so early ownership remains valuable.
Outcome inflation cannot excuse weak portfolio mathematics because time and dilution compound against the investor. Menlo models a seed position falling from 10% to roughly 3.5–4% by exit, or about 60% dilution; slow companies also damage IRR and require more financing and option-pool expansion. Because AI companies pay economic “taxes” to Nvidia, hyperscalers and foundation models, Venky says private venture must beat the accessible “Mag Seven” by roughly 1,000 basis points to justify fees, carry and illiquidity.
Strategic-acquisition “downside protection” is dangerous because buyers have no obligation to protect the cap table. Venky compares today’s confidence with dot-com acquisitions of pre-product companies and asks why a strategic acquirer would honor the cap table when it might hire the founders directly. At 30–50x, he favors selling perhaps 10–15% to lock in gains and enable a longer hold, but Menlo generally will not sell its entire position while remaining aligned with the founder.
AI’s capital intensity can justify faster deployment, but it does not repeal vintage risk. Menlo Eight invested across roughly ten months in 2000–01 and remains the firm’s only fund that has not returned capital; its $1.5 billion Menlo Nine and $1.2 billion Menlo Ten also underperformed expectations and drove LP departures. LPs now want DPI, yet many also need AI exposure as a hedge against software-heavy private-equity portfolios threatened by AI.
The durable edge Venky claims is institutional humility: remove ego, allocate capital well and keep running. He will accept a smaller allocation or a higher-priced tranche if it can make money for LPs—“the rest of this is all noise”—and tells LPs to evaluate the windshield by interviewing founders, not merely study five-to-seven-year-lagging performance. His closing principle: “There’s no limit to what a person can do as long as they don’t care who gets the credit.”

Deep dive

1. You have to dance, but every seed check is only an option

Harry’s provocation is that venture no longer resembles venture when teams report that “we can’t find anything under $100 million”—not the valuation, but the round size. Venky concedes that core AI labs may seek billions and that the moment is “very disorienting, confusing,” but cautions: “A dot is not a line.”
The market-timing lesson comes from firms that made money early in the dot-com boom, stepped away in 1996–97, missed 1997–99, then re-entered in 2000. LPs had hired them to remain at the cutting edge. Venky’s conclusion: play the game, but vary selectivity, portfolio composition and position size.
His portfolio model treats every seed investment as “an option bet.” A fund needs enough at-bats to encounter an outlier, then should size up only when revenue and other quantifiable evidence support the designation. Concentrating before the evidence arrives is a materially different risk from laddering into a demonstrated winner.
Seed is particularly hard in AI “neo-labs,” while even application companies increasingly raise $10–20 million rather than the old $3–5 million. Large platforms worsen the distortion: Menlo is sometimes “buying ourselves a seat at the table,” relatively indifferent to seed valuation because its actual objective is the right to invest heavily later.

2. Metrics become theater when investors reward the theater

Harry’s pushback—worth keeping—is that the evidence itself has become murky: contracted annual revenue may not be live or annual, while a revenue run rate may multiply the company’s best day by 365. Venky agrees with the mechanism: any metric investors heavily reward “is gonna be gamed.”
The SaaS-era specimen is net revenue retention. A single $100 purchase order shows 100% retention; start with a $10 order and add $50 a week later, and the displayed expansion becomes 500%. Beyond metric selection, Venky invokes “the bezzle”—while unsure who coined it—to warn that booms also conceal creative accounting.
King-making is Soros-style reflexivity: genuine growth earns a markup; the markup brings capital, publicity and talent; those inputs accelerate growth. Copycats then mistake the markup for the cause, and investors assume one markup guarantees another. “All reflexivity will eventually stop”; leverage and a major debt default, rather than equity losses alone, may expose the break.
Tranche financing began with sound differentiation: raise lower-priced “build-with-me money” from investors who add value, then higher-priced capital from passive investors. But cycles move from “the innovators” to “the imitators” and eventually “the idiots”; once every company copies the structure regardless of quality, it becomes another fundraising technique.

3. Price is often a proxy for opportunity—and ego is an expensive filter

Paying up can mean merely paying whatever wins the deal, but it can also mean seeing a much larger total opportunity than competing investors. Venture’s asymmetry changes the calculus: invested dollars can go to zero, while a winner can return 10x or more. Consequently, “the most expensive mistakes venture capitalists make are the deals they passed.”
Venky’s haunting omission came while he served on Plaxo’s board. After Sean Parker left, Parker urged him to meet a Boston college dropout; Venky distrusted the situation and declined even the meeting, forfeiting the chance to write perhaps a $50,000 check. Parker’s tell was uncommon insight into virality, network effects and human behavior, communicated simply.
On later, higher-priced tranches, Venky refuses to be insulted because another investor—Harry’s example is Peter Fenton—entered more cheaply. He admits ego has previously affected valuation negotiations, syndicates and allocations that felt too small: “My one lead ego is to make money for my investors. If I can make money for my investors, who the hell cares?”
Founder diligence starts with the premise that “the company you build is the team you build.” Venky asks why these people chose each other from billions of alternatives, then asks how their five closest friends would describe them in three words. References test whether that answer reflects self-awareness; known weaknesses are manageable, while blind ones are dangerous.

4. Ownership matters most before the outlier becomes obvious

Venky rejects treating ownership as an isolated target: “I’d rather take 2% of a trillion-dollar company than 20% of a $100 million company.” Menlo owns less than 2% of Anthropic, demonstrating that low percentage ownership can still produce consequential exposure when the denominator is extraordinary.
Before an outlier is known, ownership preserves return potential as well as information. Menlo’s Higgsfield example paired a $5 million check with 15% ownership and then an opportunity to size up. Once everybody recognizes the winner, selection alpha has disappeared: access and position size replace discovery.
Menlo has taken Anthropic to 20% of one fund, but Venky stresses the sequence. The relevant question is not whether 20% entered in the first check; it is whether the firm laddered toward 20% as new evidence strengthened conviction. With even 10% company ownership increasingly difficult, that staged concentration becomes the route to fund-level impact.
Harry questions whether falling ownership is justified by a broad expansion of outcomes or merely a handful of enormous companies. Venky’s answer is portfolio insurance: the target remains a home run, but ownership lets a triple matter when the grand slam never arrives. Otherwise the portfolio offers only “grand slam home runs or strikeouts.”

5. Time, dilution and public-market opportunity cost dominate venture math

Menlo assumes that 10% ownership at seed may become only 3.5–4% at exit—roughly 60% dilution from subsequent financings and option-pool expansion. As a working rule, Venky suggests expecting an initial stake to be approximately halved by the end.
Time is the hidden variable. A company that compounds valuation quickly can exit sooner, raise with less dilution and deliver higher IRR: “That’s a double win.” A long holding period inflicts two costs at once—IRR deteriorates while repeated financings and hiring grants consume ownership.
Velocity also changes employee-equity economics. A $200 million company might grant a senior executive 2%, a meaningful transfer of ownership; at $2 billion, it can offer the same $20 million of value through roughly 0.1%. Fast appreciation therefore protects existing investors from both financing and human-capital dilution.
Venky’s answer to “DPI or IRR?” is both, but today IRR requires explicit attention. AI companies pay a “tax” to Nvidia, a hyperscaler and potentially a foundation model—assets investors can increasingly access through public markets without venture fees or carry. Private funds therefore need about 1,000 basis points of excess IRR to justify their structure.

6. Downside protection can disappear before the company does

Venky learned liquidity through Avanex: a $5,000 IPO purchase grew to roughly $200,000, but he rejected his fiancée’s suggestion to fund a house deposit and eventually sold after a 90% fall for about $8,000–$9,000. Harry counters that selling Salesforce early would have destroyed Emergence’s defining return; Venky’s resolution is balance-sheet context, not a universal sell rule.
Today’s casual assumption that a strategic buyer will rescue a company at $1.5 billion resembles dot-com claims that success meant a multibillion-dollar sale and failure meant acquisition for the talent stack. Venky cites Nortel’s $3.5 billion purchase of Kairos and Lucent’s $4.5 billion Chromatis purchase as the prior-cycle pattern; it stopped working after March 2000.
More M&A may occur because permissive regulation, competitive reactions and elevated public-market equity create a temporary window. That does not protect investors: buyers care about founders and technology, not the cap table, and structured transactions already demonstrate that distinction. “Why wouldn’t they just hire the founders for the same package?”
At a 30–50x return, Venky wants the fund to ask whether it should sell perhaps 10–15%, ideally alongside founder liquidity. Locking in gains makes both founder and investor more willing to hold the remainder. Menlo generally will not sell everything unless the company itself is sold or it lacks a relationship with the founder.

7. Fast deployment is defensible only when vintage concentration is explicit

Menlo’s commitment model distinguishes passive seed exposure from a large check and board seat. In the Anthropic/OpenAI case, that is why it stayed with Anthropic rather than also investing in OpenAI. Venky presents this as Menlo’s authentic choice, not an industry-wide rule.
LPs simultaneously requesting smaller funds and slower fundraising are asking for incompatible outcomes if AI really is the defining platform shift. Unlike Google, which Venky says raised less than $50 million privately, current AI businesses require compute and scaling capital; if one fund withholds it, a competitor will supply it.
Yet vintage diversification remains real. Menlo Eight deployed during a roughly ten-month period in 2000–01 and is the only fund in the firm’s 50-year history not to return capital. Venky does not prescribe a dogmatic deployment calendar; he asks GPs to explain their reasoning, downside plan and concentration transparently to LPs.
Scale itself guarantees nothing: Menlo Nine was a $1.5 billion fund raised in 2001, and Menlo Ten was $1.2 billion in 2004; both underperformed expectations and many LPs left. Today LPs demand DPI, but AI exposure remains strategically attractive because their private-equity books are often three to four times larger and heavily exposed to software disruption.

8. Capital structure determines who still has room to act

Wealth makes investors less afraid of failure and more willing to “go to the hilt.” Venky’s poker analogy is blunt: the player with the largest chip stack can see more cards and bully the table. That makes $30–100 million funds especially awkward when they require $1–3 million allocations rather than easy-to-fit angel checks.
He resists turning difficulty into impossibility. Venky estimated Sarah Guo’s fund at around $200 million and cited her entry into highly sought-after companies; BoxGroup offers another counterexample. But these are “the best of the best,” not construction templates to extrapolate: “There’s no magic strategy”—exceptional managers out-hustle competitors and bring unusual grit.
A related control problem confronts 2021-era software. Private-equity owners face AI disruption but can use majority control to act; many venture-backed “zombie SaaS companies” have dispersed cap tables where nobody owns enough or cares enough to land the ship. Venky’s range is severe: the best result may be “getting spooned”—recovering capital—while the worst is zero.

9. Capital allocation beats charisma when the stakes compound

Asked to choose between product vision and capital allocation in a scaled founder, Venky chooses capital allocation because it subsumes product judgment: leaders must direct resources toward the products that matter and the returns they can generate. A product visionary need not possess the reverse discipline.
His comparison is Mark Zuckerberg—praised for Instagram and other capital allocation—against Evan Spiegel, whom he calls a product genius while noting Snap shareholders have not been rewarded for seven or eight years. Harry’s stock-based-compensation concern sharpens the distinction between creating an admired product and compounding shareholder value.
Venky is not worried that wealth automatically retires top performers: “Money doesn’t change people, it reveals them.” People who merely acted motivated may opt out, while A players treat money as scorekeeping because they love the game. He does concede that concentrated wealth changes the Bay Area and amplifies housing prices when constrained supply meets rising demand.
His LP advice is to “look at the windshield, not the rear view mirror”: reported performance lags current ability by five to seven years, so LPs should ask successful AI founders—even those who rejected the fund—which partners they respect. Internally, complacency gets the savanna rule: lion or antelope, “you just have to run,” and the next meeting, investment and board session matter most.

How Mark Zuckerberg, Palmer Luckey, Josh Kushner & Scott Wu Think | Jeremy Stern