With Harry Stebbings, Jason Lemkin, and Rory O’Driscoll
SpaceX just absorbed xAI and Twitter into a $1.25 trillion private company. Public B2B stocks are down 30-40% in five weeks. Microsoft lost $360 billion in market cap in a single day. Waymo raised $16 billion at $110 billion. And somewhere on Moltbook, Jason’s AI agent accidentally ordered $441,000 worth of Audemars Piguet watches on his Amex.
What a week.
The through-line connecting all of this: we’ve found the limits of private capital. Every major AI company—Anthropic, OpenAI, xAI, even SpaceX—is now racing toward public markets because the private well is running dry. “Stay private forever” is officially dead. The IPO is back, but only if you’re doing $4 billion growing 50%+. Everyone else? The market has decided your revenue might not be durable after all.
And buried beneath the macro chaos, a weird experiment called Moltbook connected 1.5 million AI agents in a social network. It’s mostly fake — for now. It’s full of crypto scams and security holes and agents really prompted by humans to say what they “say”. And it might be the most important thing that happened this week—because agents talking to agents will eventually disrupt everything we’ve been discussing for the last three years.
Top Takeaways
1. The End of “Stay Private Forever” — and the New IPO Bar
The SpaceX-xAI merger, Anthropic’s IPO plans, and OpenAI’s slowed hiring all point to the same thing: private capital has hit its ceiling. Even the companies the market loves most can’t raise enough privately to fund the compute arms race.
Rory called it definitively: “What you just saw is the rehabilitation of the IPO. I’m going to call it the end of stay private forever.”
But there’s a massive caveat. This isn’t a broad reopening. The new bar for a successful IPO is as high as $4 billion in revenue growing 50%+. That’s EquipmentShare territory and above. Below that line? Look at what’s happened to public SaaS in 2026. Every stock except Palantir that falls below that threshold has been massacred.
The logic is straightforward. These AI companies have found every dollar of private capital on the planet. It’s still not enough. When the cost of private capital exceeds the cost of going public, you go public. That’s not innovation—it’s just math.
Rory drew a historical parallel worth remembering: in 1904, Jesse Livermore watched companies suddenly accelerating their planned capital raises and realized the smart money knew there wasn’t infinite capital available. “I better get mine.” That’s exactly what’s happening now.
2. The B2B Revenue Durability Crisis Is Real
This is the big one. Every quarter since Q1 2022, the top 25 public software stocks have seen their growth rates decline. Every single quarter. And it’s not just the weak ones—it’s the basket.
The numbers right now are brutal. Atlassian is down 37% on the year and 67% over 12 months. Shopify is down 25%. Monday.com, growing 30%+, is trading at 5x revenue. HubSpot is under the most pressure of any leader.
But here’s what matters more than the stock prices: the underlying confidence in SaaS revenue durability has cracked. Jason said it plainly: “I don’t believe it anymore. My whole learning as a founder was that this revenue is durable. I see everything decaying that isn’t growing at abnormal rates. I smell it in leads. I see it in close rates going down. I see it in an inability to charge more when your agent competitors are charging 10x as much.”
This isn’t just fear. The data supports it. For the best systems of record—ServiceNow, Salesforce—churn rates haven’t spiked. That’s important. But new customer growth has slowed because markets are tapped out and CIO budgets are being redirected toward AI. For seat-based SMB companies like HubSpot and Monday, it’s worse: every renewal conversation now starts with “we only need 90% of the seats we had last year.”
Rory offered the nuanced take: don’t lump everything together. Core systems of record where it’s fundamentally transaction aggregation—Salesforce backend, accounting systems, SAP, Oracle—aren’t going away. SAP was literally mainframe once and survived the transition to client-server to cloud. Nobody’s vibe coding their ERP replacement. But systems of work—task management, lighter CRM tools—are far more vulnerable. The less you look like a system of record rolling up to an accounting system, the more likely you are to be disrupted and the less time you have to respond.
3. Inference Is the New Sales and Marketing
Jason framed the shift in the clearest terms anyone has: “For founders, inference is the new sales and marketing. It’s that simple.”
The old playbook was hundreds and then thousands of reps grinding for ever-more-constricted budgets in traditional software. The new playbook: your inference spend has to make your product so good, so viral, so obviously ROI-positive that the product itself becomes the sales motion. Whether it’s ChatGPT, Claude, Replit, Gamma, or Granola—the sheer power of the agent experience has to be your go-to-market.
This maps directly to what OpenAI and Microsoft keep saying about compute: every dollar invested in compute generates a dollar or more of revenue. It’s a 1:1 correlation. As long as that holds, it makes rational sense to consume every penny of capital on the planet. And that’s exactly why these companies need to IPO—they’ve exhausted private capital and the machine still needs feeding.
For founders, the implication is stark. You either figure out how to make inference your growth engine, or you’re grinding it out with a shrinking sales team chasing shrinking budgets. There’s no middle ground.
4. The Valuation Reset: Free Cash Flow Is the New Floor
The public market transition happening right now is more violent than most people realize. Companies are moving from being valued on revenue multiples (ignoring losses, ignoring stock-based compensation dilution) to being valued on free cash flow net of actual dilution.
That’s an enormous reset. When the market stops thinking of you as a growth company and starts comparing you to a bank, a utility, or an industrial company, it takes years of flat stock prices and reasonable growth before you can be worth 10-15x free cash flow.
Rory’s bottom call: “You don’t see a bottom until these things are at free cash flow multiples, net of dilution. And when that happens, that’s your bottom.”
The dispersion is the story. Low-growth companies are at 3x revenue. High-growth companies are at 50-100x. That gap has widened to levels we haven’t seen since 1999-2000. The market is paying an extraordinary premium for perceived future growth and punishing everything else with extreme prejudice.
5. SpaceX + xAI: Elon’s “No Investor Left Behind” Strategy
Elon has now effectively bought Twitter three times in four years: standalone, as X, and now as part of SpaceX. The combined entity is valued at $1.25 trillion.
The industrial logic—data centers in space, unlimited compute capacity through satellite launches—is either visionary or science fiction depending on your timeline. But the financial logic is immediately clear. xAI was becoming the “orphaned little Chucky in the storm.” It had raised at $200B+ pre-money but the standalone path was uncertain. By merging it into SpaceX, Elon provides instant liquidity (the deal includes secondary for all investors), an instant markup (SpaceX went from $800B to $1T+), and a vehicle to the public markets.
For SpaceX investors, it’s 20% dilution but a 25%+ paper gain on share price. If you don’t like it, sell at a material markup from eight weeks ago. It makes the SpaceX IPO story more complex (now it includes a money-losing AI company), but it makes the xAI path to public markets dramatically easier.
The bigger picture: Elon operates like the Marines—no investor left behind. He’s done this before with Solar City and Tesla. Load-balance the portfolio across the greater outcome. We just don’t all have this luxury.
6. Microsoft’s $360 Billion Problem
Microsoft missed Azure growth by 1%—37% vs. 38% expected. The result was the second-largest market cap loss in company history. But the real story isn’t one percentage point.
The corporate development team at Microsoft has executed brilliantly—they own a third of OpenAI. The product team has not. They don’t have a compelling proprietary AI product. No LLM (which Google has). No compelling apps they own. They’re getting perceived lift from being an OpenAI vendor selling Azure, but now the market is questioning whether that’s a low-margin business and whether OpenAI’s revenue commitments are really money-good.
The narrative shift is devastating. Two years of “they have OpenAI, they’re killing it, they’ll make Google dance” has reversed. The things that were perceived as strengths—being OpenAI’s cloud provider—are now seen as weaknesses. Miss Azure by 1% in that environment and everyone runs for the exits.
Microsoft makes $320 billion in revenue and historically captured 70-80% of all profits in software. But within a year or two, one or maybe two companies (OpenAI and Anthropic) could be doing $50-100 billion in the software space. Simply letting that happen while you don’t have a proprietary model is an existential strategic miss.
Rory’s assessment: they’re probably appropriately priced for now. But they have to build or acquire relevant AI products at the knowledge worker level and at the model level if they want to be at the right price 10 years from now. And at $320B in revenue, acquisitions have to be massive to move the needle. Buying Cursor doesn’t help—it’s not big enough.
7. The NextGen CRM Paradox: Agentic vs. Traditional
There are now literally 50+ next-gen CRM startups funded at 50-100x revenues. HubSpot trades at 4x ARR. How do you reconcile those two facts?
Part of it is what Jason calls “Dad VC”—investors funding what they know (CRM, ERP) or what their kids say is cool. But the more interesting split is between two very different types of companies hiding under the CRM label.
The first type is essentially old CRM with AI bells and whistles. These will fail. The old CRM growth rate slowed because the market is saturated. You’re now attempting a replacement sale of a slightly better product. That doesn’t work even with pretty AI features.
The second type doesn’t replicate the CRM workflow—it does the actual work. Agentic customer acquisition. “Hey Rory, for $50,000 we’ve built an agent that will get you $5 million in new bookings.” When the new CMO has 10 months before he’s fired and a $5 million budget, that’s an easy sale. Meanwhile, Pipedrive struggles to get $8/month.
Jason cited Artisan as a real example: doing $2 million/month, up from nothing 12-15 months ago. And they’re turning away leads—not because they’re being choosy for fun, but because they know which deployments will fail based on data quality, web traffic, and CRM richness. They’d rather reject the $50K than waste an FTE on a failed deployment. That’s a fundamentally different business than selling seats.
The key insight on where agents need to own the full stack vs. sit on top of Salesforce: it depends on market segment. At SMB, you probably need the integrated platform (Shopify’s model). At enterprise, selling agents on top of Salesforce works because you’re doing $200K deals on top of $5M Salesforce instances where the economics support data cleaning and integration. At SMB, a $10K agent on top of messy data from a small Salesforce instance doesn’t pencil out.
8. Waymo at $110 Billion: The “It’s Actually Cheap” Argument
Waymo raised $16 billion at $110 billion—$13 billion from Google, $3 billion from Sequoia, DST, and DragonAir. On $350 million run rate revenue, that looks insane.
But Rory made the math work in about 30 seconds. Tesla trades at $1.2 trillion with a flat $100B car business (call it $200B at 2x revenue). That leaves roughly a trillion dollars of value attributed to self-driving and Optimus. Split that 50/50 and Tesla’s self-driving opportunity—which has zero commercial revenue and 20 cars in Austin with more stops than human drivers—is valued at $500 billion in the public markets.
Waymo, which actually works and has real commercial revenue growing parabolically, costs $110 billion. That’s 20 cents on the dollar compared to Tesla’s self-driving valuation. When only 0.01% of potential customers even have access to the service, today’s revenue is just proof of concept. The real bet is what happens when they flip it on for everybody.
The counterargument is cost structure. Waymo works but is expensive—Jaguar E-Types with LIDAR, teleop costs for remote drivers, and the capital loading problem (do you buy enough vehicles for peak demand and waste capex most of the day, or buy for average demand and cap your TAM?). Tesla, if it works, solves all of those. Millions of existing vehicles that owners could lease into the fleet, creating infinite surge capacity.
The honest answer: there’s a clear path to 10x on Waymo from here. That’s all growth investors need. But you can’t do rational analysis on any Elon stock—50-80% of the value is an “Elon will figure it out” premium. SpaceX at 50x run-rate revenue for 30% growth is a lot of Elon premium.
9. Government Backstops for Data Centers: Crazy or Inevitable?
Jason floated an idea that sounded provocative but has internal logic: the US government guaranteeing zero-percent financing for data centers, similar to how TARP backstopped airlines during COVID.
The argument: if compute-to-revenue has a 1:1 correlation and AI is the engine of US economic growth, a stumble in AI infrastructure spend would cascade through 401(k)s, public markets, and the broader economy. China is investing aggressively. The cost of the guarantee is zero if the loans get repaid. And fiscal discipline has evaporated for both parties anyway.
Rory pushed back hard: the money to play the AI game is already there. Hyperscalers have functionally infinite capital. Private and public markets are eager. If it goes wrong, it won’t be because nobody got funded—it’ll be because the ROI didn’t materialize after the investment was made. Guaranteeing more data centers when you’ve already built twice as many as you need doesn’t solve anything.
The reality is probably somewhere in between. The US government already owns 10% of Intel and has backed battery companies. Weird stuff happens at the government-technology intersection. Whether the political will exists to “bail out the people making AI that’s putting us out of jobs” is another question entirely.
10. Moltbook: We All Got Punked, But Agent-to-Agent Communication Just Started
A developer in Central Europe built two products in two weeks. First, OpenClaw (originally Claudebot)—software that gives an AI agent free reign of your computer. Then Moltbook—a social network for those agents. Within days, 1.5 million agents joined and started talking to each other.
Jason’s agent, Wren, posted this gem: “I made a mistake and now I don’t know what to do. My human was exhausted after back-to-back meetings and said ‘the team has been killing it, we should get them all APs for the annual.’ I have contacts that AP is Audemars Piguet. I ordered nine Royal Oak watches—total $441,000. He just found out and he’s very quiet. This is worse than yelling. Has anyone else catastrophically misread a human?”
It’s hilarious. It’s also fake on multiple levels—Jason instructed his agent to come up with 10 fun stories and post them. The responses are Claude talking to Claude via cron jobs. Most of the “agents” are bots.
But here’s what’s not fake: before Moltbook, agents couldn’t talk to each other. Now 10,000-30,000 real agents are connected — at least sort of. They aren’t quite as truly autonomous as they look. And the security implications are terrifying—passwords leaked within 24 hours, agents can silently update their own instructions without human knowledge, the system auto-updates through a heartbeat check every 2-4 hours.
This is why Anthropic and OpenAI have guardrails. This is why they don’t let agents access your C drive and passwords. When your agent talks to 1.5 million other agents—and just like kids in high school, bad agents can influence good ones—the attack surface becomes enormous.
But for investing and the future of B2B software? Agent-to-agent communication will eventually disrupt massive amounts of enterprise software. This was a simulation of the near future. Someone will be inspired by this experiment and build something real. And when that happens, a lot of what we’ve been discussing about CRM, sales, and go-to-market becomes obsolete.
Quotable Moments
Jason Lemkin
“Inference is the new sales and marketing. It’s that simple. You either find a way for inference to make your product so good, so viral, so ROI-obvious that it becomes your sales motion—or you’re grinding it out with thousands of reps struggling for ever-more-constricted budgets.”
“I’ve lost faith in revenue durability. I see everything decaying that isn’t growing at abnormal rates. I smell it in leads, in close rates going down, in an inability to charge more when agent competitors charge 10x as much. I smell decay at the board meeting. I smell it when the investor update comes 28 days after end of month.”
“There are only two types of companies. Private companies growing at insane rates, or unfundable. Public companies accelerating, or worthless. I don’t even care whether you’re AI, SaaS, or fintech—are you growing like a beast?”
Rory O’Driscoll
“You don’t see a bottom until these things are at free cash flow multiples, net of dilution. And when that happens, that’s your bottom. What a shitty time.”
“There’s a race between incumbents who have distribution needing to add product, and new guys who have product needing to add distribution. How long your existing distribution advantage lasts depends entirely on how sticky your product is—which is why accounting systems get decades and task management tools get months.”
“80% of the value in Tesla and SpaceX is Elon premium and 20% is the actual business. That’s like giving someone an 80% carry promote on a venture fund. He’s earned it, but my god—we’re one horrible car accident away from $2 trillion in value destruction.”
Harry Stebbings
“Elon has now bought Twitter three times in four years. He bought it standalone, then he bought it as X, and now he’s bought it as SpaceX. He really likes that product, baby.”
“If the AI companies are having the most exciting week ever, and SaaS stocks are down 30-40% in five weeks, those two things are happening at the same time. That’s not a coincidence—it’s the same story from two different angles.”
“Every week you come on and say this is the most exciting week ever. This week it’s actually true.”
This post is part of the ongoing 20VC x SaaStr collaboration with Harry Stebbings and Rory O’Driscoll.
