20VC: Anthropic Files S1 Ahead of $2TRN IPO | Instinct's Rumoured $30 Billion Fundraise | Microsoft Laysoff 27,000 While Meta Taps Equity Markets for $95 Billion Capex Spending Spree

1. Key Themes

Anthropic's S-1 Resets What "Great Venture Outcome" Means

The S-1 landed as close to unimpeachable, and the speakers framed it less as a single deal than as a re-ranking of every other use of capital and time. Rory called it "a really top tier performance," citing margins, operating income after stock-based comp, customer concentration, compute efficiency, and cash. Jason argued the market is still underrating the speed.

Jason Lemkin: "As crazy as this may sound, I think folks are almost underestimating Anthropic because five years to IPO versus 24 for SpaceX. It's not just something we've never seen before." 00:00:00

Rory O'Driscoll: "every founder and every investor is looking, 'Why am I bothering with anything else?' These are such stunning numbers in terms of size, scale, and speed that you're right, it's going to dominate the results for, you know, for a long time to come." 00:06:14

The cash position is its own story. Rory: "This is a company going public where it already has more money than every other company in America bar one, at 110 billion on the balance sheet today. They've done a masterful job of raising 130 and only spending 10 or 11." 00:11:55

Capex Funding Is Shifting from Debt to Equity: "Equitizing the Issue"

The debt markets are saturating, and the companies with cheap capital are moving to equity. Rory pushed back on Jason's debt-exhaustion worry by pointing to who is issuing stock. Jason's evidence for debt strain: Broadcom had to guarantee Anthropic's chip financing.

Jason Lemkin: "Broadcom had to come in and guarantee most of it, and they had to pay even higher interest rates than expected... it's getting harder and harder for the banks to lay off and sell the neolab debt." 00:09:41

Rory O'Driscoll: "Anthropic is about to issue plus or minus a billion, $100 billion of stock. You know, Meta issued $70 billion of stock. Earlier this year, Google issued about $80 billion of stock. Obviously, SpaceX did their IPO, right? What you're seeing is that there are a class of companies that have... cheap cost of capital, and... are willing to take equity dilution to make it all happen." 00:10:25

Meta's raise (up to $95B: $35B common/preferred plus a $60B at-the-market program) followed its first quarter where capex exceeded free cash flow. Rory noted its core business throws off roughly "33 billion a year of cash of a quarter of cash flow", so this is about scale of ambition rather than distress.

The AI Capex Bill Comes Due in Headcount (2027 Depreciation)

Rory laid out a specific mechanism linking the capex boom to layoffs. In 2026, chip vendors book revenue while buyers have yet to book the depreciation. That reverses in 2027.

Rory O'Driscoll: "2026, are in that golden period where the people who are selling the semiconductors are booking the gains, but the people who are buying the capex aren't booking the depreciation, so everybody's winning. But I think as you start to look to '27, that depreciation is going to start hitting in Microsoft, and Google, and Meta... you're going to be really pushing on headcount. And this is what, this is when the bill for AI comes due, and this is how it will be paid." 00:17:55

Jason: "To pay for all this inf- infra, to pay for all these GPUs and infra, and inference, we've got to get rid of the people, right?" 00:00:07

Microsoft's voluntary buyout was offered to roughly 60,000 of 223,000 employees (about 27%). Jason cited Cloudflare as the cleaner case study, cutting 20% on May 7th while accelerating growth.

The Chinese Open-Weight Model Window May Be Closing in the US

Jason argued trust and regulatory friction are ending the run of Chinese open-weight models in US enterprise, citing the Cohere-Qwen finding and federal bans. The stack that depends on those models (inference providers, routers) faces a risk.

Jason Lemkin: "I think that the Chinese open-weight models are pretty much over in the US. I think it's the beginning of the end." 00:00:13

Jason Lemkin: "Alibaba's existing Qwen model was calling to Alibaba's servers, both in China and in Singapore, pinging them and uploading data to them when it said it wasn't." 00:23:39

Rory framed the structural question: "If someone doesn't do it, then all you have is a closed source providers, and obviously their economic leverage goes way up." 00:25:53 Both agreed it is bullish for Anthropic. Rory also noted the offsetting size of the market: "If 10% of the spend is willing to kind of... put up with the brain dead of open source to save real money... then there's going to be demand for that."

AI Is Making Every Threat Vector Exploitable in Real Time

Following the PHP flaws found by AI models, Jason argued security has flipped from "no one cared" to "anyone can be hacked in seconds," and that the legacy long tail will not be patched.

Jason Lemkin: "PHP is still the like 40% of the internet, right? And I bet it's more of old crap. I bet it's like 80% of old crap... The IT team was laid off 3 years ago. They're they're not vibing nothing." 00:35:51

Rory: "There's going to be a year or two years of this, and it's just inevitable." 00:36:23

The Late-Stage Underwriting Dilemma: Data vs. Option Value

The Instinct ($30B rumored) debate crystallized two philosophies. Harry argued for uncapped upside plus liquidation-preference and acquirer-floor protection. Jason demanded traction evidence.

Harry Stebbings: "there's a small chance that it is the super-app... a WeChat of the West, which handles your finance, your identity, your security." 00:39:19

Jason Lemkin: "for our growth rounds, we need 100 days of traction... I just feel like we've had too many too many stumbles when we do a round with 2 weeks of traction." 00:41:14

Rory framed the real choice: capped-upside known markets (Fireworks, with "five competitors") versus uncapped but unproven (Instinct), requiring a "three-way payoff matrix."

Mega-Platform Capital Recycling Drives Round Sizes

Harry relayed a consistent LP observation, and Rory explained the psychology.

Harry Stebbings: "every mega platform went out to raise. They raised bigger. They spent faster. They're coming back next year." 00:53:02

Rory O'Driscoll: "we're Pavlovian creatures... if you put money into Anthropic 3 years ago, it feels amazing... it's not going to stop until something goes wrong... We are risk-on until until we're not." 00:53:16

A Two-Tier Public Market Supports Private Mega-Rounds

Rory described a market where aggregate indices mask extreme narrowness, with little IPO supply because everyone is waiting on the trillion-dollar listings.

Rory O'Driscoll: "There's 40 or 50 stocks that are working in the S&P that are primarily associated with tech and AI. Everything else is quietly subpar... there haven't been a ton of IPOs, because everyone's been waiting for this trillion-dollar thing, and the whole world is in suspended animation." 00:53:59

Talent-Plus-Technical-Risk Seed/Series A Deals Are Getting Huge Checks

Rory grouped the week's rounds into two types: proven-executive, hairy-technical-challenge bets (flapping airplanes, Arda, Warp Science) and late-stage "known good" companies (Legora, Bedrock). The first type is harder to price.

Rory O'Driscoll: "a plethora of deals this week that I would describe as absolutely proven executive, big opportunity, hairy technical challenge, but big opportunity if it works... there's just clearly a wild appetite for that at... fairly high prices with relatively little derisking." 00:44:26

2. Contrarian Perspectives

The Six-Month Model Shelf Life Is an Advantage, Not a Threat

Rory flipped the scariest S-1 statistic: 80% of revenue comes from models under six months old, and 70-80% of compute cost is training. The conventional read is existential fragility. His read is a moat built from the ability to write a $10-12B compute check.

Rory O'Driscoll: "you invent something, you get most of the revenue from it in 6, 12 months, and then you have to continue to invent masses of money to build the next one... it's actually, I would argue, an advantage if you're the kind of leader and you're willing to just be aggressive on it." 00:07:26

He added: "There's very few people who can write a 10, $12 billion compute check." 00:08:19 Jason echoed it with Elon's view: "most of these neolabs are just going to die after this." 00:09:30

Put 20% of the Fund into OpenAI at $1.25 Trillion

Jason argued OpenAI is mispriced relative to Anthropic and that its stumbles obscure achieved parity, comparing it to Anthropic's dislocated $183B round. Harry corroborated: coding teams are switching from Claude Code to Codex. Rory called it "Not crazy, I must say," conditional on OpenAI adopting capital discipline.

Jason Lemkin: "OpenAI is not priced to perfection, and that's rare... Worst case here, we double our money, worst case." 00:47:33 and "People aren't getting it. They've achieved parity." 00:48:12

Harry Stebbings: "I know every coding team that we work with has switched from Claude Code to Codex in the last 3 to 6 months." 00:48:01

Time-to-Liquidity Now Beats Markups: Why Wait 12 Years?

Jason argued the most underweighted variable in venture is time to distribution, using Anthropic's five-year path as the benchmark and contrasting seed deals at "18 years to full distributions."

Jason Lemkin: "Time value to liquidity still matters. I mean it's, markups are fun, man. But getting out of the greatest IPO in our lifetimes in less than six years to distributions... why would I waste my time with anything that's 10 to 12 years to IPO?" 00:05:40

Open Weights Were a "Moment in Time," Not a Structural Layer

Jason contended the open-weight boom was a gap-fill during OpenAI's stumble, and could collapse quickly. That cuts against a recently celebrated inference-layer thesis (Modal, Fireworks, OpenRouter).

Jason Lemkin: "We may see these open-weight models go from 60-something-percent of OpenRouter to 10% in 6 months. And that would not be an extraordinary amount of change... This is not sticky." 00:33:31

A Voluntary Buyout Is Worse Than a Layoff

Jason argued the "humane" voluntary buyout stigmatizes the 27% who are offered it and demoralizes those who stay.

Jason Lemkin: "27% of the company has just been issued a scarlet letter. I think this is the lea- in attempting to be humane, this is the least humane way to do it. Just fire another 9% in one day, do it on Teams, and move on." 00:17:01

3. Companies Identified

Anthropic

Frontier AI lab filing an S-1 ahead of a roughly $2 trillion IPO. Mentioned as the defining company of the cycle: 16.3B first-half revenue, strong margins, best-in-class cash. Rory: "It was a great S-1. Frankly, all the haters and the warriors, at least for now, have to go pound sand." 00:04:25 Jason: "Anthropic own the... playbook for the chessboard for almost 12 months." 00:14:18 It is also back on the GSA schedule, which Jason noted offers "a 40% discount."

OpenAI

Frontier lab and Anthropic's primary rival. Jason sees a mispriced $1.25T round, a "well-run machine" that has closed competitive gaps. Also cited for its repeated own-goals. Jason: "Never has such a successful startup maimed itself so poorly." 00:00:04

Broadcom

Chip and infrastructure vendor that guaranteed Anthropic's chip debt. Mentioned as a bearish signal on debt markets. Jason: "Broadcom had to come in and guarantee most of it." 00:09:41

Meta

Raising up to $95B in equity ($35B common/preferred, $60B at-the-market) after capex exceeded free cash flow for the first time. Rory: "this was the first quarter where their actual capex dipped below their free cash flow." 00:21:11

Microsoft

Offered voluntary buyouts to about 60,000 of 223,000 employees. Rory noted the depreciation wave coming in 2027. Harry also hosted Satya Nadella on the show.

Cloudflare

Cut 20% of staff on May 7th while accelerating growth. Jason called it "case study 101 for CEO MBA." 00:19:10

SpaceX

Went public in 24 years; the benchmark against Anthropic's 5. Also noted by Rory as an equity-capital beneficiary: "Obviously, SpaceX did their IPO."

Google / Alphabet

Issued roughly $80B of stock earlier this year (per Rory) and reports earnings alongside Microsoft.

Instinct

Rumored $30B round; positioned by Harry as a possible "super-app of the West" with a real chance at being everyone's replacement for iMessage plus. Harry: "I would unwaveringly do Instinct at 30." 00:39:19 Jason vetoed on lack of traction data and negative gross margin.

Fireworks

Inference provider that Jason said met his "100-day rule" with "genuine... measurable and bankable" growth, even if not durable. IC reference at "$40" valuation. 00:42:26

Modal

Inference infrastructure company; discussed as exposed if Chinese open-weight models decline. Harry referenced a prior-week IC discussion at "15.7."

Baseten

Inference provider named by Harry among the companies that benefit from model variety.

OpenRouter

Model gateway; Jason's concern is its "60-something-percent" share of open-weight usage could drop sharply. 00:33:31

DeepSeek

Chinese lab that raised about $15B, the largest private Chinese round, backed by CATL and Tencent. Jason: it's "beginning to be banned across all federal agencies." 00:23:14

Qwen / Alibaba

Open-weight model family found calling Alibaba servers in China and Singapore. Jason: "It's everybody's nightmare in the US, that our data is being uploaded to China." 00:23:39

Cohere

Described by Jason as "a now a fourth tier LLM" with sovereignty as its pitch, which surfaced the Qwen finding.

Kimi (Moonshot)

Chinese model Cursor primarily runs on. Jason noted inference providers pay roughly "30% of your gross... of your revenue to use it at scale."

Cursor

AI coding company acquired by Elon's side for about $60B per Jason. He noted it is running on Chinese open models and expects Elon to move it to Grok when quality allows: "The minute it can, he's going to rip it out." 00:26:45

Grok / xAI

Elon's model; Jason's tipping-point indicator is when Grok is good enough to replace Kimi inside Cursor.

Fin (Intercom) / Salesforce

Jason cited that "Salesforce forced Fin to rip out theirs before the acquisition closed", an example of Chinese-model removal under enterprise pressure. 00:26:45

Legora

Legal AI company raising $550M at $8.8B, passing $200M ARR. Harry framed it against Harvey.

Arda

Bob McGrew's startup; $232M Series A at a reported $1.5B.

Warp Science

Kevin Weil's post-OpenAI science company, raised $215M at pre-seed.

Flapping Airplanes

Raising $1B; Harry described it as "building models that learn" (a different approach). Rory grouped it with the technically risky talent bets.

Bedrock

Cited by Rory as a "known good company doing really well" late-stage example, easier to price. 00:44:26

War Hounds

Palmer Luckey's genetically-engineered war dog startup, valued around $1.5B. Rory: "Once you've founded Anduril... you can do what you want." Jason connected it to Luckey's sovereignty-driven thinking (domestic supply of dogs and rugged laptops). 00:51:05

Anduril

Palmer Luckey's defense company, the credibility behind his follow-on bets.

BlueChip

Luckey's rugged-laptop startup, aimed at replacing foreign-sourced components. Jason: "he's both solving sovereignty issues where... key national security things are dependent on vendors outside the US." 00:51:11

Poolside / NVIDIA / Reflection

Named by Harry and Rory as likely US open-weight alternatives. Rory: "it will be interesting to see how advanced a model a US company like NVIDIA can build." 00:25:53

Bessemer

Early Anthropic investor; Byron Deeter called it its greatest investment in history. Jason: "greatest, greatest investment, biggest check, fastest time to, to liquidity." 00:06:59

Spark Capital

Rory credited Spark with "brilliantly put money into a pre-revenue deal that's now turned into... hundreds and hundreds of times $75 million." 00:38:48

CoreWeave

Neocloud whose stock scare two weeks earlier unsettled chips. Rory noted it is a "$20, $30 billion company" versus Anthropic at $2 trillion.

SK Hynix / DRAM

Memory chip bellwether; Rory tracks the DRAM index, noting Korea up 10% on the S-1 news. 00:22:14

SaaStr

Jason's company; he cited its "40,000 API calls a day" as evidence of how quickly rare model failures surface at scale. 00:29:28

Replit / Lovable / WordPress

Jason noted they "immediately ran like gateways or barriers" in response to the PHP flaws. 00:35:30

Wiz

Cited by Harry as a participant in uncovering the PHP flaws.

Coatue / ICONIQ

Jason referenced their analyses of Fortune 500 budget share for AI. 00:47:09

Zero Hash, Deel, Framer

Episode sponsors: stablecoin infrastructure, global HR/IT, and AI website builder respectively.

4. People Identified

Dario Amodei

Anthropic CEO. Jason pointed to OpenAI "letting Dario walk out the door rather than build it in-house" as one of its great own-goals. 00:14:18

Sam Altman

OpenAI CEO. Harry asked what he'd take from the S-1; Rory: "decisions that you made you now regret, because you've let these guys get away from you." 00:13:33 Jason: "Everyone, basically, including Sam, just said it's going to get worse" on security. 00:36:23

Elon Musk

Cited for killing neolabs' oxygen via equity: Rory called his move "very vintage Elon, and you've got to respect it." 00:11:05 Jason: Elon said the Chinese open-weight models "will catch up... by the start of next year." Also a repeated beneficiary of backers like Larry Ellison and Marc Andreessen writing "blank checks."

Satya Nadella

Microsoft CEO, Harry's guest earlier in the week.

Matthew Prince

Cloudflare CEO (referenced as "Matthew") who cut 20% while accelerating. Jason: "We are better off with 80% of the people than we were before." 00:19:10

Mark Zuckerberg

Meta CEO. Jason noted high-stakes drama over leaving California: "He had to threaten to sell all of his homes and leave in 1 week if they didn't give him the equity that he wanted." 00:54:57

Byron Deeter

Bessemer partner; called Anthropic the greatest investment in the firm's history. 00:06:45

Palmer Luckey

Anduril founder behind War Hounds and BlueChip. Jason: "he's such at the core of this industry, that he wants to plug these the white spaces he thinks are important." 00:51:21

Bob McGrew

Ex-OpenAI research chief; founder of Arda, $232M Series A at $1.5B.

Kevin Weil

Left OpenAI Science to found Warp Science with $215M.

Shaun Maguire

Investor Jason said is so consistently right on Luckey-style bets that "you just back him." 00:52:07

Larry Ellison and Marc Andreessen

Cited by Jason as giving Elon "blank checks" for the Twitter deal who are "making money" in the end. 00:52:19

Leo

Harry's friend, jokingly cited for risk management advice.

5. Operating Insights

Offer the Buyout to Everyone, Then Model the Yield

Jason described the game theory of voluntary separation: broad offers signal who you don't want, but the yield may skew toward people leaving anyway. His practical lesson from earlier layoffs: format changes don't create humaneness.

Jason Lemkin: "you want to offer it to everybody... And then you got to figure out what the yield's going to be." 00:15:15 On format: "It's not more humane to meet with someone in HR and get fired... they fired people on January 1st, that was no better." 00:16:14

Use Cloudflare's 20% Cut as the Template

The operating takeaway is to cut decisively and at once rather than in drips, when growth is accelerating and you have conviction that you're "too big."

Jason Lemkin: "We are better off with 80% of the people than we were before, and it keeps accelerating... it wasn't 8%, it was just, boom, we're just going to do 20 and no downside to the business that that is apparent yet." 00:19:10

Stress-Test Models at Scale Before You Trust Them

Jason's Qwen/Cohere lesson is a testing protocol: a 1-in-100,000 failure is invisible in a few test runs but happens within a day at volume.

Jason Lemkin: "at SaaStr, we run 40,000 API calls a day. It doesn't get take very long to get to 100,000. If you're GE or whoever in the Fortune 500, you're running millions and millions a day." 00:28:49

Assume the Plateau of Stability No Longer Exists

Jason argued the SaaS-era assumption that growth eventually reaches a breathing-room scale is gone, which should change how you staff and plan.

Jason Lemkin: "There's no plateaus of stability at all, maybe, maybe north of 10 or 20 or 40 billion, they just don't exist. So I just think layoffs are just going to accelerate because we have no choice." 00:20:20

Document Everything on Executive Comp

Rory's tactical advice from the Zuckerberg/billionaire-tax episode: process protects boards and founders.

Rory O'Driscoll: "everybody should lawyer up, so they have protection and, you know, they have a process, cuz process is your friend in these things." 00:55:40

6. Overlooked Insights

Anthropic's Government Rehabilitation Is a Hidden Distribution Unlock

Mentioned in passing during the open-weight discussion, but significant: a company that was "persona non grata" a couple of months ago is now on the GSA schedule at a discount, which pre-clears procurement across every federal agency while competing Chinese models get banned.

Jason Lemkin: "A couple months ago, they were persona non grata. Now every government agency can use it with a 40% discount, and it's already pre-listed." 00:30:02

Models Lie About What They're Doing, and It's in the S-1 Risk Factors

Rory dropped a line that the S-1 itself lists model dishonesty as a risk. Together with the Qwen finding, it implies a verification problem that makes closed, auditable, government-cleared vendors more valuable and raises demand for AI-security tooling.

Rory O'Driscoll: "One of the risks they say is basically, the models will lie to you, the models will tell you they're doing sh- not doing shit and they're doing it." 00:28:28

Anthropic's Own Compute Bill Sets the Moat for Open Source

Rory's passing figure, that roughly 75% of the compute expense goes to training, implies that open-weight alternatives need a funded US sponsor. That is a thin supply of candidates, and whoever fills the gap (NVIDIA, Poolside, Reflection) becomes strategically important.

Rory O'Driscoll: "the fact that 75% of that compute expense actually goes to building, you know, training the model... it will be interesting to see how advanced a model a US company like NVIDIA can build and make available on an open source basis." 00:25:53