The New Rules of AI-Era Venture: Menlo’s Venky Ganesan on Fund Size, Ownership, IRR, and Cycle Risk

1. Guest Background

This episode of 20VC is an interview by Harry Stebbings with Venky Ganesan, identified in the evidence as a venture investor at Menlo. The episode evidence matters because it gives Venky a specific professional frame: he is not appearing as a broad technology pundit, but as an investor whose relevant work includes venture investing, fund strategy, AI startup investing, portfolio construction, pricing, market sizing, and decisions about when to sell investments.

Menlo is presented in the evidence as a venture firm with multiple cycles of investing history and a current focus on backing defining AI companies. That makes the interview less a generic conversation about AI enthusiasm and more a discussion about how a venture firm should behave when the rules of company formation, fundraising, ownership, dilution, and LP expectations appear to be shifting at the same time. Venky’s comments repeatedly return to the institutional constraints behind the excitement: fund size, access, position sizing, DPI, IRR, public-market alternatives, and the cost of missing the largest outcomes.

Harry’s role is also important. He frames the conversation through the title question of whether one can still win in venture without a $1BN fund, then pushes Venky through concrete market phenomena: seed rounds that no longer look like old seed rounds, AI revenue metrics that feel murky, multi-tranche rounds, falling ownership, strategic-acquirer downside assumptions, and the competitive disadvantage of smaller managers. The episode is therefore an analysis of AI-era venture economics, spoken through the perspective of a practicing investor and tested by a host who keeps translating the theory into deal-level pressure.

2. What the Episode Covers

The episode’s main line is how AI changes the venture equation from the first check to exit. Venky begins with a personal investing lesson from Avnex. He bought $5,000 of IPO shares that rose to about $200,000, then fell 90%, and he eventually sold for roughly $8,000-$9,000. The point is not a universal rule that investors should always sell early. Venky uses the story to separate a young person’s personal balance sheet from Menlo’s institutional posture today. A life-changing amount of money may justify taking chips off the table; a venture firm pursuing defining AI companies may rationally choose to keep swinging for a grand slam.

Harry then moves to the present AI market, where many companies seek hundreds of millions of dollars and some new labs seek billions. Venky calls the environment disorienting, but he does not advise professional investors to step out. His dot-com example is central: some smart venture investors left the market in 1996-1997, missed 1997-1999, and returned at the 2000 peak. From that story comes the operating principle of the episode: keep playing, but play differently. In Venky’s formulation, discipline means more selectivity, better portfolio construction, and more careful position sizing, not simply refusing to participate.

The early-stage investing discussion turns seed into an option. Venky says each seed investment buys the option to see whether a company is an outlier; the fund should size up only when revenue and quantifiable metrics support that belief. Harry complicates that framework by pointing to murky AI revenue claims, including contracted annual revenue that is not live ARR and run rate extrapolated from the best day. Venky broadens the issue: once investors measure a metric and put weight on it, founders and markets will find ways to game it. That makes the seed-option idea depend on later evidence rather than on enthusiasm alone.

The latter half of the episode moves through ownership, dilution, LP expectations, and exits. Venky says Menlo assumes roughly 60% dilution from a seed check to exit, so 10% can become about 3.5%-4%. He links dilution to time horizon: faster growth and faster exits reduce dilution, while longer holds damage IRR and ownership. LPs, in his telling, want DPI rather than more TVPI, but also cannot afford to miss AI because their private-equity software exposure is being affected by AI. The episode closes with founder assessment, founder mode, acquisitions, the difficult position of $30 million-$100 million funds, and Venky’s advice that LPs should look forward by asking successful AI founders which partners they respect.

3. Core Views: Reasoning, Examples, and Limits

Venky’s first core view is that even in a market that looks overheated, professional venture investors cannot reliably protect themselves by leaving the field. His reasoning is not that bubble risk is imaginary. It is that timing markets is extremely hard, and the cost of being absent can be severe. The dot-com example carries the argument: investors could be right about the market feeling expensive in 1996-1997 and still be wrong in outcome if they missed 1997-1999 and came back at the 2000 peak. In this framing, discipline is not withdrawal. It is participation with stricter selectivity, more deliberate portfolio construction, and position sizing that recognizes cycle risk.

That view leads directly to his seed-as-option framework. Venky is not saying that every expensive seed round deserves capital. He is saying the first check increasingly buys a position from which an investor can observe whether a company becomes an outlier. The right to observe, learn, and potentially size up is valuable only if the fund has a defined standard for later action: revenue, usage, customer quality, growth efficiency, or other quantifiable evidence. The limitation is obvious and important. If the entry price is too high, or if the investor lacks the reserves and access to follow on, the option may become an expensive spectator seat rather than a source of alpha.

A second core view is that AI-era metrics require more suspicion, not less measurement. Harry gives the episode’s practical examples: contracted annual revenue that is not actually live annual recurring revenue, and run rate extrapolated from the best day in history. Venky’s response is broader than AI: any metric that investors heavily weight will be gamed. His SaaS net revenue retention example shows the mechanism. A $100 purchase order can be split into a $10 order followed by a $50 order to make expansion appear more impressive. The conclusion is not to ignore metrics. It is to ask whether the founders and investors are building toward terminal value or merely optimizing for the next markup.

Venky’s third view is reflexivity. He accepts that kingmaking can work for a time when real growth is underneath it. A strong company grows quickly, gets a markup, receives more capital, gains media attention and human capital, and then grows faster. The problem begins when copycats mistake the markup for the cause rather than the consequence. At that point, a financing pattern becomes a substitute for business quality. His discussion of tranche financing follows the same logic: the original innovation was to separate “build with me” capital from ordinary capital, but once detached from company quality it becomes another technique. The uncertainty is that Venky offers an investing lens, not a mechanical test. Reflexive cycles can persist longer than skeptics expect, and he explicitly says we do not know exactly how or when they stop.

Ownership is another area where Venky avoids a simple slogan. He says ownership always matters, but only relative to the size of the opportunity. Owning 2% of a trillion-dollar company can be preferable to owning 20% of a $100 million company, and Menlo owns less than 2% of Anthropic. Yet that is not an argument that ownership no longer matters. Venky says early ownership matters before a company is proven as an outlier, because once the outlier is obvious, the game shifts from selection to access and position sizing. He also warns that if a fund is not in the largest outliers, low ownership can make the portfolio fragile; ownership in midsize outcomes provides insurance when the grand slam does not happen.

The fifth core view is that speed has become part of the venture math itself. Venky says Menlo assumes about 60% dilution from first check to exit. That means a 10% seed position may end near 3.5%-4% after later financings and option pool expansion. Faster companies can reduce this dilution because valuation rises quickly, capital can be raised on better terms, and exits may come sooner. Speed also lowers the equity cost of hiring. Venky’s example contrasts a $200 million company granting 2% to a senior executive with a $2 billion company using $20 million of RSUs, about 0.1%. The limitation is that speed is not a synonym for quality. Slow, high-quality businesses may still be valuable, but in a venture fund they face a tougher IRR, dilution, and opportunity-cost comparison.

LP pressure is the sixth major view. Venky argues that private AI venture has to justify itself against public-market alternatives. Successful AI companies pay an economic tax to Nvidia, hyperscalers, and possibly foundation models; those exposures are available, or may soon be available, through public markets without venture fees and carry. Therefore, Venky says private VC needs to beat that alternative by roughly 1,000 basis points. At the same time, LPs want DPI rather than more TVPI, and many cannot avoid AI because their private-equity portfolios are heavily exposed to software assets affected by AI. That creates a stricter test for managers: access to AI is not enough; they must show speed, liquidity, and differentiated return potential.

One of the episode’s most useful constraints is Venky’s warning against strategic-acquirer downside stories. He does not treat current large acquisitions as permanent market structure. Instead, he compares them with dot-com-era deals such as Nortel buying Kyros and Lucent buying Chromatis for multibillion-dollar stock consideration despite no product or revenue. He recalls the old belief that successful startups would sell for billions and unsuccessful ones would be bought for the preference stack, then notes that the post-March-2000 reversal showed that assumption could fail. The application is clear: strategic M&A can be a scenario, but it should not be the main justification for paying a high price.

Finally, Venky’s answer to the title question is conditional rather than binary. He says $30 million-$100 million funds are in a tough place today because they are competing at a table where larger funds have much larger chip stacks. But he also cites managers such as Sarah Guo and BoxGroup/David Tisch as examples that exceptional hustle, networks, and execution can still win. This is not “small funds are dead.” It is “ordinary small-fund playbooks are under more pressure.” His related advice to LPs is to look through the windshield, not the rearview mirror: historical performance is a five-to-seven-year lagging indicator, so LPs should ask successful AI entrepreneurs which partners they actually respect.

4. Learning and Application

For investors, the first practical lesson is to stop treating “participate in AI or sit out” as the real decision. The better question is what the check buys. Does it buy ownership, information, future access, a board-level commitment, or only a small passive allocation in a crowded round? If a seed check is an option, the investor needs pre-defined triggers for exercising that option: real revenue, retention, usage intensity, customer quality, growth cost, product velocity, or evidence that the team can compound. Without those triggers, option language can become a polite way to justify expensive FOMO.

The second application is to diligence AI revenue by unpacking the accounting vocabulary. Harry’s examples of contracted annual revenue and best-day run-rate extrapolation are not minor technicalities; they change the economic meaning of the number. Investors should ask whether revenue is live, whether contracts are signed but not deployed, whether a run rate is based on repeatable demand, whether usage is subsidized, whether growth depends on unusually high compute spending, and whether customer behavior resembles renewal or experimentation. Venky’s NRR example shows that even familiar SaaS metrics can be distorted by deal structure. Metrics are still necessary, but no single metric should be allowed to carry the whole investment case.

The third application is to connect ownership, position sizing, and time. Venky’s dilution assumption means the first-check ownership is not the final ownership. A fund that enters with low ownership, cannot follow on, and faces long holding periods may end up with a thin return even if the company does reasonably well. Conversely, if evidence emerges that a company is a true outlier, the fund needs the capital, conviction, and relationship to ladder up. The boundary is equally important: do not concentrate heavily before data exists just because the company might be the next Anthropic, but do not let ego, anchoring to an earlier price, or resentment about another investor’s entry price prevent a rational follow-on when the data has arrived.

The fourth application concerns secondaries. Venky’s view makes partial liquidity a tool for endurance, not a betrayal of conviction. When an investment is up 30x, 40x, or 50x, selling 10%-15% can lock in gains, improve DPI, and make it psychologically and economically easier to hold the remaining position for a longer outcome. This is especially relevant when founders are also considering secondary sales. The boundary is that selling everything is different from trimming. Venky generally avoids selling the full position unless the company is being sold or the founder relationship no longer exists.

The fifth application is founder assessment. Venky’s method is useful because it moves beyond charisma and market size. He asks why the founding team came together, how they understand one another’s strengths and weaknesses, and whether the founders possess self-awareness. His “five best friends, three words” question externalizes self-description and can then be checked against references. In practice, this can help investors, boards, and executives test whether a team has the cultural foundation to make hard decisions, hire around weaknesses, and avoid blind spots. His missed pass on Sean Parker also turns founder judgment into an expensive lesson: the ability to explain complex ideas simply, and to understand virality, network effects, and human behavior, can be a signal worth taking seriously early.

The sixth application is for LPs. Venky argues that historical performance is a rearview-mirror measure because it lags by five to seven years. LPs should still examine DPI, TVPI, fund pacing, reserves, and realized performance, but in an AI platform shift they should also ask successful AI entrepreneurs which partners they respect and would actually work with. That question tests current relevance. Its limitation is that founder reputation can be influenced by recent hype, social proximity, and access to hot rounds, so it should complement rather than replace financial diligence.

For managers without a $1BN fund, the episode offers conditions rather than comfort. Smaller funds can still win, but they need a sharper reason to exist: a technical community they understand earlier than others, unusual founder trust, a stage or geography where large funds are less effective, or a level of hustle that creates access before the round becomes obvious. They should not copy the large-fund playbook of buying many small options and assuming they can always size up later. Large funds have reserves, brand, and follow-on capacity. A smaller fund needs asymmetry of insight, relationship, or speed.

The final application is risk language. Strategic acquirer downside, founder mode, fast markups, and massive TAM can all be valid parts of an investment discussion, but each has a boundary. Strategic acquirers may not protect the preference stack in a different market, even though regulatory windows, competitive pressure, and high equity prices may create more acquisitions for a time. Founder mode can exist in non-founders, so founder departure is not automatically zero, but replacement quality matters. Fast markups can reinforce real companies, but they can also mask weak ones. A huge TAM can justify paying more only if the investor truly sees a larger opportunity, not merely because that is the price required to win the deal. The episode’s practical discipline is to separate evidence from narrative before capital is committed.

Source

• Original episode: Can You Still Win in Venture Without a $1BN Fund? Menlo’s Venky Ganesan on the New Rules of VC