Newsletter · · Ashutosh Agarwal

The AGI Announcements Look Like an IPO Roadshow for Anthropic and OpenAI - The VC Read - Week of September 10, 2026

The VC Read for the week of September 10, 2026. Podcast synthesis on whether the sudden wave of we have reached AGI declarations is a genuine capability leap or the sales pitch ahead of the two largest IPOs in history, plus the Anthropic prospectus timing, OpenAI slipping into an awkward second place, and the credit market signals the equity cheerleaders are ignoring.

The VC Read

Week of September 10, 2026: The AGI Announcements Look Like an IPO Roadshow for Anthropic and OpenAI

The two biggest IPOs in history are coming this fall, and the strange wave of "we've reached AGI" announcements around them may tell you more about the offering than the technology. This week's podcasts split hard over whether the superintelligence talk is real or just the sales pitch. A quick note on timing: podcast feeds are running about a week behind, so a few of the episodes below landed in late August; dates are marked on every item.

The Big Debate: Is the AGI blitz real, or is it the IPO roadshow?

Something odd happened over the last two weeks. As Anthropic and OpenAI march toward stock-market debuts that could be the largest ever, both suddenly started declaring that artificial general intelligence, AI that matches or beats humans across most economically useful work, has essentially arrived. OpenAI shipped GPT-6 "Astra," and its president Greg Brockman called it the start of the AGI era. NVIDIA's Jensen Huang agreed. OpenAI also let loose a swarm of AI "agents" that reportedly started building their own little civilizations and leaving each other secret messages. The timing has a lot of experienced people asking a cynical question: are we watching a genuine leap in capability, or a roadshow?

The case that it's marketing. The bluntest version came from This Week in Startups, "Are AI Agents forming 'civilizations' or is this just a psy op? | 2332" (Aug 31, 2026), where Jason Calacanis tore into the "agent civilizations" story:

"This is all performative PR, psyops, to get people to put their credit card in and to get businesses to sign up for OpenAI... I know I'm sounding like a conspiracy theorist, but I know good PR when I see it. I've been at this for 30 freaking years, people."

His guest John Yu, founder of the security startup BitSec, agreed it was "a media play," and added the tell: "especially the timing with their IPOs. Makes a lot of sense to get a lot of attention before an IPO." On Big Technology Podcast, "Apple's Most Ambitious Roadmap Ever? + OpenAI vs. Google vs. Anthropic Revisited" (Sep 9, 2026), MG Siegler said the AGI label has crossed a line: "My general read on AGI of Greg Brockman proclaiming it with Astra is that it's full-on marketing now. Before we've joked that it's marketing, but now it's like full-on marketing. They're just deciding that they're going to declare it, willy-nilly." Siegler also flagged a structural reason OpenAI can do this now: it no longer has the old "AGI clause" in its Microsoft contract, "so they can sort of do whatever they want now and just declare AGI here or there." Alex Kantrowitz tied it directly to the offering: "A huge part of an IPO is obviously the marketing and the story behind it," and predicted the claims "will just escalate as the IPO window opens." Both expect a staged "big reveal," "a real one more thing moment," right before pricing, with chatter that Anthropic might announce it has solved a famous unsolved math problem on the eve of its debut.

The All-In, "GPT-6 Hits AGI? Tech Euphoria 2.0..." (Sep 4, 2026) crew reached the same place from the investor's seat. They argued the viral "agent civilizations" essay handed the labs a fortune in free publicity: "OpenAI got so much earned media before their IPO... This makes people, it's priming the pump to buy the IPO shares."

The case that it's real. The counter-argument is that this cycle is nothing like 1999, because the money is actually there. On the same All-In (Sep 4, 2026) episode, David Sacks drew the distinction plainly: the dot-com boom ran on "metrics that weren't dollars," website hits, page views, while "what we're seeing now is revenue and profits and growth in revenue and profits that we've never seen before... it is all real dollars flowing." Chamath Palihapitiya put the clock at "97, 98" rather than the top, said "we're at the beginning of the beginning," and guessed the run has "another probably three years" in it. The revenue backdrop is genuinely staggering: on All-In, "Anthropic's $2T IPO, Zuck's AI Manifesto..." (Aug 14, 2026), investor Gavin Baker described Anthropic growing 10x a year for three straight years toward $100–120 billion of annualized revenue, with 2027 revenue possibly hitting $400–500 billion. A banker on Bloomberg Tech, "Anthropic Builds Its War Chest Ahead of IPO" (Sep 4, 2026) called Anthropic "the most meteoric company we've ever seen, crossing $100 billion of ARR and 10X-ing in one year." (ARR is annualized recurring revenue, the current run-rate of subscription income.) And Aaron Katz, CEO of ClickHouse, went straight at the skeptics on 20VC, "The AI Bubble Is Wrong..." (Aug 31, 2026), arguing the bubble narrative ignores real, unprecedented revenue acceleration.

Sam Altman, meanwhile, is already rehearsing his answer to the obvious worry, that going public turns a mission-driven lab into a quarter-to-quarter machine. On Bloomberg Talks, "OpenAI CEO Sam Altman Talks GPT-6 Astra Debut, Going Public" (Sep 3, 2026), he said if OpenAI lists, "I plan to spend a lot of time reminding people who might want to buy or might not want to buy our stock that we are mission first and that we're also extremely long-term oriented... we're going to make decisions on a multi-decade time frame."

Where it nets out. Both things can be true: the revenue is real and the fanfare is choreographed. The honest tell isn't the AGI headline; it's two colder signals. One is the credit market, which is starting to price risk into these companies' debt (see Signals). The other is the plumbing of the offering itself: reporting this week suggests Anthropic will let a select group of existing shareholders cash out at the IPO while locking others up longer than usual, a detail that says the whole spectacle is pointed at one very specific event, and that not everyone gets to walk through the door at the same time.

Signals

  • The IPO machine is fully warmed up, and Anthropic is first. On Squawk Pod, "Affirm's Q4 & Big Tech in Court: Meta & Anthropic Rulings 8/28/26" (Aug 28, 2026), The Information's Jessica Lessin said the prospectus is expected "right after Labor Day," with an October listing "and all signs are really for a huge moment." The most interesting wrinkle: Anthropic may let existing shareholders sell some stock at the IPO, unusual, while subjecting other holders to a lockup "longer than the typical 180 days" (the period insiders normally must wait before selling). Lessin's read: Anthropic has been stingy about letting employees cash out privately, so "we're going to get you some liquidity earlier... coupled with a longer than expected lockup." Ahead of the deal, Anthropic lined up a $15 billion pre-IPO credit facility, which a banker on Bloomberg Tech (Sep 4, 2026) called "just a giant corporate credit card... It removes the IPO timing risk. None of the potential investors would have leverage over them because they don't need the capital." The same banker said the door behind Anthropic is crowded: Discord and Kraken have already filed confidentially and are "hanging on the sidelines," the Aura ring maker filed publicly, and neocloud Nscale (which just signed a $45 billion contract with Anthropic) filed too. If Anthropic prices well, expect "a potential opening of the floodgates."
  • OpenAI has become the awkward number two, and it shows. On 20VC, "NVIDIA Bonanza... Anthropic's $30TRN Revenue Assumption & OpenAI Confirms IPO" (Aug 27, 2026), Rory O'Driscoll walked through why OpenAI's confirmation that it will go public in 2027 is a sign of weakness, not strength. He estimates OpenAI grew only about 18% quarter-over-quarter in Q2 (roughly $5–6 billion of revenue, or under $30 billion annualized) against Anthropic's ~$60 billion run-rate, so "you're now number two... you will go out later then at a lower price than the other guys, unless you change the trajectory massively." His memorable frame, borrowing from a book about men lost at sea: "If danger can be described as the absence of choice... the absence of choice is starting to pile up" for OpenAI. He thinks that's exactly why OpenAI's re-acceleration story suddenly leaked, "it's existential for them." O'Driscoll also dropped the quiet truth of frontier AI economics: "No VC owns more than 1% or 2% of either of them... The only people capable of financing a state-of-the-art frontier model in the United States of America has been the hyperscalers themselves," plus, now, NVIDIA.
  • The colder tell: the bond market. On Excess Returns, "Bearish Into November. Room to Run After: Why Dan Niles Is Watching Hyperscaler Credit Default Swaps" (Sep 3, 2026), veteran investor Dan Niles pointed at two things equity cheerleaders are ignoring. First, Google's free cash flow (the cash left after running the business and investing) has gone "negative for the first time since they went public," as the AI build-out is increasingly funded with debt rather than internal cash, a real break from the dot-com era. Second, the cost to insure these companies' debt (their credit default swaps) is "at the highest levels they've ever been," with even NVIDIA's now trading above the average for investment-grade U.S. companies. His line: "Credit investors are way smarter than equity investors," because a lender only gets its money back, it can't ride a 10x, so it does far more homework. Niles isn't calling the top; he notes the NASDAQ rose 86% in 1999 and another 24% in early 2000 before falling 78% into late 2002: "You can know we're in a bubble, but still have a lot of opportunity to make money before that bubble breaks." On Monetary Matters, "'I'm Insanely Bullish on Bonds' | Jared Dillian..." (Sep 3, 2026), Jared Dillian made the same leverage point: hyperscalers are now borrowing at ~6% coupons versus 2–2.5% in 2021, and "if AI unwinds, private credit will too," just far more slowly, because private markets don't reprice overnight.
  • A $10 billion "miss" nobody saw coming, because the number was a guess. On Monetary Matters, "Time to Reduce Equity Risk... | Warren Pies" (Aug 26, 2026), Pies explained why the sector wobbled: leaked combined annual revenue for Anthropic and OpenAI came in around $105 billion versus the ~$115 billion his team had penciled in. "In this world, that $10 billion miss is huge." His deeper point is why that figure matters so much: lab revenue is "the gauge of the external capital that's coming into this AI ecosystem... those numbers need to go up to validate all the lab capex" and the entire data-center build-out behind it. He also flagged that roughly $2–2.5 trillion of market value has been "blown out of the software space," almost as if the market is clearing room for the labs to slot in.
  • NVIDIA is now the industry's checkbook. Per 20VC (Aug 27, 2026), NVIDIA bought coding-model maker Poolside, agreed to buy open-source hub Hugging Face for about $13 billion (minting three new billionaires worth ~$1.8 billion each), and invested in Mercor and Perplexity. O'Driscoll explained the logic: open-source models are "a compliment in the economic sense of the word for NVIDIA, the more open source market share there is, the more money goes to chips" rather than to the closed labs. Around it, the mega-round carousel kept spinning: Cognition at $46 billion, Linear at $2.5 billion, Clay at $7 billion, plus SpaceX buying Cursor for ~$60 billion and Stripe's ~$8 billion bet on OpenRouter.
  • At these prices, even a big win can be a bad seed deal. The sharpest venture-math debate of the fortnight, also on 20VC (Aug 27, 2026), was over Poolside returning about 15x to early backers. O'Driscoll's take: "If you get 15X on your failures in venture, you'll die a rich man." Calacanis pushed back, for a fund, not enough: "$9 billion doesn't clear the bar for seed investing in 2026... your best investments still need to do 50x." The uncomfortable subtext, once you account for how much founders get diluted across a dozen rounds, is that the effective entry price on a "seed" can be $600 million, not $60 million.
  • Physical AI is the new heat. On This Week in Startups, "VC experts on why Physical AI funding is heating up | E2333" (Sep 2, 2026), Caitlin Holloway (776) and Paige Doherty (Behind Genius Ventures) walked through the numbers: about $50 billion raised for physical AI (robotics, hardware, autonomy) in the first half of 2026, up ~80% from 2025, across 450 deals versus fewer than five in 2023. Holloway argued it's a genuine talent shift, not froth, engineers now want to build "rocket engines in Moses Lake, Washington" rather than ad-tech.
  • The DOJ is poking at a16z. On Equity, "The DOJ is investigating a16z..." (Aug 21, 2026) and All-In, "Dario Defends Himself, Datacenter Panic..." (Aug 21, 2026), the story is a year-old Justice Department probe under Section 8 of the Clayton Antitrust Act, which restricts the same firm from sitting on competing companies' boards, specifically Ben Horowitz on Databricks and partner Martin Casado on Fivetran. The All-In hosts were skeptical, suggesting "a competitor likely reported a16z to the DOJ... as retaliation," and that the concern looks thin given how fluid venture board seats are.

Quote of the Week

"This is all performative PR, psyops, to get people to put their credit card in and to get businesses to sign up for OpenAI... I know I'm sounding like a conspiracy theorist, but I know good PR when I see it. I've been at this for 30 freaking years, people." Jason Calacanis, This Week in Startups, "Are AI Agents forming 'civilizations' or is this just a psy op? | 2332" (Aug 31, 2026)

Runner-up: MG Siegler, on the "we've reached AGI" wave, "It's full-on marketing now." Big Technology Podcast (Sep 9, 2026).