Sameer Dholakia on Bessemer's AI Investing Playbook

Sameer Dholakia has felt magic exactly three times in a thirty-one-year career in software, and each time a trillion-dollar market followed within a few years (05:03). The first was the early internet, when he graduated into Silicon Valley in 1995. The second was the rise of cloud computing, when he watched VMware move a running computer workload from one machine to another without interrupting it. The third came during a sabbatical in the winter of 2021, when a young founder showed him a marketing tool built on top of GPT-3. "Magic," he thought again (06:53). Dholakia, now a growth investor at Bessemer Venture Partners and formerly the CEO of the email infrastructure company SendGrid, believes this AI cycle is bigger than the internet and cloud cycles combined (06:53).

The scale of his conviction shows up in the numbers. Bessemer just raised $5.75 billion across two new funds: a $4 billion growth fund and a $1.75 billion seed and early-stage fund (00:24). That seed number alone is almost double what a well-known peer fund just announced, and it underlines a blunt claim Dholakia makes about the market today: with AI valuations where they are, and with many AI companies needing heavy compute spend, a small seed fund can not take enough shots on goal to find a winner (28:17).

"Intelligence is gonna be bigger as a market than oil because oil powers all the machines. Intelligence powers all the humans." — Sameer Dholakia [07:19]

That line explains why Bessemer went all in on Anthropic, the AI lab behind the Claude models. The firm joined Anthropic's Series D at a $15 billion pre-money valuation (08:02), a round led by Menlo Ventures, which Lightspeed and Bessemer joined. Dholakia says it was, by his partner Byron Deeter's account, the lowest-voted deal in Bessemer's history at the time (10:18). The thesis was a "cloud redux": just as cloud computing consolidated around a handful of capital-intensive hyperscalers like AWS, Azure, and Google Cloud, frontier AI would consolidate around a short list of labs that could afford the scaling laws, with hundreds of applications built on top of them (08:35). His biggest regret is not writing a bigger check. He says he and his partners underestimated how large the business could become by "factors" (11:22).

Paying for the Future, Not the Past

That Anthropic round meant paying about 150 times the company's current annual recurring revenue, or ARR, which is the yearly subscription revenue a company is already collecting (17:56). Dholakia says a price like that would "make most venture capitalists throw up" if you only looked backward. His answer is to look at NTM, or next-twelve-months revenue: his estimate of what the company's revenue will be a year from now, not what it is today (18:24). He treats 10 to 15 times NTM revenue as a reasonable entry multiple for companies growing at rates he has never seen in 31 years in software (19:15). He applied the same logic to Harvey competitor Lagora, leading a $1.8 billion round when the legal AI company had just $20 million of ARR; it has since grown to roughly $200 million (19:19, 17:56).

Price is not the only variable. Dholakia says he pays close attention to the quality of revenue: gross and net dollar retention, meaning how much of a customer's spend sticks around and grows over time. He would pay a much lower multiple for a consumer or "prosumer" business, where churn can kill a company, than for a sticky enterprise product (19:52). Thin gross margins do not scare him either, as long as he can defend why they will expand; he says many of his past investments had margins at entry that were "factors" lower than what they reached within 12 to 24 months (21:04).

"I absolutely believe capital is a moat. I've always said that markets pick winners, and then it becomes a self-fulfilling prophecy." — Sameer Dholakia [23:59]

That belief, that raising a war chest is itself a competitive advantage, is the thread running through the episode. Dholakia argues that compute, the computing power needed to train and run AI models, is scarce and expensive enough that companies without massive funding simply cannot compete at the frontier (24:20). Big rounds create brand, brand attracts talent, talent wins customers, and the cycle reinforces itself, much as it did when SendGrid and Twilio became the default developer tools for sending messages in the 2010s (24:20). He is careful to add a caveat: capital is necessary but not sufficient. A company still needs a product customers rave about (25:19).

Dholakia extends the same consolidation logic to the venture industry itself. He argues that venture capital, like most industries that professionalize and scale, is splitting into two viable positions: giant global platforms like Bessemer that can write checks from pre-seed through pre-IPO across every sector and geography, or extremely specialized firms that add differentiated value in one narrow niche. The middle, he says, is "a hard world to make money" in (29:56, 30:49).

Teammates, Not Colleagues

Dholakia's second obsession, alongside pricing, is culture, a subject he says most CEOs claim to care about without doing much about it. His guiding line, used for over a decade, is that "teammates and friends" outperform "colleagues and coworkers" because teammates do not want to let each other down (40:57, 42:02).

"They're not working for themselves. They're working for their teammates, and they're working for their team, and you never wanna effing let down your teammates." — Sameer Dholakia [42:02]

At SendGrid, he built this around four explicit values he called the four Hs: happy, hungry, humble, honest (40:57). Harry suggested that a value only works if people could reasonably argue against it, a premise Dholakia disagreed with. "Humble" at SendGrid meant favoring the word "we" over "I" in interviews, to the point that interviewers literally counted pronouns; that would have been a miserable standard for some candidates, and that was the point (43:01). He also insists culture needs shared time that has nothing to do with work. SendGrid flew its entire staff to Mexico every January, then spent a second day pairing people off for kayaking, mountain biking, or just sitting by a pool (44:42). When something goes wrong, he says, colleagues point fingers or head for the door; teammates who have spent real time together instead say, "we've got to figure this out" (45:21). He wants the same directness inside Bessemer's investment committee, where he believes partners should tell each other bluntly when a price looks wrong rather than soften it into mush (47:20).

That same appetite for a market with no obvious ceiling shapes how Dholakia talks about what comes next. He expects a "permanent structural shift" toward companies staying private longer, simply because there is now so much private capital chasing deals, but he also expects a real unlock in IPOs in 2027 and 2028, driven by public investors who want direct exposure to dozens or hundreds of AI companies, not just a handful (52:15, 52:59). He is particularly animated about Fireworks, an AI infrastructure company he says is led by CEO Lin Qiao (referred to in conversation as "Lynn") and president George Hu, and about Waymo, the autonomous driving company Bessemer put nearly $200 million into, which he argues could eventually be worth more than a ride-hailing business by selling autonomy as a subscription on every car, not just robotaxis (68:20, 70:08). He closes on biology: citing Anthropic CEO Dario Amodei's essay "Machines of Love and Grace," he says he believes AI could compress a half-century of drug discovery into a decade (71:31). It is, in other words, one more moment that feels like magic, and Dholakia has learned what tends to follow those.