Summary

The Anthropic call: GC put “a few hundred million” in at the $60B round less than a year ago — revenue then “under a billion,” publicly guided to grow ~9x — and did it again at $180B, a round 5x oversubscribed. Taneja’s math: at ~20x ARR versus peers raising at “50 to 100 times,” it was “the cheapest round that got done this year on a multiple basis,” and if next year’s revenue lands near $27B, the same multiple makes it “a half a trillion dollar company by the end of next year.”
Core doctrine, stated flat: “venture capital can’t scale and performance at the same time” — more money doesn’t create more Patrick Collisons. So GC caps venture funds at a size that can “deliver four to 5x,” routes scale into creation and customer-value vehicles, and defines winning as the biggest AUM from the fewest companies — Chanel, not Walmart.
Stripe is the template: seeded 2010, invested 14 times in 15 years, roughly $1B in for a sub-10% stake worth over $5B — “I think Stripe’s going to be a trillion dollar company… probably a 25-year hold.” The corollary sin is under-doubling: he told a GC partner sitting on a decacorn, “you’re going to make over a billion dollars on this investment and you’re an idiot — you gave up making the second billion.”
Jobs are the under-priced macro: “everywhere you offshored for labor benefit, you’re going to onshore for AI productivity” — GC says it bought a 3,000-person Philippine call center in a company called Crescendo on that thesis. It’s “a 5-year problem,” not 12-18 months; one consulting client asked for a plan to reach 100,000 employees “but only 10,000 of them are humans.” Governments comfort themselves society will slow it down; Harry adds, “I don’t think we have time.”
Growth bar reset: “Triple triple double double is definitely dead. I tell our investors don’t bring that to me… you got to go like 1 to 15 to 20 to 100” — Mercor went 1→500M in 17 months. The unresolved variable is durability: “we never had so much scale without just taking durability for granted,” and some of the fastest growers “will also not be around.”
OpenAI post-mortem: he passed on the structure and regrets it daily — but the dilution math illustrates the toll: per Harry’s cited chart, the $200M invested at the $1B valuation returned only ~25x, versus “hundreds of x” for GC’s best first checks. Microsoft “maybe took more risk,” made the highest multiple on ~$20B — and Harry said Microsoft now uses Anthropic for the majority of its suite.
His biggest change of mind is indexing over picking: “when you know the trend’s going to win, but you don’t know which one’s going to win, you’re better off backing all of them” — he did Stripe but skipped Square, tried to pick the winner in AI, and says “in hindsight we should have just gone and indexed,” as he cited likely Yuri Milner and likely Lightspeed doing.
Price discipline is usually a conviction tell: “Price only hurts once”; “investors use price as a reason to pass because they couldn’t gain conviction elsewhere — it just makes them sound pragmatic.” Returns come from concentration: of 200+ investments over 25 years, ~60-70% of returns sit in about 10 companies.

Deep dive

1. CEO and managing director both — seed is “the right to exist”

Taneja carries both titles deliberately: “General Catalyst is a business — but it wouldn’t be a business if it wasn’t venture capital at its core.” The stated ambition is startlingly small-ball: “we want to be one of the best seed firms in the world.”
Harry’s challenge — can you honestly justify seed hours on “a $200 million vehicle in a $25 billion pool”? The answer: orient to ownership and relationship, not check size, and bring on seed groups including Jeannette and La Famiglia, plus Venture Highway in India.
The stakes in his own words: “if we don’t do early stage investing well, we will lose the right to exist — and we’re paranoid about that.”

2. “Venture capital can’t scale and performance at the same time”

Against Doug Leone’s high-margin-boutique-to-low-margin-industry lament: the industry’s only innovations have been on three axes — stage, sector and geography — “make the funds bigger, put them in different geos.” The right question is retooling the proposition so founders build bigger companies.
The core belief: “just because we have more money doesn’t mean there are more Patrick Collisons or Sam Altmans.” Founders are the zero-sum constraint — so the escape is to “manufacture more outliers than the ones that naturally exist on the power law,” not to buy everything already on it.
Architecture follows: venture funds stay capped at a size that can “at least deliver four to 5x funds,” while creation (incubation, rollups) and customer value funds carry the other capital solutions. “The reason to scale capital isn’t to be a low margin business — it’s because you want to build the very best ones and really lean into them.”

3. Stripe: 14 rounds in 15 years, sub-10%, trillion-dollar hold

Seeded in 2010, “I’ve invested in Stripe 14 times in the last 15 years” — about $1B in, still sub-10% ownership, worth more than $5B today. The call: “I think Stripe’s going to be a trillion dollar company. You just got to give it 10 years… we’ll have a 25-year hold in some form or the other.”
His most memorable founder meeting: he asked Patrick Collison who the ideal customers were — “he said they haven’t been born yet.” His reaction, via the Sixth Sense ring-drop: “oh crap, I don’t even have a complete view of the world.” Every payments expert called it niche; he backed the person anyway — “markets expand also.”
Livongo, built in GC’s offices, made “a few billion” — returning one fund roughly 3-4x and another close to 1x — yet Stripe still ranks as GC’s best investment.

4. Embrace serendipity at seed, be intentional on macro — the Coinbase scar

The confessed miss: Paul Graham showed him Coinbase’s seed round — “a Bitcoin ATM, what is that?… it still haunts me.” Brian was amazing; his own thesis-brain killed it. Lesson: at seed, back great founders “regardless of our view of the world.”
Intentionality lives one level up, in the “global resilience” theme — every region rebuilding defense, energy, industrials, health, and financial sovereignty. GC claims to be the only firm invested in a defense prime in the US, Europe and India — likely Anduril, Helsing, and Rafi — because each region “needs to create its own AI deterrent solutions.”

5. Jobs are the most under-discussed macro shift

His four-part model of national transformation: AI for deterrence (“without peace, you don’t have capitalism”), healthcare, AI diffusion into business — then jobs: “there is an immense reskilling that needs to happen. People are starting to give lip service to it, but it hasn’t hit people yet.”
After Harry cited an MIT study as showing ~95% little impact, Taneja grants the study merit because enterprise AI needs four things at once: data and infrastructure readiness, models trained on “your secret sauce,” workforce transformation, and “courage at the top.” On the org design: “some humans are going to manage AI agents, some AI agents are going to manage humans — imagine how the org charts have to change.” All four rarely co-occur — “that’s why these things are hitting a wall.”

6. The rollup thesis: offshored for labor, onshored for AI

The one-liner thesis behind GC’s AI rollups: “everywhere you offshored for labor benefit, you’re going to onshore for AI productivity.” Live case: GC says it bought a 3,000-employee call center in the Philippines in a company called Crescendo — and his first question was what those people, and every middle class built on offshore labor, will do next.
Timing: “this is a 5-year problem” — company-building has physics (team, customers, acceleration) — “but 5 years is also not a long time.” The anecdote that lands it: a consulting CEO’s client’s plan to reach “100,000 employees but only 10,000 of them are humans. The rest are AI agents.” Provocative — but “a non-trivial probability” over 10-15 years.
The London thought experiment: if every nurse, lawyer, and accountant here becomes “an AI agent of some company in the United States,” you “hollow out the service sector just like we hollowed out manufacturing jobs for globalization” — Jeannette’s European-champions argument for capturing AI productivity onshore.

7. Governments are betting society slows it down — “I don’t think we have time”

Most impressive: Singapore (“blown away by the depth of thought”), Greece’s prime minister, and meetings with Starmer — but nobody is “thinking comprehensively.” Heads of state take comfort that if AI is that disruptive, “society will just slow it down.” Taneja says market forces are stronger; Harry adds, “I don’t think we have time” unless there’s more intentionality.
Against Harry’s free-market maximalism: “capitalism is a privilege.” Nationalism rose because tech’s productivity gains never passed through to society — and even venture’s boom is smaller than it feels: total value created in venture versus the “max 7” “is noise.”
The policy role he’ll accept: keep a level playing field so healthcare gets a vibrant ecosystem rather than “some company that comes along and just controls healthcare.” Beyond that, “be very free markets oriented.” The deeper worry: build with abundance or it’s “not sustainable in the very long term — you and I will make a lot of money… but what do we create on the other side?”

8. US is well positioned — the question is whether the world buys American

On Trump-era America: “we have energy, we have AI, we have the largest market, the largest entrepreneurial ecosystem… in the short term we’re actually increasing our moats.” The risk sits downstream of tariffs and disrupting the world order: “what is the appetite of the world to embrace companies coming out of the US and let them be global leaders?”
The China race is close: “China and US are very comparable in what they are in AI today — a few months lead here and there.” DeepSeek’s open-source models get used in the US “because they’re better” — capitalism will force adoption of whichever AI wins.
Second-mover advantage, spelled out: a company starting in the GPT-5 era versus GPT-4 gets “this potent force unfair advantage,” because technical debt “used to be on the order of a decade of coding, not a year of coding” — early go-to-market leads may prove “anemic” against fresher stacks.

9. Anthropic at $60B: “risk adjusted, the best price round you could have done”

GC entered less than a year ago at the $60B round with “a few hundred million,” when coding made Anthropic distinguishable: “OpenAI to me is more of a consumer company with ChatGPT… Anthropic kind of became an apps company… with coding as a use case.” It was the first time these bets felt like businesses rather than “some abstract AGI goal” with tremendous burns and dilution. Revenue was “under a billion if I recall,” publicly guided to ~9x — “they’ve done way better than we thought.”
Another few hundred million went in at $180B, 5x oversubscribed — and still, he argues, “probably the cheapest round that got done this year on a multiple basis”: ~20x ARR while being at roughly 10x the scale of peers raising at “50 to 100 times ARR.” Hedged as spoken: “durability of everything in the models is highly unclear.”
The forward math (his comp framing, not company guidance — “you’re not also Anthropic CFO… I’ve got to caution”): if next year is ~$27B, that’s still 200% growth, and the same 20x makes “something like” $550B — “it’s a half a trillion dollar company by the end of next year… if they hit their numbers, I don’t see why that won’t happen, just public market comps.”
On margins: “margins already are not an issue for Anthropic.” A coding agent replaces junior engineers making $80-100K, so pricing power is real; the prize is “$500 billion of payroll is developers… about 10 trillion of white collar jobs.” His analogy: everyone said clouds would commoditize — they run 70s margins. Endgame: “a couple global ones and a couple of sovereign ones in every geo… not everybody’s going to make it.”

10. OpenAI: the structure he overthought — and Microsoft’s trade of the decade

He saw the early round: “Sam’s a force of nature… the guy can bend reality and he has. I just couldn’t get my arms around the structure.” Now: “this is a daily conversation I have with myself… I do regret it” — mostly for the lost front-row learning. It wouldn’t have precluded Anthropic; plenty of investors hold both.
The dilution math off the chart Harry cited: if roughly right, the $200M at the $1B valuation made only ~25x — “our best companies, our first rounds were hundreds of x.” The dilution came from the nonprofit’s share plus compute: Microsoft “maybe took more risk in a lot of ways,” made the highest multiple on ~$20B, got the AI halo and the Azure draft — “an amazing investment. Is that an enduring one? No.” Ambitions now collide, and Harry said Microsoft now uses Anthropic for the majority of its suite.
Harry’s frame, left standing: with stock comp and dilution, this is “the greatest transfer of wealth from venture capitalists to founders and team members.” Taneja’s counter: “never bet against Sam, but he’s doing a lot of things — the phone, the data center, the infrastructure.” Anthropic is focused, hasn’t raised as much, and — citing Arthur at a company likely called Mistral — “you can waste a lot of compute.” Its investors “will probably end up doing better on a multiple basis” than OpenAI’s early ones; Dario’s evolution from research lead to company builder is “really impressive.”

11. Sovereignty Shapes AI Competition

He was the first VC Arthur ever met — a video call taken from a Paris park bench. His feedback: “I’ve never had such a bad pitch, and you are competing against Sam Altman, the mother of all fundraisers. This is not going to end well.” Two years on, Arthur has aggregated capital, the models have “caught up,” and the customer posture turned commercial — “I am bullish on what they will do, even though I was anxious about it.”
The positioning: absent “the overhang of these two monsters” (OpenAI and Anthropic), a company likely called Mistral would be “the hottest startup in the world” on scaling and valuation. And in the West, the company is truly dedicated to enterprise open source — “it’s not Meta, they’re not an enterprise company.”
Harry asked for one success story where sovereignty was the #1 driver. Answer: every US defense prime — likely Lockheed Martin, likely Raytheon, and Boeing — was “built off of sovereignty,” and “AI is that strategic a technology.”

12. “Triple triple double double is definitely dead”

Verbatim: “I tell our investors don’t bring that to me… going from 1 to 3 to 9 to 27 is not interesting. You got to go like 1 to 15 to 20 to 100” — likely Mercor, whose seed GC led, went 1→500M in 17 months. Why it’s possible: for the first time “every CEO in every industry in every country is thinking about what do I do with this technology — cloud wasn’t like that, certainly PCs weren’t, internet wasn’t.”
The open question is durability: “we never had so much scale without just taking durability for granted.” On Lovable (likely; “Anton’s done a great job”) the naysayers ask whether it’ll be around; on likely Mercor GC has “huge conviction” — but “some of these companies that grow really fast will also not be around.”
The 20%-growth SaaS generation — “some founders’ life’s work,” in his partner’s framing — sits in “the purgatory where we need innovation”: too slow for venture, too small for public markets, yet profitable if they stopped investing in sales and marketing. That’s precisely what the customer value fund exists to serve.

13. Price only hurts once — and price-passing is a conviction tell

“This is peak ambiguity.” His advice for navigating it: a true north — GC healthcare tests every decision against proactive/affordable/accessible; Europe against resilience. He pities investors learning the craft now: “you have this great revenue growth to lean on, but no durability… the signals to determine if your decisions were right or wrong — you have none.”
Joel Cutler’s house line: “Price only hurts once” — like a Gucci bag (Harry: “whenever I see my mother with a Chanel bag, I’m reminded of the dent it caused”). In 25 years, “I am yet to see some investor, at least in our firm, ever nail price… we make all the money when it’s better than what we thought.” The dagger: “investors use price as a reason to pass because they couldn’t gain conviction elsewhere — it just makes them sound pragmatic… then you don’t know if you love this company.”
Harry’s capped-upside hypothetical ($2-4B data business at 80 vs 140 pre) gets rejected at the premise: if it’s truly capped, “you shouldn’t be doing it anyways.” When he did Stripe, every payments expert called it niche — “either companies are completely mediocre, or they’re great and you’re not willing to stretch because you’re not willing to believe what the world’s going to look like.”
Concentration is the engine: 200+ investments over 25 years, and “60-70%… it’s like 10 companies.” His recent self-own — congratulating the GC partner who led a now-decacorn: “you’re going to make over a billion dollars on this investment and you’re an idiot — you gave up making the second billion. You didn’t double down.” House rule: he tries not to exceed 10-15% of a fund in one company; truly great companies may force cross-fund investing.

14. LPs: endowments to sovereigns, $16T of retail at the door, zero fee distributions

The strategy break: endowments wanted single-strategy managers and to build portfolios themselves; GC’s counter was “back us to make the founders successful… then we’ll create alpha.” Then came US state pensions (“create wealth for everybody in the US”), and now sovereigns as genuine partners in national AI transformation, not just capital.
Retail: 40 Act evolution and 401(k) changes will open “$16 trillion of retail capital.” He wants a trickle that scales: access to “SpaceX and Stripe, you’re not going to regret it” — but “what you don’t want to do is take retail and put it into the bottom quartile of venture funds that lose money.” Robinhood’s tokenized-access work gets a nod.
The fee stance: GC distributes no fees — everything is reinvested — because partners wanting bigger funds for bigger distributions is “just not the culture we want.” Partners make less than Harry’s floated $3-5M; the filter is “depends on if they’re focused on performance or salary… deliver your dream and you’ll make more money than anywhere else.”
On post-IPO stock: LPs sell programmatically the moment you distribute, so GC holds where its time still compounds value — and paces distributions, since dumping too much at once “could hurt the price of the stock, which hurts the rest of it.”

15. Losing means you’re in the right fights; the one change of mind is indexing

Arriving in the Bay Area: “I lost the Series A of Stripe, of Samsara, of Snap — and the first one I won was the Series A of Gusto.” His Boston partners feared losing would devastate him; his retort: “if I’m not losing, I’m not winning” — the best founders pick among five to seven great firms, so a win rate over 30% means the right fight. To teammates who never lose: “you’re just in the wrong pond, buddy.” He also missed Dropbox entirely — tried to hire Drew Houston instead of funding “this file storage company”; the first million would have returned ~$2B.
The biggest strategic regret is picking over indexing: he did Stripe but said “let’s not do Square,” then tried to pick the winner in AI. “In hindsight we should have just gone and indexed those” — he cited likely Yuri Milner and likely Lightspeed as examples. The logic: “when you know the trend’s going to win, but you don’t know which one’s going to win, you’re better off backing all of them.” Asked what he changed his mind on most in 12 months: “this idea of indexing” — though, joking, he hasn’t moved yet: “I’m going to wait till the next time I miss it in AI.”
The receipts and the endgame: the $500M fund holding Livongo, Snap, Circle, and Gusto is tracking “13 to 15x” — one of GC’s two or three best. Ten-year vision: “a strategic conglomerate where every part of GC is in service of founders.” And his resolution of Harry’s Chanel-vs-Walmart binary: the biggest AUM with the fewest number of companies — “I want the value of the capital you raise to be the biggest but the amount of money you raise to be smallest. That’s when you’ve created the most alpha.” (His compass, from Princeton endowment’s Andy Golden: “run your own race.”)
Quick-fire keepers: money doesn’t make him happy — “a byproduct of the impact I want to create”; parenting in the AI era — “teach them to ask questions, not solve problems”; and the closing horizon: GC will probably invest “300 billion, 500 billion” over 20 years shaping what AI does for society — “I want to get it right… that I did right by the world.”

Dylan Patel on the AI Chip Race - NVIDIA, Intel & the US Government vs. China 刘夜：在“双减”废墟上用 Talkit 重启人生