Summary

The news: Mercury founder Immad Akhund has closed his first institutional fund at $26M, partnering with Yash Toshi (ex-EQT Ventures), formalizing ~350 angel investments made since 2016. The design is deliberately non-threatening: 60 companies at ~$150K average non-lead checks — “isn’t it better if Sequoia or Founders Fund is leading the round and I get to invest alongside them?” — powered by the fact that “30 to 40% of all startups use Mercury.”
Founder advice that cuts against Harry’s book: take the highest price. “We did our Series B at 120x, which was not rational. This is 2021, but we did it and I would do it again” — $120M into a 40-person company. The discipline is downstream: raise enough at the high mark (a $1B valuation on a $50M raise is the real mistake), then don’t spend it — while conceding VCs actively push founders to spend.
He is unapologetically biased toward serial founders with a chip on their shoulder — “being an entrepreneur is irrational, but being a serial entrepreneur is especially irrational,” and that’s the signal. Best specimen: Truebill, first investment at a $16M valuation to a $1.25B exit (>30x, 2016–2021), from the repeat founders behind Webs.com. His biggest miss ran the other way: passing on Scale AI’s 19-20-year-old founders — “I thought I could run this company better… I was just so wrong.”
“AI is overhyped and overvalued” at seed: the same idea pitched a fourth time at a $40M valuation, and Harry observes that “there were always two or three competitors in everything five or six years ago, but now there’s 15,” all raising $10M+. Deeper problem: labor-replacement revenue priced at a third of labor cost will compress to “a tenth or maybe even a 20th of the labor cost” once competitors on the same foundation models undercut — “the margin compression is just inevitable.” He’s rotating into space and hard tech, which he says are better than AI SaaS companies at seed.
Seed economics have shifted from unicorns to decacorns: at $20-25M entry prices post-dilution, “I’ve seen as low as 8x from a seed investment to unicorn — this is awful.” A 10x fund requires $10B+ outcomes, and angels need at least 20-30 checks — “five bets” is not a portfolio.
Venture’s future is a barbell with a dead middle: multi-stage funds will IPO (“we heard some stuff about GC doing it”), more money enters because outcomes are now trillion-dollar scale, small-check investors do fine — “I don’t know what happens in the middle.” Public markets: he says most people see ~$10B as the minimum, so tenders and secondaries substitute.
Notable change of mind: 12 months ago he was “very skeptical” of advanced superintelligence; now “the advancement in AI has just been relentless… it’s probably going to happen sooner than we think” — yet he’ll have more engineers in five years, not fewer: “I have infinite ambition.”

Deep dive

1. The news: a $26M debut fund built around non-lead checks

Immad closed his first institutional fund at $26M “last week or the week before,” partnering with Yash Toshi (ex-EQT Ventures), a friend who invested in Mercury seven years ago. It formalizes ~350 angel investments made since 2016, mostly via an AngelList rolling fund; five or six new checks are already out.
The construction: 60 companies at an average $150K non-lead check, plus occasional $1M “conviction checks” for founders he’s known for years, and selective — never automatic pro-rata — reserves. The strategy is designed not to compete with lead investors: as an active CEO he can’t lead rounds anyway, so he rides alongside the lead instead of fighting it.
Why now: LPs told him they’d back a real fund but not the AngelList vehicle, and Mercury’s reach — “30 to 40% of all startups use Mercury” — pushed deal flow past what he could process. “I don’t like doing things badly, and I felt like I was being a bad angel investor.”

2. The raise: three weeks, one $7.5M check, and “it’s very boring”

Core allocations closed in about three weeks: three anchor LPs, fund-of-funds investors making up ~60% of the LP base — Harry name-checks one, likely Cendana, and gets a yes — the rest entrepreneurs and GPs, biggest single check $7.5M of the $26M. The slow part was the LPA — “a multi-party negotiation for these esoteric terms” that took a month and a half. “This is a silly process.”
His biggest surprise: fund pitching teaches nothing. “It’s very boring… pitching a fund, there’s not much to learn in the process” — unlike pitching a company, where the questions reshape your story.
Harry’s ethical challenge — a founder owes VCs a company, not a portfolio — drew a two-part defense: transparency (“Mercury is my main job… this has always been part of the story”) and genuine symbiosis: he thinks 100% of Mercury’s first 30 alpha customers were companies he’d invested in. And why bother, when, as Harry calculated, 20% carry on $26M is $5.2M — an amount he could instead realize by selling Mercury secondary? “It’s fun to build institutions.”

3. 350 angel deals: remove your ego, expect silence

Lesson one: stop pitching founders your better idea. Young founders will nod along, “and obviously that’s not their idea… you really have to remove your ego and really listen” — you’re along for their journey, not a major part of it.
The formative data point was likely Rappi (“a DoorDash for LatAm”): invested at a ~$20M cap, unicorn within a year and a half — and they never spoke to him again. “What is the point of being an investor if my best investments don’t even talk to me?” His resolution: that IS the point for a capital returner. The model now — “send me a text and if I have time I’ll talk to you,” in practice three or four founder calls a week, because “time is about energy, not time.”
The mirror-image error: passing on young founders who don’t push back. As a part-time partner he saw Scale AI when the founders were ~19 and 20: “I thought I could run this company better… I don’t see how they’re going to figure it out. And I was just so wrong.” Judge founders against their life situation — intimidated youth is not a red flag, contra Harry’s test.

4. Serial founders with a chip — and naive outsiders

Forced into one camp: “I just prefer serial founders. I have such a bias towards them” — specifically a serial founder with something to prove, not one who could retire on $100M. His own chip: a $45M exit after four pivots. “Being an entrepreneur is irrational, but being a serial entrepreneur is especially irrational” — and that irrationality is itself the qualifying criterion.
On domain expertise he takes the other side: he prefers naivety to ten years in the market. He knew nothing about fintech when starting Mercury, and after likely Andreessen Horowitz led the seed, not one dedicated fintech seed fund would follow — “all fintech funds saw was the problems… which in hindsight is ironic.”
The best specimen of both lessons: Truebill — first investment at a $16M valuation, $1.25B exit, >30x from 2016 to 2021, built by the repeat founders of Webs.com in brutal consumer fintech. Twin takeaways: repeat founders somehow own difficult competitive spaces (Rippling likewise), and timing — they sold in December 2021, and “I wish I’d taken some more chips off the table in 2021.”

5. Take the highest price — the case Sequoia proved

His self-declared contrarian take: founders can’t resist the top valuation, so do it right. “We did our Series B at 120x, which was not rational. This is 2021, but we did it and I would do it again” — $120M raised into a 40-person company. Two rules: raise enough money at the high mark (a $1B valuation with a $50M raise is the real mistake), then don’t spend it.
Harry invokes his running Parker Conrad debate: most young founders, handed the money, spend it. Immad partly concedes and redirects blame: “The VCs are like, hey, go spend the money… they really want to go for the home run.” The fix is “frameworks for how you can do it but still be successful — which is tricky.”
Sequoia’s Series C is his proof the best firms earn the price: he pitched more than 20 funds, and “Sequoia does not take their position lightly — they did the most work,” the most customer diligence, the most data-room effort. “I don’t want a term sheet just because someone’s FOMOing.” The kicker: partner Sonia later told him she was 95% convinced after the first meeting — the work came after conviction, not before.
It was ultimately the highest offer, though “I probably would have taken it if it was a little lower” — the good firms match or beat the top bid anyway, and at that stage “it’s not about the money for either side.”

6. Seed math now: unicorns return 8x — hunt decacorns

His single piece of advice to aspiring angels: this is “a rich person’s game,” and one or two checks make no difference — you need capital for at least 20-30 investments, both to iterate as a picker and because the return distribution demands diversification.
The arithmetic behind the fund: at $20-25M entry prices, post-dilution, “I’ve seen as low as 8x from a seed-stage investment to unicorn — this is awful.” Hence: “What we’re really doing as seed investors is unicorn hunting — or at these current valuations, you’re hunting for decacorns.” He wants a 10x fund, which requires $10B+ outcomes: “I’m not happy to get a 2x or 3x.”
The reserves fight is the episode’s best disagreement. Harry: run no reserves, do SPVs with deal-by-deal carry into the winners. Immad: founders dislike “FOMO SPVs,” and reserves let him act on signal — “you know within six months whether that company is going to kill it.” Harry’s rebuttal from his own fund one — Linear, Linktree, Captions, NexHealth, AgentSync “were all pretty slow burns.” Immad’s hedged concession: Airtable took a three-year ramp, so “maybe six months is too early — but you definitely know before everyone else, because you’re in the company.”

7. AI’s revenue-quality problem: labor-replacement margins compress 10-20x

The revenue he’s most skeptical of is labor-replacement revenue — “we’re a third of your labor cost, just install us.” The ROI sale works until three or four competitors on the same foundation models pitch the same thing and a customer notices one charges half. End state: “the actual margins are going to compress massively and we’ll end up at a tenth or maybe even a 20th of the labor cost as actual eventual revenue… the margin compression is just inevitable and there’s very little moat and network effect against it.”
Cursor’s claimed ~$400M is the sustainable kind in his read: SaaS-priced well below value delivered, and “engineers don’t like changing tools.” But even there, competition sets price — “if they try to charge for the productivity gain… Windsurf will come along and charge $20.” You end up charging against competitors, never against value created.
On whether moats are dead: “We’re still in the flashlight-apps era of AI.” Eventually the old SaaS defensibility — brand, multi-product, enterprise connections — reasserts itself, as it did for HubSpot and Salesforce. “We’re just in this moment of extreme change” where “no one knows anything. Everyone’s trying everything.”
The headcount corollary: Mercury (a Cursor shop internally — “I haven’t heard too many people using Windsurf”) will have more engineers in five years, not fewer. “If my engineers get more productive, I’m going to come up with more things to do… I have infinite ambition.”

8. Seed AI is overhyped — he’s hiding in space tech

“I think AI is overhyped and overvalued” at seed: “It’s the fourth time I’ve heard the pitch of the same idea,” raising at $40M with a little traction and great investors attached. Harry’s observation is that “there were always two or three competitors in everything five or six years ago, but now there’s 15” — all raising $10M+. His rule: a seed investor “can’t be doing too many investments at the top of the hype cycle — the same thing happened in 2021.”
He hasn’t quit — four of his last eight deals are AI — but the bar is now a second-time founder deep in a specific domain applying AI to it (his example: a prop-tech deal with “a really specific AI application”), or traction so real “this already seems like a rocket ship” and you pay the valuation.
Where he’s gone instead: space and hard tech, which he says are better than AI SaaS companies at seed. His method for new sectors: make one or two learning investments (likely Momentus in 2016/17 — the SPAC didn’t work out) and talk to the ten people at the edge. The map as he draws it: basically three existing space markets — rockets (SpaceX dominates; his bet is Stoke Space, whose initial round was probably near a $20M valuation and which he says has now raised more than $100M), imaging from orbit (a ~$40B market; he’s in likely Albedo), and communications (Starlink). Bio he skips entirely: “I speak to someone and they’re like, we’ve cured cancer, and the next person — we’ve cured cancer. You sound good, but I don’t know.”
His quickfire change of mind sits in tension with the seed caution: 12 months ago he was “very skeptical we’re going to get advanced superintelligence”; now “the advancement in AI has just been relentless… probably going to happen sooner than we think” — though he still won’t commit to the next five years.

9. Mercury’s scar tissue — and the $100B bull case

Biggest strategic regret: launching the credit card two years late (2022 instead of ~2020). He’d assumed Brex owns cards, Mercury owns banking — then Brex entered banking and Ramp launched a card. Corrected, emphatically: among Mercury’s own customers its card is now “not just bigger — it’s completely dominant” over every other corporate card.
His competition doctrine after two decades: “Every single year there’s been some competitor that was better funded and seemed really scary. 99% of the time it didn’t matter.” He bans competitor talk internally — proposals must cite the customer or the long-term vision — and shrugs off Harry’s counter-positioning frame (Ramp’s save-more inversion of Brex): “I want to be your first bank account,” 200,000+ customers, a different business at inception stage.
Harry’s rude-but-fair question — framing Mercury at $5B and asking why Brex and Ramp are valued at roughly double — “Each company has to prove out its own thing… people understand enterprise SaaS and payments more than they understand banking” — where, at scale in the US, it’s basically Mercury, “Chime and a few others.”
The $100B case: US banking is a $2T market and financial software another $500B — and they should be one market. “The only reason these markets are separate is banks don’t know how to build software.” In ten years your bank does invoicing, bill pay, and employee spend, fully integrated — “for how big this opportunity is, this seems very uncompetitive.” His other retrospective: Mercury’s $6M seed on a $23M valuation was its highest-dilution round ever; $3.5M would have covered it.

10. Venture’s future: multi-stage IPOs, a squeezed middle, broken public markets

The forecast: “It seems inevitable that a few of these multi-stage funds will IPO” (he cites the GC chatter), and more money enters venture, not less, because outcomes exploded — “we have trillion-dollar companies; when I started investing, a hundred billion was huge.” The shape is a barbell: small checks fine, multi-stage fine — “I don’t know what happens in the middle.” He concedes Harry’s cost: multi-stage seed products force him to pay higher prices, “that definitely impacts your returns.”
Against Harry’s claim that multi-stage funds killed seed: “They made it difficult for you, but not for me necessarily” — there are maybe eight or nine billion-dollar funds with a brand, and they back one flavor (big-company execs, repeat founders). Hungry first-timers who “figure stuff out” remain open field for smaller checks.
Why nobody IPOs: the cost and rules of being public, plus few active public investors left — outside the S&P 500 and the index complex, “you can hardly get an analyst to look at you,” even at $5B. “Most people are saying 10 billion is probably the minimum.” Private liquidity substitutes meanwhile: Mercury just ran an employee tender, and against Revolut-style secondary tightness he’s relaxed — “I don’t want this to be a lottery ticket… I want it to be actual valuable stock.”
One founder-craft answer worth keeping: write the culture down at three or four people. Mercury’s six attributes each carry a real trade-off — “we look for humble people… a successful exec with a massive ego, we never hire those people” — and that day-zero act still holds a cohesive culture near 1,000 employees, “because we did it at day zero and it’s very hard to do it later.”

Market Overview | May 13, 2025 Huawei's AI Supernode: Market Impact with Magic Compute's Xu Lingjie