
Anthropic's Record IPO Fades From Its Open as the AI Selloff Spreads to Asia
Transcript
3,283 words · Plain textWelcome to Super Intelligence News and Report. Today is Friday, November 20th, here are the headlines.
Maya Today the market is pricing what AI is worth, while lawmakers, critics and even insiders ask what it costs. Start with the market. Anthropic's debut didn't stop the AI selloff. On Thursday the Nasdaq closed at a seven-month low and Nvidia hit a new fifty-two-week low. On Friday it spread to Asia. SoftBank Group closed down almost fifteen percent, its worst day since August 2024, and Korea's Kospi and TSMC both took heavy losses. CNBC tied SoftBank's drop to its big OpenAI bets. Why it matters: investors are separating AI companies from the businesses that funded them, and anyone holding OpenAI exposure is getting hit.
Theo Rough week to be the one holding the OpenAI stake.
Maya Very rough. Next, chips and China. On Thursday, House Foreign Affairs Chairman Brian Mast released licensing data showing Commerce approved about eight hundred eighty thousand Nvidia and AMD AI chips for Chinese buyers in six months, December through May. Alibaba, Tencent, Baidu and BYD are on the list. Nvidia says it has since stopped shipping H200s to China. Separately, The Information reports that Moonshot AI cancelled a Blackwell chip rental in Indonesia after the new US rule on remote chip access. That part is reported, not confirmed. Why it matters: licenses granted and chips actually reaching China are now two very different numbers.
Theo And if the Moonshot report holds, the remote-access rule is already biting.
Maya Staying with governments. On Thursday the European Commission confirmed what its spokesperson Thomas Regnier called an AI-generated cyber campaign, first detected in late October. According to The Register, staff payroll, leave and medical reimbursement systems were down for more than ten days. The Register links the attack to ARTEX, a publicly available security-testing tool. That link is reported, not confirmed; the Commission hasn't named the tool, or said whether personal data was taken. Why it matters: a major government body is confirming an attack carried out with an off-the-shelf AI tool.
Theo Here's a friendlier version of software acting on its own: the desktop-agent race just reached the Mac. On Thursday Google added computer use to the Gemini Mac app, so Gemini can organize files, fill in forms and work across Mac apps in the background. It's English only, for US subscribers on AI Plus, Pro and Ultra. Same day, Bloomberg's Mark Gurman reported that Apple is building its own agent for device control, apps, browsing and coding, based on new job listings. That's early-stage and reported, not confirmed, and Apple hasn't commented. Why it matters: the company that owns the operating system may end up competing directly with the agents running on top of it.
Maya On the deals side, The Information reported Thursday that Nvidia is in advanced talks to buy a startup called Inception for more than half a billion dollars, plus earnouts. That's reported, not confirmed, and neither company has commented. Inception makes Mercury, which is the interesting part.
Theo It's a diffusion language model. Instead of writing one token at a time, it refines lots of tokens in parallel, and the whole point is faster answers.
Maya And the co-founders, Stefano Ermon, Aditya Grover and Volodymyr Kuleshov, would reportedly join Nvidia, with the tech feeding its Nemotron models. Why it matters: even without Huang, Nvidia is still buying ways to make AI cheaper to run. Now to Washington. On Thursday four senators, Mark Warner, John Kennedy, Angela Alsobrooks and Mike Rounds, two from each party, introduced the AUDIT AI Act. This comes via Politico; it's reported, not confirmed, and we haven't seen the bill text. Public companies outside finance would have to disclose, in their 10-Ks, 10-Qs and offering documents, the special-purpose vehicles, joint ventures, leases and guarantees behind their AI infrastructure. The SEC would get about six months to write the rules. Politico says Warner is worried about circular deals, where companies finance their own customers. Why it matters: the off-balance-sheet debt behind AI data centers could become public.
Theo Bipartisan, too. That's not nothing.
Maya It isn't. And some pushback from inside Google. On Thursday more than five hundred forty Google employees and contractors signed a petition from the Alphabet Workers Union called Let Students Search. They want K-12 schools to be able to use Search without the AI features. Right now a district that turns off AI has to give up Google Search entirely, and the New York Times reports districts in New York City, Los Angeles and Chicago are looking at alternatives. Google says separate on-off switches are coming early next year, but Search would still include AI. Why it matters: it shows how hard it's become to get Google Search without the AI.
Theo So even with the switch off, some of the AI stays on.
Maya That's what Google is saying.
Theo Meanwhile Google is also looking up. On Thursday Blue Origin announced that Google picked its New Glenn rocket for the first dedicated heavy-lift launch of Project Suncatcher, Google's effort to run AI computing on solar-powered satellites. The target is early 2029, and each satellite would carry roughly a data-center rack's worth of compute. Google isn't calling it an operational orbital data center; this mission is about proving it can work. Why it matters: Google is putting a date on AI compute in space. But Blue Origin's launch pad was badly damaged in May, so that date could slip.
Maya A rack in orbit by 2029, while the Senate is still asking who's paying for the racks on the ground.
Theo Fair. Last one, into the ideas pile. People on X are passing around an essay Shane Legg shared, called What comes after AGI, by Google DeepMind researcher Tim Genewein, published Thursday by the DeepMind Institute. It lays out four routes to superintelligence: scaling, new paradigms, coordinating many agents, and recursive self-improvement. It estimates effective compute is growing at least tenfold a year, and argues that capping one input, like training compute, won't stop progress and could cause sudden jumps later. The site notes it isn't Google's official view. Why it matters: it's an argument from inside Google DeepMind against relying on compute caps alone.
Maya Interesting timing, given the chip-licensing fight. Next we go deep on Anthropic's first day as a public company: a record debut that faded from its open.
Now for the Main Episode
Maya Here's the question to hold onto. When investors paid up for Anthropic on its first day, did Wall Street endorse a trillion-dollar valuation and a safety-first pitch? Or is the slide from the open, in a market that kept falling around it, the more honest signal?
Theo I came into this week not sure which. So start at the beginning. How did they get to the first trade?
Maya About a week before the listing, Anthropic set its range at seventy to seventy-eight dollars a share. On Wednesday it priced above the top of that, at eighty-three. It sold nine hundred eighty million shares and raised about eighty-one billion dollars.
Theo And then there's a second pile of money on top.
Maya A separate twenty-six billion dollar private placement with suppliers. Together, about a hundred and seven billion. The National calls that the largest raise on record, ahead of SpaceX's eighty-six billion in June. So roughly a quarter bigger, five months later.
Theo Hang on. If a big chunk of that total came from suppliers, is it really a record for public demand?
Maya That's the right thing to flag. The IPO piece on its own, about eighty-one billion, is smaller than SpaceX's listing. The record only happens because companies that sell to Anthropic put in twenty-six billion. Not a scandal, but a different kind of money than a pension fund buying shares.
Theo And the kind of money the senators in our headlines want to see in the filings.
Maya We'll come back to that. So Thursday, the stock opens just before noon Eastern. First print, a hundred eleven fifty. That's thirty-four percent above the offer, and for a moment it implied a fully diluted value of about two trillion dollars.
Theo Two trillion. For the people who make Claude.
Maya Briefly. It traded heavily, about eight hundred forty million shares, which one market site puts at more than half again what SpaceX did on its first day. Then it ground lower all afternoon and closed at ninety-seven seventeen.
Theo So up on the offer, down on the open.
Maya Up seventeen percent on the offer, about thirteen percent below the open. Then after hours it slipped again, to somewhere around ninety-one or ninety-two dollars, another five or six percent. The figures differ a little by source.
Theo So anyone who bought the opening print was down about a sixth by evening. And the valuation? I've seen one and a half trillion, two trillion, everything in between.
Maya Depends what you count. At the close, on shares actually outstanding, about one point two seven trillion. Count every option and dilutive share and The National gets about one point seven five. The splashier headlines mostly use the offer price or the intraday high. If you want one number to carry around, it's roughly one point three trillion.
Theo Still a trillion-dollar company on day one.
Maya Still. But here's the detail that matters most, and it's our own math from the reported figures. About thirteen billion shares outstanding, nine hundred eighty million sold. So only about seven percent of the company is actually trading. And PitchBook's Harrison Rolfes says orders ran about six times the shares available.
Theo Six buyers for every seat, and not many seats.
Maya And heavy demand chasing a small float can lift a price by itself. So what we can't untangle yet is whether that pop was demand for Anthropic, or scarcity of Anthropic.
Theo Let me add what I think buyers believed they were getting, because it comes from my side of the house. This priced right as OpenAI's GPT-six point two was having a rocky launch. Gizmodo argued the day before the debut that Anthropic might be the real winner of OpenAI's release. Early testers said the new model needed too much pushing, or stopped long tasks early. And the developer Theo Browne, no relation, said he'd still pick Opus five point five for his own work.
Maya So the idea that Anthropic is pulling ahead was at full volume the week it went public.
Theo Full volume, but thin. That's a few days of early impressions, not measurements. Nobody has settled which model is better for long agentic work. So some of that premium rests on vibes from a handful of developers. I build with these tools, and I wouldn't put a trillion dollars on vibes.
Maya And it shows up on the other side of the ledger. On Friday SoftBank closed down almost fifteen percent, after already losing about thirty percent in October, and CNBC tied the drop to its heavy OpenAI bets. One reading is that markets are treating Anthropic's gain partly as OpenAI's loss.
Theo How confident can we be in that link?
Maya Not very. The whole AI trade was falling on Friday, the Kospi and TSMC included. How much of SoftBank's day is about OpenAI specifically, nobody can say yet.
Theo Okay, that's the money. What did Dario Amodei actually say on the day?
Maya That being public would strengthen Anthropic's commitment to safety, not weaken it. That it won't bend to business or political pressure. And that it'll pace training and deployment until it's confident its systems are safe and under human control. He called the strategy a race to the top again, where rivals compete on capability without cutting safety. The Grok X report counts about one and a half million views on Anthropic's main post. So this wasn't a footnote. It was the message on the biggest day in the company's history.
Theo And I'll give him some credit here. That pacing promise is testable from where I sit. Release cadence is public. If a rival ships something big and Anthropic holds a model back, we'll see it. If it rushes one out the next week, we'll see that too. It's a claim about behavior, not a values statement.
Maya Testable, sure. But think about who it's promising now. Last week Amodei answered to investors who'd signed up for the mission. Now it's anyone with a brokerage account, and they get a quarterly report card. The first time revenue misses, nobody's asking about human control. They're asking about the next release.
Theo Though his argument is the opposite. A giant pile of public money gives you room to say no.
Maya It's an argument. It's just never been tested at this scale. And the sharpest criticism came from someone who knows the company from the inside. Dustin Moskovitz, the Facebook co-founder, one of Anthropic's earliest investors and a former board observer, told The New York Times that he and his wife, Cari Tuna, invested to promote safety, not racing. He said AI is moving too fast, and that Anthropic and everyone else at the frontier should slow down and stop racing.
Theo On the day of the listing.
Maya On the day of the listing. And he isn't cashing out and complaining. In early 2025 they moved their stake, worth billions, into a nonprofit that hasn't been named, so they wouldn't personally profit from selling it. The Times notes he's the biggest funder of AI-safety nonprofits.
Theo That's hard to wave away. He's basically saying a race to the top is still a race.
Maya That's how I hear it. The name of the strategy concedes the premise.
Theo I'd push back a little. Moskovitz wants everyone at the frontier to slow down together. Amodei's bet is that nobody else will, so you'd better be in front, setting the standard. Those are different theories of how you get safety. The IPO doesn't settle which one is right. It just raises the price of being wrong.
Maya That I'll grant you. And the street made Moskovitz's point less politely. More than a hundred protesters outside Anthropic's San Francisco headquarters, calling for a ban on AI and on recursive self-improvement, systems that improve their own successors. Mission Local reports police arrested four people, including Stop AI co-founder Guido Reichstadter.
Theo That's one of the four routes in the DeepMind essay from our headlines. So protesters and a DeepMind researcher are pointing at the same mechanism, from very different places.
Maya So let's weigh it. The bull case isn't weak.
Theo Go on. I'll probably agree with more of it than you do.
Maya Probably. Six times oversubscribed, per Rolfes. And the stock held a seventeen percent gain on a day the Nasdaq fell more than two percent to a seven-month low, Nvidia hit a new fifty-two-week low, and Yahoo Finance's live coverage had the Dow down about eleven hundred points. It also closed above both outside estimates. PitchBook had pegged it at seventy-nine dollars a share, and Morningstar's fair value is ninety-two.
Theo And Dan Niles was cheering, right?
Maya Per the Grok X report, which we haven't independently confirmed, the investor argued on X that the debut shows strong appetite for a pure-play AI company even in weak markets.
Theo I think that pure-play point is the real insight. Until this week, if you wanted frontier models and nothing else, you couldn't buy it. You bought Nvidia, or SoftBank, or a cloud company, and got a pile of other stuff attached. Now there's one ticker. Some of that demand is just people finally getting the thing they wanted.
Maya Which cuts both ways. A first-of-its-kind asset with seven percent of its shares trading is exactly where scarcity does the most work. Rolfes himself says the first year will test the price. PitchBook expects twenty twenty-seven to be the weakest year in its forecast, as locked-up shares are released in stages around Anthropic's first earnings reports.
Theo So the supply arrives right when the first real numbers do.
Maya At the same moment. And we don't know the date of that first report, or how fast insiders can sell. The last record listing is a caution too. One Nasdaq dot com commentary notes SpaceX has mostly traded sideways since June. One data point, not a rule.
Theo And was Anthropic supposed to rescue the sector?
Maya Some people hoped so. Vijay Valecha, chief investment officer at Century Financial, said before trading that the listing was being watched as a verdict on the wider AI trade. And a Taiwan-based trader quoted in coverage said many expected AI stocks to rebound if Anthropic went well. It went well. They kept falling.
Theo So if it was a verdict, it was: Anthropic, yes. Everyone else, no.
Maya Or the sector's problems are just bigger than one listing. Nvidia has slid for weeks since Jensen Huang resigned and the China export questions piled up. And the Fed's November Financial Stability Report named AI the top financial-stability risk. One good IPO doesn't change what the Fed is worried about.
Theo Which brings us back to the senators.
Maya The AUDIT AI Act, from Warner, Kennedy, Alsobrooks and Rounds, introduced the same day Anthropic started trading. It's via Politico, reported, not confirmed, and we haven't seen the bill text. Public companies would have to disclose the special-purpose vehicles, joint ventures, leases and guarantees behind their AI infrastructure. Warner's worry, per Politico, is circular deals, where companies finance their own customers.
Theo And twenty-six billion dollars from suppliers is at least worth a look under that lens.
Maya Worth a look, and I want to be careful. We don't know whether the bill would cover Anthropic's supplier placement or its compute deals. Nobody does yet. But if it does, how this company's compute gets paid for becomes a lot more visible, and that's exactly what investors will want to see before the lockups end.
Theo From the builder side, that's the piece I care about. Developers picking Claude for real work are betting the API keeps getting better without getting pricier. That depends on those compute deals. If they turn out to be fragile, it's not just shareholders who feel it.
Maya So where do you land? Endorsement or warning?
Theo Closer to endorsement than you, I think. Not of a trillion-dollar valuation, but of the company. Holding a seventeen percent gain while the Nasdaq hits a seven-month low tells me people want this specific business, and the developer preference is real, even if it's early. Where I agree with you is that day one doesn't price the thing that matters most to me, which is whether the pacing promise survives a rival's big launch.
Maya I'd put it more bluntly. Day one endorsed demand for a scarce asset. Seven percent of the company, six times oversubscribed, a first-of-its-kind ticker. That tells you very little about whether one point three trillion is right. On the valuation, the fade from the open, the after-hours drop and a sector that kept falling are the more honest signals.
Theo And on the safety pitch?
Maya Untested. Amodei made the promise loudly, Moskovitz said it isn't enough, and the market didn't really price either view. That test comes with the first missed quarter, or the first time a rival ships something that makes Claude look behind.
Theo So the near-term tell is today's session, the first full one, coming off that after-hours slide with Asia selling off around it. We'll have how that played out, and how far the Asian selloff carried, on tomorrow's show. The longer tell is the calendar: the first earnings date, and how the lockups get staged around it.
Maya That's when the float stops being seven percent and starts being the real market. Wall Street got to say what Anthropic is worth on a day when almost nobody could sell. The real verdict comes when everyone can.
Thanks for listening to Super Intelligence News and Report. Please subscribe and share it with your friends.
Sources
37 pages this episode's research used- Everything That Happened in AI Today (Thursday, November 19, 2026)2026-11-19 · theneuron.ai
- Everything That Happened in AI Today (Friday, November 20, 2026)2026-11-20 · theneuron.ai
- Anthropic shares close 17% above IPO price after volatile Nasdaq debut2026-11-20 · thenationalnews.com
- Photos: S.F. anti-AI protest hits Anthropic on day of $1.5 trillion IPO2026-11-19 · missionlocal.org
- One of Anthropic's Biggest I.P.O. Winners Feels a Bit Uneasy About the Whole Thing (NYT linkpost)2026-11-19 · lesswrong.com
- What comes after AGI? From AGI to Superintelligence2026-11-19 · institute.deepmind.com
- Anthropic ANT Closes at $97.17, Up 17% on Its First Day of Trading2026-11-19 · invezz.com
- Anthropic IPO Live: Stock Closes Day At $97—Up More Than 17%2026-11-19 · forbes.com
- Amodei vows to prioritise AI safety over business, political pressure after Anthropic's IPO2026-11-19 · newsdrum.in
- Stock Market Today, Nov. 19: Nvidia Slides as Investors Weigh CEO Transition Fallout2026-11-19 · fool.com
- Stock market today: Dow sinks 1,100 points, S&P 500 and Nasdaq tank 2% as Anthropic IPO grabs spotlight2026-11-19 · finance.yahoo.com
- SoftBank suffers worst day in over two years as AI sell-off hits Asian tech stocks2026-11-20 · cnbc.com
- Asian Chip Stocks Plunge Despite Anthropic's Record-Breaking IPO Debut2026-11-20 · gulfnews.com
- Anthropic could be the real winner in OpenAI's latest model release2026-11-18 · gizmodo.com
- Chairman Brian Mast Exercises Legislative Authority to Publicly Release AI Chip License Information for China2026-11-19 · foreignaffairs.house.gov
- US signed off on sales of nearly 900,000 AI chips to China in six months2026-11-19 · aljazeera.com
- China's Moonshot drops $150M Indonesia Nvidia deal days after US crackdowncryptopolitan.com
- Hackers use AI tool to disrupt European Commission staff systems (syndicated The Register)news.lavx.hu
- European Commission confirms cyberattack, says hackers used AI tooltherecord.media
- Google gives Gemini the ability to use your Mac for you2026-11-19 · techcrunch.com
- Apple is building an AI agent that can control apps, browse the web and write code, Bloomberg reportsmlq.ai
- Apple plans autonomous agent to handle tasks across its devicesinterestingengineering.com
- Nvidia in advanced talks to buy diffusion AI startup Inception for over $500M - reportseekingalpha.com
- AUDIT AI Act: 4 Senators Want AI SPVs and Leases Disclosedfourweekmba.com
- US senators seek greater disclosure of AI financing risks2026-11-19 · socialnews.xyz
- Google Workers Urge Bosses to Give Schools Search Without AI2026-11-19 · bloomberg.com
- Google Employees Petition to Let Schools Use Search Without AIgizmodo.com
- Google Selects Blue Origin's New Glenn for Project Suncatcherblueorigin.com
- Google signs deal with Blue Origin to launch AI computing satellites2026-11-19 · mediasat.info
- Nick Bostrom: How Should We Navigate Superintelligence (a16z podcast)a16z.com
- Hy4 flash (Hugging Face model card)huggingface.co
- Tencent Opens Hy4 Flash Weights With 1M-Token Context at 20x Lower Costalphasignal.ai
- Research acceleration: The jagged frontieropenai.com
- Advancing chip design with our modelsopenai.com
- More banks say customers using AI agents are on their own even if they get hacked2026-11-18 · techcrunch.com
- Xi, Trump, and Putin Summit Makes History but Not Much Else2026-11-19 · time.com
- Perplexity Launches Orchestrator API for Failover and Model Routing2026-11-19 · marktechpost.com