Video · 2026-08-06 · 1h 18m · 6 moments

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

✦ AI generated

GUEST Nikesh Arora GUEST Rory O'Driscoll

timeline · colored by role

Claim

No one outbid Bending Spoons for Airtable because PE firms see SAS as facing a fundamental change in long-term growth expectations, not a pricing dislocation—and their own portfolios already hold enough turnaround problems.

The panel explains that no PE rival outbid Bending Spoons for Airtable because SAS multiples reflect a fundamental long-term-growth reassessment, and PE firms already carry heavy software-restructuring portfolios.

transcript

Nikesh Arora: You know you see dislo is this a pricing dislocation or is the fundamental change in the long-term growth rate that people expect out of SAS. If it's a fundamental change in long-term growth rate people assess, then the multiples are right now, you know, and that's I think where the market is grappling with this... the people you mentioned, uh, Jason, the the PE guys, they might have a full roster of stuff they'd like to sell to Bending Spoons as opposed they'd like to buy against Bending Spoons.

provides context · 1

Claim◆

Leo Aschenbrenner was absolutely right on the AI trend—the capex data supports his memo—but absolutely wrong on portfolio construction; accumulating high-volatility stocks with forex leverage made his wipeout mathematically inevitable.

Nikesh Arora distinguishes between Aschenbrenner's correct macro thesis on AI capex and his flawed portfolio construction—high-volatility assets stacked with forex leverage that made a wipeout statistically inevitable.

transcript

Nikesh Arora: I mean, I think he was absolutely right on the trend. I mean, absolutely right on the trend and remains to date right on the trend. In other words, the data just last week about capex absolutely supports his memo. So conceptually right on the trend and then absolutely wrong on portfolio construction. If you accumulate a portfolio of high volatility stocks with forex leverage, the math makes it clear your probability getting wiped out once is just very high. This is as simple as that. Absolutely right on trend, absolutely wrong on portfolio construction. It's almost like it was inevitable.

Claim

AI-driven security defends at a fundamentally new speed: adversaries find vulnerabilities in seconds where it once took months, while organizations' time-to-detect remains days—so the security infrastructure of a year ago is wholly unfit for the year ahead.

Nikesh Arora argues AI collapses vulnerability-discovery time from months to seconds while average patch time is 55 days and detection/response is 4 days, making existing security infrastructure unfit for purpose.

transcript

Nikesh Arora: These things are finding vulnerabilities in split seconds and then turning around and building an attack on the back of that. So I think the the fundamental speed at which cyber attacks will happen and need to be defended changes... That is fundamentally not true. Now, we we found 14,000 vulnerabilities in open source in the last 14 weeks testing open source packets... The average time to detect and respond is 4 days. How are you going to get it down to a minute? So, it's not a fear problem. It's a capability problem. It's an infrastructure readiness problem.

Mechanism

The critical variable in AI security going forward is context—not the raw model—because organizations that build proprietary training context can stick any model on top of it, making model distinctions and even frontier-model dependence less relevant.

Nikesh Arora argues that the competitive moat in AI shifts from model intelligence to proprietary context and training data, enabling model-agnostic infrastructure, echoing but differing from Satya Nadella's architectural point.

transcript

Nikesh Arora: the part which we will be build, we are starting to build and we will be building for the next 3 to 5 years is context... All that knowledge, all that learning is being captured by me in effectively vector DBs and in context learning systems... then I can stick any model I want on it. And the model distinction will not matter because the context will become as important or perhaps more important.

Prediction

Compute supply—not demand—is the next dislocation risk: capex can be committed, but regulatory, permitting, and geographic constraints on data centers could throttle supply and knock on to the semiconductor cycle.

Nikesh Arora identifies supply-side constraints—regulatory approvals, permitting, states restricting data centers, European barriers—as the likely next dislocation, since demand is effectively infinite.

transcript

Nikesh Arora: I think the next dislocation could happen is in the supply of compute. You can bring all the capex to bear but I think to your point the valors may not get their their regulatory uh set of approvals. Europe may not allow data centers. You may find 30 states with you know picket fences which say no data centers in my state. So there's a supply problem that happens in comput side which could have a knock-on impact on all our infrastructure buddies in the semiconductor space saying holy doesn't look like all the stuff they're building is going to go out as fast as we thought it was going to go out.

Example◆

Demo Deploy succeeded because of capital discipline, profitability, and being on the right side of a platform shift—the lesson being that in the physical world AI takes far longer to arrive than expected, but when it arrives it is transformative.

Rory O'Driscoll attributes Demo Deploy's low-stress sale to disciplined fundraising, profitability, and riding a platform shift toward drones and robots—noting physical-world AI takes longer than expected to mature.

transcript

Rory O'Driscoll: when you have capital discipline and you know, modest fundraisers, you you're set up for success, not failure... The trend was our friend, not our enemy... in the physical world, AI takes a lot longer to happen. I mean, when it happens, it's amazing, but it's clearly, you know, I look back 10 years ago, I thought drones would have exploded 5 years ago. They're really starting to explode now, as are robots.

Highlight slides

Right on trend, wrong on construction✦ from: Leo Aschenbrenner was absolutely right on the AI trend—the capex data supports his memo—but absolutely wrong on portfolio construction; accumulating high-volatility stocks with forex leverage made his wipeout mathematically inevitable.Capex thesis vs. portfolio construction✦ from: Leo Aschenbrenner was absolutely right on the AI trend—the capex data supports his memo—but absolutely wrong on portfolio construction; accumulating high-volatility stocks with forex leverage made his wipeout mathematically inevitable.A Low-Stress Sale Built on Discipline and Platform Timing✦ from: Demo Deploy succeeded because of capital discipline, profitability, and being on the right side of a platform shift—the lesson being that in the physical world AI takes far longer to arrive than expected, but when it arrives it is transformative.Physical-World AI: Slow to Mature, Transformative On Arrival✦ from: Demo Deploy succeeded because of capital discipline, profitability, and being on the right side of a platform shift—the lesson being that in the physical world AI takes far longer to arrive than expected, but when it arrives it is transformative.

Related episodes

Video · a16z Podcast

Ben Horowitz on the Global Race for Tech, Power, and Influence

→

Audio · All-In Podcast

SpaceX-Cursor Deal, SaaS Debt Bomb, New Apple CEO, SPLC Indictment, Colon Cancer Spike

→

Video · Lenny's Podcast

The rise of taste, human authenticity and judgment in an AI world | Adam Mosseri (Head of IG)

→