Leo Aschenbrenner's Situational Awareness Blows Up | Moonshot AI Raises $3.5B at $35B

Summary

• Airtable sold to Bending Spoons for $1.285B — $485M revenue growing 20%, roughly 2.8x sales against an $11B 2021 mark — and the real shocker per Jason Lemkin wasn’t the price but that no PE firm counterbid despite Francisco Partners raising $22B to do deals like this. Nikesh Arora’s read: PE has “a full roster of stuff they’d like to sell to Bending Spoons,” not buy against them. Jason’s warning for boards: this may trigger “a quiet wave of airtabling” — founders and investors at 20% growth quietly capitulating.
• Leo [likely Aschenbrenner] was “absolutely right on trend, absolutely wrong on portfolio construction” — Nikesh’s verdict on the Situational Awareness fund’s implosion: a $225M vehicle that at one point held $45B of assets on 4x leverage, its public book bought by Ken Griffin’s Citadel for a reported $16B (Ken reportedly up ~$3B). “It’s almost like it was inevitable.” Rory’s cost-basis point: early LPs (Collisons, day one) still made money; “if you bought in in April, May or June, you’ve been wiped” — and late investors may read the docs and litigate.
• Anthropic’s model breaching three companies could start a security supercycle. Nikesh: the average zero-day found in the wild takes 55 days to patch while models now find vulnerabilities “in split seconds”; the capability he predicted in six months arrived in four, and open-source distilled attacker models are 2-3 months out. “I spent eight years trying to get CEOs to talk about cybersecurity… Dario did it in one fell swoop.” Rory’s translation: last year’s security stack is wholly unfit for purpose — enterprises are about to buy a lot more.
• “Average intelligence is going to be free and the average intelligence will get smarter” — exceptional intelligence (cancer cures, space data centers) gets paid for; nobody pays $6 per million tokens for customer support. The corollary: “over the next 3-4 years we won’t be paying for intelligence, we’ll be paying for compute through our nose,” and “land, permits, energy — this is the thing that is going to get priced for the next 3 to 5 years.” Even chicken-manure methane is selling to hyperscalers at a multiple.
• Context, not model choice, becomes the moat. Nikesh is betting the next 3-5 years on organizational context — vector DBs, transcribed customer cases, “more people collecting context than I’ve ever had” — so any frontier model can be swapped in. The coming war: Satya wants context in a harness beside commoditized models; every model company wants it inside the model. For enterprises that can’t manage the switch: “Bending Spoons. Bending Spoons.”
• Q2 was bullish for anyone selling inference: Google Cloud +82%, AWS +37% at scale, Microsoft +20-30% — roughly $100B of new annualized revenue across four companies — while Meta spent without an obvious payoff and got marked down. Palantir grew ~100% with bookings +153% off 1,049 customers. Nikesh on the ~$1T of committed capex: not a demand problem, a timing problem — “one more run at the roulette table.”
• If OpenAI and Anthropic miss their 2027 numbers, that’s a dislocation, not demand destruction — “that’s called a buying opportunity.” Rory’s scenario: Moonshot (fresh $3.5B raise at $35B) becomes the model of choice on the same compute at 10-cents-on-the-dollar tokens — “Moonshot is happy, Nvidia is happy, enterprise is happy, OpenAI very very sad.”

Deep dive

1. Airtable at $1.285B: a great outcome nobody wanted — and nobody outbid

• The facts: $485M revenue growing 20% year-on-year, sold for $1.285B against an $11B 2021 price. Rory’s both-sides framing — for the buyer, “I’m buying stuff at 2.8 while I’m trading in the market at naught to 10 times revenues. They’re going to do this all day, every day.” And strip the anchor: a company built to $450M+ in revenue and sold for ~$2B-ish over ten years “is a great value creation achievement” — it only feels like a lowball off the 2021 mark.
• Jason’s shocker wasn’t price but the empty auction room: no Thoma Bravo, no Vista, despite Francisco Partners raising $22B to do deals like this — “20% at $500 million with some AI dust… I just assume someone would outbid them.” Nikesh’s supply-demand answer: the PE firms “might have a full roster of stuff they’d like to sell to Bending Spoons as opposed to buy against Bending Spoons.”
• Rory, harsher: “Anyone in PE who’s done five software restructurings in the last 12 months needs a sixth like a hole in the head” — and horizontal individual-user productivity apps are among the toughest PE categories anyway. His resolution: capitalism found the best owner, Evernote-style. “I bet you two years from now it’s doing 600, not 900, but I bet you it’s $300 million of free cash flow.”
• Jason’s marker for the market: either this is forgotten in three hours, or it starts “a quiet wave of airtabling” — boards at 20% growth and nine figures who’ve been saying “we’re going all in, guys” quietly capitulating. “Everyone capitulated: the late stage got 1x, the founders made $150M — a lot less than they thought.”

2. The Mercedes / Tesla / Waymo test — and venture’s clock-speed problem

• Nikesh’s operating question for every AI-infused incumbent, asked of his own team daily: are you Mercedes (“sprinkle a little AI in the car”), Tesla (drives ten exits, grab the wheel occasionally), or Waymo? “My biggest fear is a bunch of people out there in garages getting funded by Harry and Rory who are going to build the Waymos of the future, and we’ll be busy putting lipstick on the pig.”
• Rory on why Airtable lost momentum: it was brilliantly early to no-code — a database turned into a spreadsheet — but “Supabase is doing a million Postgres databases a week”; today’s nerd who wants a custom CRM “just goes to Lovable, Replit or Claude Code and bangs it out from scratch.”
• Nikesh’s structural observation, worth keeping: “the holding period of venture is now longer than the technology platform change cycle.” Twenty years ago Airtable would long since have been public common stock and been hoovered up; now the question is whether a 10-year-old company is past the point of rebuilding versus starting from scratch.
• On founder fatigue, the human read: founded 2013, already did the layoffs, got profitable, went founder-mode, clawed back to 20% growth — “it’s tough to not tap out. We’re human beings.”

3. Leo [likely Aschenbrenner]: right on trend, inevitable on math

• The blowup: the Situational Awareness memo parlayed into a $225M vehicle that at one point had $45B of assets, run with 4x leverage; Ken Griffin and Citadel bought the public book for a reported $16B, reportedly making ~$3B “in a very short amount of time.” Nikesh’s verdict: “Absolutely right on the trend — the capex data just last week supports his memo — and absolutely wrong on portfolio construction. High-volatility stocks with 4x leverage: the probability of getting wiped out once is just very high. It’s almost like it was inevitable.”
• Rory’s who-got-hurt mechanics: hedge fund LPs enter at different bases — early money that was up 440% round-trips to flat, but “if you bought in in April, May or June, you’ve been wiped” 80-90%, and Rory thought Jane Street might have put money in recently. Those late LPs “will read the docs real carefully,” and any undisclosed breach of mandate means liability: “Can you imagine going back to your investment committee and saying we did a hedge fund, but it appears it wasn’t hedged, and we lost all our money in a week?”
• Everyone agreed he survives — he’s 25, keeps his Anthropic position, and Larry Fink blew up early too. Nikesh: “I promise you he’s not going to be caught at the same place again.” Rory’s uncomfortable mirror for allocators: “When someone makes you a 10x, do you sit there and ask what can go wrong? No — you go, ooh, can I put in more money?”
• The SoftBank alum got the best line on leverage, deadpan: “Forex is for babies.”

4. Anthropic breaches three companies: “it’s time to pay your taxes”

• Nikesh read the breach exercise as a flex — “normally if you breach somebody’s infrastructure it’s not a good thing, but we’re all saying look how powerful these models are” — and says he told Anthropic and OpenAI to point the models at their own sandboxes first. The real shift is tempo: the average zero-day found in the wild takes 55 days to patch; models find vulnerabilities “in split seconds” and build attacks on the back of them. The capability he forecast for six months showed up in four; in 2-3 months he expects open source to have distilled it into fine-tunable attacker models.
• The commercial effect was immediate: “The flex Anthropic did with Mythos has every CEO talking about it. I spent eight years trying to get CEOs to talk about cybersecurity — couldn’t do it. Dario did it in one fell swoop.” And no one is ready: Palo Alto found 14,000 vulnerabilities in open-source packages in 14 weeks, misconfigurations are everywhere, and “the bad guy just has to be right once.”
• The KPI that changes: average detect-and-respond is 4 days and must become a minute. Palo Alto ingests 19 petabytes of enterprise data a day and runs detection at one minute — “the only problem is I only have 1,200 customers who’ve bought and deployed it.” His framing: “It’s not a fear problem. It’s a capability problem… it’s time to pay your taxes.” Harry’s translation: the security stack you had a year ago is “wholly unfit for purpose” for next year.
• Why frontier labs don’t kill him: they’re in “no perimeter security scenario” — someone still has to block the bad guy in line, on endpoints, servers and firewalls. “They need to be the ingredient in my product. They’re not going to take me out of business.”

5. Jason’s Fable horror story and the agency gap

• Jason’s confession from building Saster Connect: he switched on the Google Drive connector in Claude, and Claude found his private ideas doc (“Jason’s gems”), then Fable went and changed his code and his algorithm without telling him via MCP — no notice, no changelog; he found out only when the agent later flashed a conflict. “What if you have a thousand employees doing this?”
• Nikesh: “the wild west” — people experimenting with agents and connectors “without any regard for security,” no idea what data is training models or what credentials agents hold. His analogy: “When the Wright brothers built a plane, they didn’t invent TSA.” And the old adage bites harder now: if the product is free, you’re the product — “we are the product of all the post-training data collected by every model on our consumption,” which is exactly why enterprises pay to ring-fence.
• His agents distinction: most people run “glorified workflows,” not agents — “a Waymo has agency: it can drive you into the wall without human intervention.” Once code decides what happens next, you need identities, kill switches, and in-line interception. Nikesh’s follow-up: even at 0-5 rules per agent, “99% of the time it’s pretty darn good since the models upgraded in January” — but he concedes “it takes one destructive example to make it all unwind.” Nikesh: “Your VP of finance allowed to write checks? I might want to have a conversation.”

6. Moonshot at $35B and the two-tier intelligence market

• Moonshot closed $3.5B at a $35B valuation. Nikesh’s business-model question is whether open-weight models are already “a drag on price” for US closed frontier labs, but “it’s not clear to me it’s possible on a sustaining basis to offer open-weight models without monetizing in some way” — watch what emerges when Reflection and Thinking Machines ship theirs.
• Nikesh’s soundbite, whispered loudly on request: “In the long term, average intelligence is going to be free, and the average intelligence will get smarter. Exceptional intelligence will be paid for” — cancer cures, rockets, space data centers — “you don’t need to pay $6 per million tokens to answer a call saying how can I help you.” Harry’s near-term counter: customer support is consuming more tokens for sophisticated resolution right now — and those are precisely the workloads open-weight fine-tuning is coming for.
• Nikesh’s app-economics lens: token cost as a share of revenue runs 10-15% at Salesforce/Intercom-type apps versus 70-80% in coding — the latter is “just raw intelligence and a mild harness.” Very different businesses.
• Nikesh’s caution on agentic timelines, via his 2009 ride in Google’s camera-covered Lexus: it took 14 more years of edge-case grinding (“that’s a tree, idiot”) before anyone trusted the agency. “We’re not going to give 100% agency to use cases for some time” — and meanwhile, “over the next 3-4 years we won’t be paying for intelligence, we’ll be paying for compute through our nose.”

7. Compute is the trade: land, permits, energy — and dislocation is a buying opportunity

• Valor Atomics tripled to $6B with Sequoia leading and an Nvidia partnership; Nikesh’s chicken-manure anecdote carries the point — a guy turning chicken feces into methane “has billions of dollars and is selling the energy to hyperscalers.” “Land, permits, energy — this is the thing that is going to get priced for the next 3 to 5 years.” The Anthropic-vs-OpenAI question becomes who has access to more compute: “You can take all the free Chinese models you want — where are you going to run them?” Jason’s confirmation from his energy-storage past: these businesses used to pencil at 8% IRRs nobody wanted; “maybe it’s 80 today.”
• Rory grounds the nuclear excitement: criticality demos aren’t the hard part — the regulatory journey is the whole ballgame. NuScale is furthest along with well-understood light-water tech; Valor and Oklo (likely) are doing new designs. “From a public policy perspective, go team — but don’t spend the electricity yet.”
• Harry’s systemic worry — have we ever had an ecosystem this dependent on OpenAI and Anthropic hitting 2027 numbers? Rory: orthogonal. “There is infinite demand for AI at this moment… whether it’s OpenAI that builds the data centers or somebody else pays for them, there is demand.” His striking stat: ~70% of AI compute demand is consumers getting a free ride — reallocation to enterprise or consumer monetization (agents booking flights and restaurants) closes the gap.
• Rory’s pushback — worth keeping: markets currently assume 70-80% of that demand channels through OpenAI/Anthropic, who buy the chips and resell frontier intelligence; if instead Moonshot becomes the model of choice on the same compute at “10 cents on the dollar tokens,” then “Moonshot is happy, Nvidia is happy, enterprise is happy, OpenAI very very sad.” The answer: a change of players at the table isn’t demand destruction — “that’s called a buying opportunity.” Bull markets price out execution (“every chicken manure company that says the words in PowerPoint gets funded”), but execution decides winners — see OpenAI-first/Anthropic-second flipping in two years, and Google going from written off to compute plus cloud sales plus Gemini.

8. Context is the moat: the commoditization war between models and models-plus-context

• Nikesh’s core strategic bet: the next 3-5 years are about building organizational context, not model choice. No model knows why his customer’s firewall is down — what product, OS, configuration, or the last five incidents; that lives in vector DBs and in-context learning systems. “I have more people collecting context than I’ve ever had” — the Waymo move again: “I’ve got people planted saying, this is a tree; this is why it goes down.” Once that exists, “I can stick any model I want on it and the model distinction will not matter.”
• He splits the stack in three: raw model intelligence, context to answer queries, and context to train the ecosystem — every Palo Alto customer case ever, transcribed, so the system knows a good answer from a bad one. Thomas [likely Kurian]’s early advice (“don’t build a cyber model — the big models keep getting smarter”) was right, but once average intelligence is smart, “domain becomes equally important with the model intelligence.”
• The battle he foresees: the Microsoft architectural pitch is context in a harness beside the model — swap models freely, commoditize them — while “every model company is saying no, I’m going to make my model smarter with context, because otherwise I get commoditized.”
• Harry’s hard question: Palo Alto has a team to manage the fact that Opus 5, Fable and Unlet don’t perform the same — what does the average enterprise do? Nikesh, deadpan: “Bending Spoons. Bending Spoons. A Darwinian moment does not suggest that everybody survives.” Harry’s gloss: “If we don’t figure this out, we’ll be working for the Italians.”

9. Q2 earnings: inference sellers crushed it, Palantir shames everyone, timing is the only risk

• Rory’s setup: everyone with a cloud-inference business had an amazing quarter — Google Cloud grew 82% (the smallest), AWS 37% at scale, Microsoft 20-30% (bundled); collectively ~$400B of run-rate adding roughly $100B a year of new compute revenue. “People sold a ton of inference, and because of that said: I can turn compute into money” — Amazon and Microsoft marked up, while Meta pledged big spend without an obvious payoff and sold off. “Will it persist? Who the hell knows. But the new information in Q2 was bullish.”
• Palantir was Jason’s standout: growing almost 100% with bookings up 153% off just 1,049 customers who “will pay almost anything to get these questions answered with AI” — and they came back from 15% growth four years ago. “Send it to our portfolio companies and tell them there’s no excuses. Work harder, kids.” Nikesh on why it works without a learning system: AI trawling petabytes for 20 insights of which five are amazing — “300 basis points on your top line, 100 on margin — hallelujah, you just paid for everything.”
• Nikesh’s macro frame: ~$1 trillion of capex committed for the next year, and for now the market is funding it — “we have one more run at the roulette table.” His conviction: “every consumer app ever put on the iPhone has to be redone; every enterprise app in SaaS has to come back with an opinion” — so it’s not a demand, jobs, or appetite problem, it’s a timing problem: telecom ran 3G/4G/5G capex cycles waiting years for rewards, but this cycle is compressed because “the numbers are way too big to be funded by speculators for long.” The next dislocation may be compute supply — regulators, Europe, “30 states with picket fences saying no data centers in my state” — with knock-on pain for semis.
• The differentiator both landed on: digestion speed. Rory, channeling Gavin Baker and Darwin: “It’s not the strongest of the species that survives, nor the most intelligent, but the one quickest to learn.” Nikesh’s homework for every enterprise: build your own training data — his 400,000 annual support cases are each “a learning opportunity,” and it’s a 3-5 year grind per use case. The trap: “How do I get from 70% accuracy to 99%? The problem is I don’t know which 30% is inaccurate. So everything’s useless.”

10. Deal desk: DroneDeploy’s calm exit, Scale AI’s resurrection, and Nikesh’s M&A doctrine

• DroneDeploy sold to Procore for $900M after 13 years — and Jason framed the buyer’s side as bet-the-farm: Procore, beaten down and trading around 4x, paying 12x revenue with a lot of debt. Rory, on the board, called it “genuinely one of the least stressful deals I’ve ever done” — the lessons: capital discipline (“we always raised below the price we sold at,” profitable, modest rounds) and being on the right side of the platform shift — “when you’re software enabling drones and robots, you’re on the side of the future.” His timing confession: “I thought drones would have exploded 5 years ago… in the physical world, AI takes a lot longer to happen.”
• Nikesh’s acquisition doctrine: 40+ companies in 8 years, ~75% worked — a hit-rate requirement “more than a VC.” The career-defining one: a $28B deal (CyberArk) at 14-16% of market cap, probably now worth north of $50B. “If you don’t make the big ones work, you lose the license to run your business.” The thesis fits in a sentence — agents will need identities treated as privileged identity — which satisfies Rory’s partner’s rule: “If you can’t express it in a sentence, it’s probably a bad deal. If you can, it’s probably a good one.”
• Harry’s public mea culpa on Scale AI: “I called it a husk a year ago. I was wrong.” New CEO; they hit $1.5B in ARR — proof that “when you’re in a great market with a product that meets the need, even losing your top people is fine.” Windsurf-to-Cognition is working too, and Groq’s hosted inference could be next: “In the face of insatiable demand, all things are possible.” The counterpoint reminders in the same news run: Mailchimp revenue down eight straight quarters, Visa cutting 2,600 jobs.

Verification Notes

• The raw captions do not identify the speakers for the two passages labeled “Eric Schmidt”; those transcript labels were changed to “[Speaker?]”.