Markets Shrug at 10% Extinction Odds With a 0.1% Dip

Source: 20VC with Harry Stebbings | Published: 2026-09-17T13:55:57Z

Jason Calacanis argues Dario's P-Doom open letter is IPO roadshow risk-disclosure prep for Anthropic, while markets absorbed the existential warning with a barely-there 0.1% slide in semiconductor stocks.

Dario's Open Letter Might Be a Living S-1 Risk Disclosure

Jason Calacanis offered the sharpest read of the episode: this letter isn't a policy discussion — it's the warm-up act for Anthropic's IPO roadshow.

His logic: a former Anthropic employee who worked there four to eight weeks posted on Twitter that "many of my colleagues and I believe there's a 10% chance AI destroys humanity within a decade." Put that sentence in an S-1 risk factors section and institutional investors get very uncomfortable. So Dario is surfacing it now — letting society hash it out before the roadshow, so by the time he's in New York pitching investors, it's already old news.

"This isn't even my cynical read. I think this is exactly what his job as CEO requires him to do."

It's not an unreasonable interpretation. Dario carefully separates three categories of risk in the letter: cyberattacks, economic displacement, and loss of control over AI. He doesn't endorse the 10% P-Doom figure — he merely "responds" to it. The language is precise. Legal probably reviewed every word.

If There's Really a 10% Chance of Human Extinction, the Feds Would Have Acted Already

Rory O'Driscoll's rebuttal to the P-Doom argument is simple and devastating:

Substitute "AI" with a nuclear reactor in downtown San Francisco — one that's safe for now, but carries a 10% chance of destroying the entire world within five years — and publish that statement. How many seconds before the U.S. military shuts your company down?

Answer: not many.

So there are only two conclusions: either the federal government is asleep at the wheel, or federal officials look at the P-Doom crowd and privately judge them as "overexcited teenagers — we'll step in if something actually goes wrong." Markets confirmed this. After the Anthropic news broke, semiconductor stocks dipped slightly while cybersecurity stocks rallied 10%. Markets processed a 10% probability of human extinction with a 0.1% price move.

Three Risks, Only One Worth Taking Seriously

The episode dissected each of Dario's three risk categories in turn.

Cyberattack risks are real, but the cat is out of the bag — China and Russia already have these capabilities, and no regulatory framework can claw that back. The economic displacement argument was dismissed as crowd-pleasing: from the agricultural revolution to the industrial revolution, no technological advance has ever been halted because it would "eliminate jobs."

The third risk is where it gets genuinely interesting: we might lose control of AI — recursive self-improvement, systems evolving into something nobody can predict or manage.

Rory's take: if you admit you're building something you can't control, you either believe the problem is ultimately solvable (so keep building) or you believe it's unsolvable (in which case you have a responsibility to stop). David Sacks was more direct: if you genuinely believe there's any probability of human extinction, that's your responsibility as CEO — not the regulator's. Shut the company down, or bring in someone who can actually solve the problem and step aside.

Meta Muse's Real Advantage Has Nothing to Do With Features

The night after GPT-6 Astra launched, Zuckerberg pulled together a massive AI team and gave them one mandate: build a consumer AI assistant. Not 20 people — 500 people, running hard, engineers responding to user bug reports on Twitter in real time, online Saturday night and Sunday morning.

After trying it, Jason's verdict: this is the definition of "great software" — you can't see all the invisible work it does, but everything just runs smoothly. He had Muse draft pitch emails for his show team and rebuild an entire WordPress backend from scratch. Muse delivered.

But the real moat isn't there. At the feature level, Muse gives each user a virtual machine — roughly 2 CPU, 2 GPU, 8GB RAM, 100GB storage — at a cost of about $3–4 per user per session for comparable products. Meta's advantage is that it already owns the infrastructure at scale, running on its own Muse LLM, fast and cheap. Nobody can compete on that dimension.

One VC Argued Both Sides of the Same Deal in the Same Room

Instinct is raising $1 billion at a $10 billion valuation. Jason was asked to simulate an IC (investment committee) discussion live: would you invest?

He spent three minutes making the bull case: Instinct's founding team is among the strongest he's encountered; trust and data privacy are Meta's Achilles heel; and while Muse covers email, calendar, payments, and cross-app workflows, it requires users to hand over their most intimate personal data to a company with a documented history of misuse — Instinct carries none of that baggage as an independent platform. Founder Noah Shinn is described by everyone who's met him as a "generational talent," exactly the profile that acquirers like OpenAI and Google would pay a premium for. Even without a clear path to profitability, the exit scenarios look attractive.

Then he pivoted and spent three minutes making the bear case: Meta has unlimited compute, its own LLM, cost advantages, and speed advantages. Instinct's sluggish response times are already a warning signal. If you can't count on M&A as an exit, the IPO path isn't obvious. Venture capital's cardinal rule is don't bet on deals that only work if someone acquires you — that's too unpredictable. You might be Clem's best friend, he introduces you to Jensen, and Jensen resigns the next day.

His conclusion: buying in at a $50M valuation in April was a phenomenal risk/reward play. Buying at $500M in May or June was an interesting bet. At the current $2.5B Series B — and investing at the proposed $10B new-round valuation — the exit size required to justify the multiple keeps climbing, and the largest AI acquisition in history was Cursor at $60 billion. Without serious revenue, you're not getting a $40 billion takeout.

Musical Chairs With Two Seats Left

Miro was acquired by Bending Spoons for $1.35 billion. In 2021, it was valued at $17.5 billion.

Jason's analogy: musical chairs — one seat removed each round, only a few spots left. Miro got one. When you sell at 2.4x or 2.5x your last round, it almost certainly means only one buyer made an offer. Any capable acquirer — Salesforce, Francisco Partners, Thoma Bravo — could have easily bid a touch higher. One offer means the market has lost interest in this asset class.

Bending Spoons' playbook is simple: acquire, raise prices dramatically, absorb 30–40% customer churn, and whoever stays is someone who actually needs the product. The first to leave are users who were sold on the product by a sales team. The ones who stay can't live without it. This is a fundamentally different business philosophy — VC syndicates can't execute this, because aggressive layoffs and steep price hikes aren't in their DNA. These assets may genuinely be worth more in Bending Spoons' hands than they ever were under the original investors.

The problem: the chairs are running out. Rory's read: if you're worse than Miro, you don't get any of the remaining seats. Those companies won't be acquired — they'll enter zero-growth mode and slowly fade out.

Mistral Raised €3 Billion, But This Isn't Competition — It's Sovereignty

Mistral closed Europe's largest-ever tech funding round: €3 billion. Samsung led. ASML participated in the prior round.

Rory's read is unambiguous: don't interpret this as Mistral mounting a serious challenge to OpenAI or Anthropic on frontier models. That's nonsense. This is about AI sovereignty.

Europe made a calculation: this technology is too important to leave entirely in American hands. Same as the decision 30 years ago to build Airbus — not because Europe thought it could beat Boeing on pure market competition, but because "we cannot be a vassal state on strategically critical technology." It took 10–15 years to build a genuine competitor. Airbus eventually surpassed Boeing.

One catalyst: the U.S. government recently demanded Anthropic cut off Fable model access to all other countries — not just Russia. Everyone. The day that happened, every European policymaker understood they could no longer depend on American AI companies. If you're a Mistral shareholder, you should be writing a thank-you note to the current geopolitical climate. It just made you several billion euros richer.

Canva and Adobe's Real Problem Is Transitioning From Growth to Value

Adobe appointed a new CEO — an internal promotion, no surprises. Jason's verdict: nothing will change. Milk the existing product lines' high margins, layer in some AI features, no fundamental shift.

The more interesting story is Canva. Australian disclosure requirements have revealed that Canva's growth has slipped from 30% to 20%. That number still sounds decent, but it triggers a brutal valuation mechanism.

Rory laid out the logic: above 30% growth, every mistake is forgivable and you get revenue multiples. Below 30%, you have to switch to EBITDA multiples. The pain of that transition: you go from 6–7x revenue to 20x cash flow. To hold your valuation flat, you need to first get your cash flow margin to roughly 30% — and that process takes three or four years.

Box went through this. Canva is standing at that threshold now. The good news: strong founders and genuine user value. The bad news: navigating this transition as a private company is brutal. Investors rotate out, valuation logic shifts, the whole process feels like running in place — and private market patience is far shorter than public market patience.