Latest / Super Intelligence News and Report / Anthropic's IPO Hits a Wall: Investors Push Valuation Toward $1 Trillion
Transcript
- Super Intelligence News and Report — Anthropic's IPO Hits a Wall: Investors Push Valuation Toward $1 Trillion Welcome to Super Intelligence News and Report. Today is Saturday, November 7th, here are the headlines. Theo: Today's thread: the models keep getting faster and scarier, and everyone from voters to judges to IPO buyers is getting harder to win over. First up, OpenAI. On November 5th it introduced GPT-6 Cyber, a version of GPT-6 Astra trained to find vulnerabilities, build exploits and write fixes. It's the model behind that Linux exploits research post we covered, now named and shipping. Maya: Shipping to whom, though? Theo: Only to vetted offensive security testers in its Daybreak Red program, through the API and Codex. OpenAI rates it Critical for cyber under its Preparedness Framework, a step up from High for the last version. A Firefox engineer says it found 81 previously unknown high-severity Firefox bugs. So top-tier offensive capability is now a gated product, and access decides who benefits. Skeptics on X are grumbling that it's yet another powerful model most people can't touch. Maya: And speaking of things the public can't see. Theo: Right. Also on November 5th, the nonprofit METR published an outside risk assessment of Anthropic, built on internal data and staff interviews it's had access to since August. It finds Anthropic's capabilities are improving about twice as fast as a year ago, and it judges, with stated uncertainty, that about 80 percent of that speed-up comes from Anthropic using AI in its own research. The catch is that Anthropic redacted most of the evidence METR considered most informative. Anthropic's McCandlish says those details could help rivals go faster. Maya: So the first outside check on AI speeding up AI research is mostly a black box. Theo: Mostly. And METR says Anthropic's own public report from August understated the acceleration. Your turn, and it's Washington. Maya: Three House chairs want Jensen Huang in a closed-door briefing on November 18th. That's Moolenaar from the China Select Committee, Mast from Foreign Affairs and Crawford from Intelligence. Their letter asks about his reported interactions with Alice Huang, a Megaspeed executive it calls a publicly reported Chinese intelligence asset, and about Nvidia's protections against espionage. He has until November 11th to accept or offer another date. Nvidia says regional cloud customers get only ordinary business information. Why it matters: export-control politics are now aimed at Nvidia's CEO personally, in a week when chip stocks are already fragile. Theo: Not great timing for anyone near a listing either. Maya: Which brings us to Anthropic in court. On Friday, November 6th, Magistrate Judge Alex Tse in California trimmed some claims but let key parts of a proposed class action over Claude Max go forward. His point was that customers could reasonably read 5x or 20x more usage as a monthly promise, when the multiple actually applies per five-hour session. The plaintiffs figure Max 20x gets only about six times Pro's weekly usage on one model. He did say Anthropic didn't tighten limits as a bait-and-switch. Theo: Two links away from the sales page is a long way. Maya: A judge noticed. Usage-limit marketing is now legal exposure for every AI subscription business, and this lands a month before Anthropic's planned IPO. Next, Meta has shelved its new business terms indefinitely. Critics said they would have let Meta crawl the websites of businesses using its products. They were due November 20th, after already slipping from October 30th, so this is the second delay. It follows a Norwegian court ruling, a formal probe by the UK's Competition and Markets Authority and a European Commission review. The old terms stay, and the competition authority says it's still looking. Bottom line, publishers and regulators stopped a platform from getting AI-era data rights through the fine print. Theo: Meanwhile Amazon would like its agent to be the only one in the store. On November 6th it started letting invited US customers text Alexa for Shopping. It searches, compares, tracks prices and buys, either with your approval or automatically when a target price hits. Sponsored products can show up in its replies, free to advertisers for now. Amazon says the assistant already has more than 400 million users, and access expands to millions more in the coming weeks. TechCrunch points out Amazon blocked Meta's Muse shopping agent from its site in September. Maya: So agentic shopping stays inside Amazon's walls, with ads in the answers. Theo: Exactly. Now, does any of this show up in the economy? Maya: Not much yet, says the Bank of England. Its November report estimates AI has added about half a percent to UK productivity over three years, so a tenth or two of a point a year. Firms expect closer to a full point a year from here. The Bank's regional agents say AI is making firms less likely to hire or replace staff, but not more likely to fire. So far the payoff shows up for the people selling AI and as slower hiring, not as broad productivity growth. Theo: And the backlash isn't waiting for the productivity data. Maya: No. On Friday, about a hundred protesters broke into a Microsoft data-center construction site in Amsterdam, AP reports. Some climbed a crane and lit flares, and others clashed with police. It's part of a Stop Dirty Data Centres campaign across 16 countries, from Dublin to Warsaw to another Microsoft site in Denmark, demanding a moratorium on large data centers in Europe. The pushback we saw in US ballot measures is now physical and international. Theo: Europe's builders, meanwhile, are selling control. On November 5th Mistral expanded Mistral Studio with managed agents that run code in sandboxes and hand work to subagents, plus apps, event-triggered workflows, observability, evals and an MCP server for deploying from coding agents. It runs in Mistral's cloud, in your own VPC or on-prem. People on X are saying agents and apps are in public preview and workflows are generally available, with on-prem support later this month. Reported, not confirmed. The pitch is enterprise control, not raw model power. Maya: Any hidden gem today? Theo: WorkOS. In a November 6th update on its October breach, the identity company says the intrusion into an internal sales tool involved an AI-assisted offensive framework that automated reconnaissance and exploitation. It cites Wiz, Google and CrowdStrike seeing AI agents in more real attacks. A named victim confirming agentic attacks in the wild is the flip side of OpenAI gating GPT-6 Cyber. Maya: Lovely. Theo: Lighter one: Cursor now lets you talk to your coding agents. Describe a change out loud, ask what it's doing, hear spoken progress updates. Separately, GitHub now lets admins manage Copilot code review through its APIs and shows the AI credits each review burns. Coding tools are shifting from typing prompts to supervising agents, and they're starting to show the cost per task. Maya: And one from the hype file? Theo: People on X are passing around clips of Greg Brockman at the Journal's tech conference saying death could become a choice, and predicting big strides in curing disease within 12 to 18 months. Reported, not confirmed. The original remark hasn't been independently verified. Replies called the timeline hype with no clinical path. Still, OpenAI leadership is putting near-term dates on medical claims just as investors are questioning AI valuations. Maya: Speaking of which, that's where we're going next: the selloff has reached Anthropic's IPO price, and OpenAI's round came up short too. Let's dig in. Now for the Main Episode Maya: So here's the question we're going to try to answer. Can Anthropic still pull off what would be the biggest IPO ever, this month, now that the market has turned hard on AI stocks? And if it can, at what price? Theo: And price is the part that's moved the most. A month ago people were talking about two trillion dollars. Maya: Right. Let's start with where things stood before this week. Anthropic is trying to be the first frontier model maker to go public. City AM reports that its last private round, back in May, valued it at nine hundred sixty-five billion dollars. The hope was that the listing would push that to about one and a half trillion. Theo: So a fifty percent step-up from the private mark just by ringing the bell. Maya: That was the plan. And the timetable, which Reuters reported earlier, was tight. The roadshow would start the week of November 9th, pricing the week after, and trading before Thanksgiving. Yesterday we told you the roadshow could start as soon as Monday. That's what this is an update on. Theo: Then the election happened. Maya: Then the election happened. Democrats swept the midterms on Tuesday, and in a lot of places voters backed local limits on data centers. AI stocks fell for two straight days. Thursday was the brutal one. The Nasdaq 100 dropped six point four percent. Bloomberg puts that at two point three trillion dollars of market value gone in one day, which is a record. Theo: Just to put that in Anthropic terms, that's more market value than the most optimistic Anthropic valuation, wiped out in one session. Maya: Yeah, that's a good way to hold it. And the VIX, Wall Street's fear gauge, hit its highest level since April of last year. So that's the backdrop. Thursday evening, the Wall Street Journal reports that major investors doubt Anthropic can get to one and a half trillion. City AM, summarizing that piece, says three prospective investors think the number has to come down, and one floated something closer to a trillion. Theo: And I want to flag the sourcing here, because almost everything this week is people familiar. Maya: It is. The Journal, Reuters, Bloomberg and The Information are all reporting from people familiar, so treat all of it as reported, not confirmed. And no price range has been filed yet. Anthropic declined to comment to Reuters. Theo: Okay. So what did Reuters actually add on Friday? Maya: Reuters had an exclusive with four people familiar. Potential investors doubt the roadshow can start next week in time for a November listing. They want a bigger discount, and they gave two reasons. One, Anthropic's record of quarterly losses. Two, its huge commitments to compute partners. And there's a third thing they're waiting on, which I think is the most interesting detail in the whole story. Theo: The October number. Maya: The October number. In its updated S-1, Anthropic left October revenue blank. Investors want to see whether growth picked up after a recent slowdown. Some expect the deal to raise about a hundred billion dollars. That would be the largest IPO in history, and that size is exactly why buyers have leverage. There's only so much demand for a hundred-billion-dollar order book, so the buyers get to set terms. Theo: Can I push on the blank for a second? Leaving a month blank in an amended filing isn't automatically sinister. The month just ended. But in this market, an empty cell reads like a question mark. Maya: Agreed, and I don't think anyone's alleging anything there. It's just that the whole debate now hangs on one number nobody outside the company has. Theo: So then Bloomberg comes in with actual figures. Maya: Bloomberg says prospective buyers are now discussing one point two to one point three trillion, and some are weighing as low as one point one. And here's the trajectory they lay out. In early October, investors pegged fair value as high as two trillion. After the company briefed them on its financials in October, that fell to about one point four or one point five. Now it's one point one to one point three. Theo: So how big is the cut, really? Because I've seen people throw around a forty-five percent haircut. Maya: That number is misleading on its face. Forty-five percent is the gap from the most bullish two trillion talk to the very bottom of the new range. Measured from the one and a half trillion target that was actually on the table, the top of today's range is only about thirteen percent lower. So the honest version is: a modest trim from the real target, and a big comedown from the hype. Theo: And what's worth noticing is that the first big drop came before the selloff. That October briefing knocked a quarter off the top-end number. The election just finished the job. Maya: That's exactly right, and I think it's the most underrated part. The market mood is the headline, but the numbers had already disappointed people. Theo: Which brings us to the filing itself. Remind me what was in the November 2nd amendment. Maya: Third-quarter revenue of about seventeen billion dollars. About thirty-three billion for the first nine months of the year. A return to an operating loss. And commitments of about eight hundred ninety-five billion. Theo: That last one is the scary one. That's about twenty-seven times this year's revenue so far. Maya: It is, but that ratio flatters the fear a bit. Those are long-dated contracts, some stretching up to two decades. It's not one year's bill set against nine months of sales. Still, you're signing up for obligations that assume revenue keeps compounding for a very long time. And that's the backdrop to Microsoft. Theo: Right, the neocloud thing. Maya: Reuters, citing Bloomberg, says Microsoft tightened oversight of some data-center projects run by Nebius, IREN and Nscale. Again, reported, not confirmed. Nebius, IREN and CoreWeave have each lost about half their market value in a month. If the biggest renter of compute is getting choosy, an investor looking at Anthropic's commitments is going to ask whether the whole compute economy is repricing. Theo: Okay, here's where my beat comes in, because I think the Ramp data is the real story underneath all this. Maya: Go. Theo: TechCrunch reported new data from Ramp, the corporate card company, on how much businesses spend on tokens. In July, spending grew forty-two percent month over month. Then across August and September combined, it grew one and a half percent. Basically flat. But usage kept rising the whole time. Maya: So people used more and paid about the same. Theo: Exactly. And Anthropic basically confirmed the mechanism itself. It said in a filing that price cuts on Fable 5.1 and Opus 5.5 reduced its revenue growth by about six hundred million dollars. So part of the slowdown investors are worried about is self-inflicted. Anthropic made its models cheaper and the bill went down. Maya: And Ramp's lead economist, Ara Kharazian, draws the bearish conclusion from that. He told TechCrunch that almost all AI spending comes from venture-backed tech companies that could slow down together, and that the era of hypergrowth may be over, because price cuts make AI cheaper instead of growing revenue. Theo: See, I half agree with him. The concentration point is real. If your customers are all funded by the same pool of venture money, they tighten together. But I don't buy that price cuts mean hypergrowth is over. From the builder side, cheaper tokens are what make agents viable. You don't run long agent loops on expensive models. Usage rising while spend flattens looks to me like a price war, not like demand dying. Maya: But a price war is exactly what a public investor is pricing in. It doesn't matter whether demand is healthy if the labs keep competing it away. That's why I think the question shifted this week. It's not just mood anymore. It's whether competition caps revenue growth. Theo: Fair. Though the Ramp data has a coda. Spending picked up again, up about nine percent between late September and late October. Maya: Which is why October revenue matters so much. If Anthropic's own October looks like that rebound, the bulls get their story back. If it looks like August and September, the buyers asking for a trillion look right. Theo: Let's talk about the buyers, then. Who's saying what? Maya: On the skeptical side, Max Kettner, the chief multi-asset strategist at HSBC, told Bloomberg Radio that about a trillion could be a more comfortable valuation. His worry is what he calls a SpaceX-style debut, where the stock pops for a few hours and then falls. Theo: That's the outcome nobody wants, including Anthropic. A broken IPO turns into a story about the whole company. Maya: Then there's Keith Lerner, chief investment officer at Truist Advisory Services, who told Reuters there's something like a buyers' strike in tech. Over the past month tech has lagged the S&P 500 by almost eleven percentage points. That's the most extreme one-month gap since May 2002. Theo: The dot-com hangover. That's quite a comparison to reach for. Maya: It's not subtle. And Stephan Kemper at BNP Paribas Wealth Management in Germany told Bloomberg the valuation may need to sit ten to fifteen percent below the low end of the rumored one and a half to two trillion range. Theo: Which, if I do that math, lands around one point three trillion. So Kemper is basically describing the top of where buyers already are. Maya: Right, and that's what makes the bullish side interesting. They're not arguing for two trillion. David Rainville, a technology portfolio manager at Sycomore Asset Management, told Bloomberg that at around one point two trillion, Anthropic shouldn't have trouble finding investors. Theo: So the bull case is the bear case's price. Maya: Pretty much. The fight isn't really whether Anthropic is worth owning. It's how much of the old number it has to give up. And Alexandre Baradez at IG made the cleverest point, I think. He told Bloomberg it may actually be better to list after the valuation excesses have already been purged. Theo: I like that one. If you list into a market that's already been flushed, you're less likely to be the thing that gets flushed next. Maya: And then there's a policy angle from Malik Ahmed Khan, a senior equity analyst at Morningstar. He told City AM that Anthropic's safety focus could help it if Democrats push for tighter AI regulation. Though he also flagged rising uncertainty for data centers, especially in Texas. Theo: I'm skeptical of that one, honestly. Safety branding might win you goodwill in a hearing. It doesn't change your gross margin, and it doesn't build your data centers. And this same filing added a risk factor saying future models may soon meet Anthropic's own definition of recursive self-improvement, AI that substantially builds the next AI. Maya: How does a public investor even read that? Theo: From where I sit, it cuts both ways. It's a capability claim, which is bullish if you think it means faster products. And it's a regulatory magnet in a Congress that just shifted. Khan's framing assumes regulators reward the careful lab. They might just regulate the most capable one. Maya: That's a fair push. I'd say Khan's point is relative, not absolute. If rules tighten on everyone, the lab that already built the compliance muscle probably suffers least. But I take your point that it's not a valuation driver this month. Theo: Okay, so here's what I really want to know. Does Anthropic go Monday or not? Maya: Honestly, nobody knows yet. The Information says Anthropic plans to go ahead with formal marketing next week, which is also reported, not confirmed. Bloomberg says deliberations are ongoing and no final decision has been made. And there's a procedural floor. SEC rules require the filing to be public for at least fifteen calendar days before formal marketing. Pricing usually comes about a week after marketing starts, though Bloomberg notes some companies wait weeks or months. Theo: And there's a middle path Bloomberg described, right? Maya: Start marketing with a conservative size and a conservative price, then raise both if demand is strong. It's a standard playbook. But Bloomberg pointed out that Cerebras and Figma both upsized their IPOs that way, and both now trade well below their IPO prices. Theo: So the playbook that makes the first day look great can be the one that sets up the slide. Maya: Exactly. Upsizing tells you demand on day one, not fair value on day ninety. Theo: And the PitchBook view was more about timing? Maya: Nalin Patel, who runs EMEA private capital research at PitchBook, told City AM that a delay protects against near-term valuation risk. But if the IPO has to happen by a certain date, a valuation cut could draw in more new investors. So it's a trade. Wait and hope the price recovers, or cut and widen the buyer base. Theo: Now here's the thing that I think changes how all of this reads. OpenAI. Maya: Yes. Bloomberg reported Friday that OpenAI closed the first eleven billion dollars of a round valuing it at one point two five trillion, with the money transferred Friday. Reported, not confirmed. In late September, OpenAI was reportedly looking for thirty billion. Theo: So OpenAI raised about a third of what it wanted, at least in this first close. Maya: In the first close, yes. We don't know what comes after. But look at where that puts the two companies. When the Anthropic talk was one and a half to two trillion, it was being priced well above OpenAI. Now the IPO range of one point one to one point three sits right on top of OpenAI's private mark. Theo: Which is a big repositioning. Over the summer and into the fall, the narrative was that Anthropic had overtaken OpenAI on business revenue and deserved a premium. That premium just disappeared. Maya: And here's where I think we might actually see it differently. My read is that OpenAI's mark is now the anchor. Any buyer sitting across from Anthropic can say, why pay more than the private price on your biggest rival? That's a ceiling. Theo: I'd call it a floor, though. OpenAI's round is private and illiquid, and it still cleared at one point two five. A public Anthropic stock you can sell tomorrow is arguably worth more, not less, at the same number. If I'm Anthropic, I point at OpenAI and say, you're getting liquidity for free. Maya: That's a real argument, and in a normal market I'd buy it. But OpenAI also came in way short of its target. That tells me private buyers are rationing too. A round that's a third of plan isn't a strong floor. Theo: Okay, that's fair. Let's say it's an anchor and leave whether it's a floor or a ceiling to the order book. Maya: Deal. Theo: Can we talk about Friday's bounce for a second? Because it complicates the doom story. Maya: Sure. AI stocks stabilized. The Nasdaq rose about one percent. Nvidia was up almost four percent and AMD almost eleven. But the S&P 500 still had its worst week since June, according to Yahoo Finance. So one good day doesn't erase a bad week. Theo: And for Anthropic, the stocks that bounced are chip companies. The stocks that got hammered and stayed hammered are the neoclouds, the people renting out compute. That's much closer to Anthropic's business. Maya: That's a sharp distinction. And there's a broader risk Reuters flagged. The market is so heavily weighted toward big tech that if a top AI company struggles to raise money, those valuations could get squeezed too. So Anthropic's deal isn't just Anthropic's problem. It's a test the whole index is watching. Theo: Which honestly might be the strongest argument for pricing it to work. A successful deal at one point two is good for everyone. A failed deal at one point five is bad for everyone. Maya: That's where I land too, actually. So back to the question. Can they pull it off this month, and at what price? My read is yes, they probably can, but not at the price they wanted. The deal everyone's converging on, bulls and bears alike, is somewhere around one point two trillion. That's still a big step up from May's round. It just isn't the victory lap. Theo: Mine is that the selloff is the excuse, but the real issue is the revenue curve. That October briefing knocked a quarter off the valuation before a single vote was counted. And the Ramp data plus Anthropic's own six hundred million says price competition is eating growth. October revenue is the whole ballgame. If it shows the late-September rebound, they go, and maybe they even get to upsize. If it doesn't, I think they wait. Maya: So the two things I'm watching. First, whether the roadshow actually launches the week of November 9th, and whether a new S-1 amendment finally puts a price range on paper, and ideally an October number. Second, if the timetable holds, pricing the following week, before Thanksgiving on November 26th. Theo: And I'd add one quieter tell. If they come out with a conservative range and talk about upsizing, remember Cerebras and Figma. Maya: Right. The question this week went from whether Anthropic is worth one and a half trillion to whether it's worth more than OpenAI. That's a much harder sell in a market that, for the first time in a long while, wants to see the receipts. Thanks for listening to Super Intelligence News and Report. Please subscribe and share it with your friends.