Newsletter · · Ashutosh Agarwal
Sprint Through the IPO Window or Wait for Calm Waters - Startups and Venture - Week of August 20, 2026
Startups and venture newsletter for the week of August 13-20, 2026. The public-market window has cracked open and OpenAI, Anthropic and Databricks are picking opposite sides, Michael Furtick used the words down round about OpenAI on live television, and Compound's Michael Dempsey warned that when the venture bid softens it does not soften 10 percent, it softens 100 percent.
Startups and Venture
Week of August 20, 2026: Sprint Through the IPO Window or Wait for Calm Waters
The IPO window just cracked back open, the biggest private companies can't agree whether to run through it, and for the first time, someone went on live TV and said the words "down round" about OpenAI. The week of August 13-20, 2026, in venture, from the podcasts.
A quick, honest note on where this week's material came from. The venture-native shows we usually lean on were mostly off-topic: 20VC spent the week on Canva's valuation and CEO pay packages, a16z ran founder interviews (Travis Kalanick, Stripe's Patrick Collison), and Invest Like the Best went deep on Big Tech and China. The sharpest venture-macro arguments this week actually came from the finance shows, CNBC's Squawk franchise, Motley Fool, The Delphi Podcast, The Full Ratchet, and Prof G, where a genuinely new debate broke out. So that's where we went.
The Big Debate: Sprint through the window, or wait for calm water?
For about a year, the whole venture conversation has been stuck on one question: is this an AI bubble? This week the question quietly changed to something more practical and, frankly, more useful: the door to the public markets is opening again, who should walk through it, and who gets hurt when the giants do?
The reason it matters now is that the two companies the entire AI economy is built on, OpenAI and Anthropic, are both circling an IPO, and they're doing it in opposite ways. And the people running the next tier of companies are watching, trying to decide whether to follow.
The case for running through the door
The clearest voice on the "go now" side was Greg Brockman, OpenAI's president, on Squawk Pod. He wouldn't confirm timing, but his whole framing was urgency: "Now is the time to act." His argument is that demand for AI is effectively bottomless, "compute is really becoming the new oil, the new limited resource of the AI age", and that if you believe demand for problems to be solved is unlimited, you build (and raise, and eventually list) as fast as you possibly can. On the wave of C-suite departures (more on that below), he waved it off: OpenAI is "a very resilient organization," he and Sam Altman are "constants," and the only reason every exit gets scrutinized is that the company is "so much in the spotlight."
The market backdrop supports the sprinters. On Motley Fool's Hidden Gems Investing, Lou Whiteman made the cleanest version of the argument: growth-at-all-costs is exactly the right strategy right now because the money is there to fund it. "Profits are not a requirement. They are not, as we saw with the SpaceX IPO. Growth is required... You need profits when the market tells you you need profits." His warning to management teams was memorable: "You don't necessarily get an exit sign three miles in advance. You don't get a warning. So you better be ready for profits when they come." Translation: windows slam shut without notice, so when one opens, you go.
And Anthropic is going. Reports this week had it holding "test the waters" meetings with investors ahead of a listing later this year, explicitly selling itself on management stability, a not-so-subtle contrast with its rival. Its revenue is genuinely exploding: leaked second-quarter numbers annualized into the mid-$40 billions, and by mid-week some reports put its run rate near $65 billion. When your growth looks like that and buyers are lined up, the argument to wait gets very thin.
The case for waiting
The most compelling counter came from someone who could list tomorrow and is choosing not to: Ali Ghodsi, CEO of Databricks, in an unusually candid CNBC interview on Squawk on the Street (Aug 13). Databricks is growing more than 80% a year and is free-cash-flow positive, the textbook profile of a company that should IPO. Ghodsi's response to "why wait?" was essentially: because I don't have to, and going public right now would be a distraction. "I don't want to go public and have to sit there" while "the markets have been really wobbly... one great print by one software company and then the whole market is up 10 percent. And then the other day, everything is down 10 percent." He'll go "when there's a little bit more stable waters, calm waters, where there's a little bit more stability and predictability."
Crucially, he named the fear directly: 2021. "Seeing that story, how it played out... in 2021 people were so excited, it was like nothing, and then in 2022 it's the opposite," he said, the backlash that clobbered the Snowflakes and Affirms of the world. His point is that IPOing into froth isn't a triumph; it's a trap you spend the next two years digging out of. If you have the capital and don't need the money, why volunteer for it?
And this year handed the "wait" camp a live cautionary tale: SpaceX. On Prof G, Scott Galloway dismantled it. SpaceX priced at $135 a share, roughly a $1.8 trillion valuation, the largest IPO ever, bigger than Saudi Aramco, at 94 times revenue on $18.7 billion in sales, "richer than any company in the S&P 500." How? Galloway argues the demand was manufactured: only about 5% of the company floated (versus the usual 10%+), and the stock was fast-tracked into major indexes, forcing index funds to buy. The result was a pop to $225, followed by a roughly 45% collapse back to about $123. His verdict: "I think this is a $10 to $30 stock," and the sky-high analyst price targets "make Henry Blodget in 1999 look responsible." Even the most hyped IPO of the decade became a warning inside two months.
The swing factor: someone finally said "down round" about OpenAI
Here's what tips this debate from academic to urgent. On Squawk on the Street (Aug 19), Michael Furtick, founder of Vertic Capital and managing partner of Heroic Ventures, said out loud what nobody in venture will put their name to: OpenAI may never go public, and there could be a down round in its future.
His reasoning was specific. OpenAI has raised nearly $200 billion, far more than Anthropic's ~$130 billion, and is growing revenue roughly 20% quarter over quarter, "but their losses appear to be growing even faster... it's not clear that there's any end in sight." He pointed to the leadership exodus, and singled out one departure as the real tell: not the chief revenue officer, but Fiji Simo, the executive brought in to instill business discipline, who took medical leave and decided not to return. "This is someone who could really see where the momentum is going... the fact that she was not coming back was a signal that this was a very hard cleanup act to complete." Then there's the structure: "They've got a $200 billion preference stack. That's astonishing. It's historic. Maybe the largest ever of any private company ever." (A preference stack, in plain English, is the pile of money owed back to earlier investors before common shareholders and employees see a cent, the bigger it gets, the harder it is for an IPO to clear at a price that makes everyone whole.) His conclusion: "I suspect there may be a financial reckoning next year."
CNBC's own reporting on Squawk on the Street (Aug 14) fleshed out the instability. Kate Rooney reported that five senior leaders have left OpenAI in six months; that CRO Denise Dresser departed after just nine months (having doubled the enterprise book), a role now folded into Brockman's; and that investors describe the culture as a "pressure cooker," a "hire-fast, fire-fast type of place" where turnover is "almost priced in." Anthropic, meanwhile, is leaning into the contrast, literally pitching management stability to IPO investors.
So where does that leave us? The bull case (Brockman, Whiteman, Anthropic) says: the window is open, capital is abundant, growth is all the market wants, go, and go first. The bear case (Ghodsi, Galloway) says: froth is not a friend, SpaceX shows what forcing it looks like, and if you don't need the money, don't hand the market a stick to beat you with. And hanging over all of it is the uncomfortable possibility that the single most important private company in the world isn't racing toward an IPO at all, it's racing to stay ahead of its own burn.
Signals
"When the bid softens, it softens 100%." The best fund-level warning of the week came from Michael Dempsey of Compound on The Delphi Podcast, in an episode bluntly titled "Most Venture Funds Will End Badly." His thesis: too many funds are playing momentum, buying because they know a marginal buyer will pay more next round, not because the company is durable. That works until it doesn't, and here's the killer distinction from public markets: "In public markets, when momentum starts to slow, there is still a bid... it might come down 10 percent in a given day, it might come down 15, it might come down 20. In venture, there's no liquidity... and thus when the bid starts to soften, it doesn't soften 10 or 20 percent. It softens 100 percent." Dempsey, who deliberately sat out most of 2021-22 and runs "maybe the slowest deploying seed fund in the United States", also punctured the trendy $100M-seed logic: "It's a fallacy to believe that the highest price things are the best things at the early stage." You find out investing is hard, he said, not in the good 36-month windows, but "in 48-plus months."
The math on mega-funds doesn't work. On The Full Ratchet, Glenn Solomon of Notable Capital made the quantitative case for staying small, and it's worth sitting with. The number of truly large exits per five-year period is roughly constant, about 2,000 total outcomes, of which the top 1% is ~20 companies, or about four a year. "Even if somebody caught every top 1% deal in a year and owned 10% of it at exit", which basically never happens, a $10 billion fund still couldn't return itself. "The math doesn't work... the laws of physics don't break." He tied it to this month's marquee sale: Airtable, which raised well north of $1 billion and sold to Bending Spoons with roughly half the deal being cash Airtable had never spent. His point about over-funding: management teams "feel the need to spend that money to try to justify the valuation... even if the metrics aren't working, we better spend it", and most that do end up in a fire-sale. (Notable, for the record, got into Anthropic at a $61.5B post-money valuation, Solomon called it "by an order of magnitude the highest valuation I'd ever paid", back when Anthropic had gone from $100M to nearly $900M in ARR in a single year.)
Retail is being sold the giants, through trapdoors. The scariest concrete story of the week was on The Journal (Aug 13): a small investor, Ram Rubaretti, put ~$17,000 into an SPV (a special-purpose vehicle, a pooled fund created to hold shares of one company) run by a firm called Late Stage Management, believing he owned SpaceX exposure that couldn't be sold before the IPO. In reality he was layers removed, his SPV owned an interest in another SPV, sourced through a Bahamas-based firm, "SPVs all the way down." When SpaceX went public, he expected his ~2,500 shares to be worth more than $300,000. Instead the portal showed his SpaceX had been sold on December 31, 2024, a year and a half before the IPO, leaving him about $45,450. The same danger showed up from the buy side: on The Lifestyle Investor (Aug 13), veteran secondaries investor Yotis Tonnelier praised how secondaries now hand employees and early backers liquidity from Series A/B onward (he cited a 30% discount on a European Waymo competitor, targeting 5x in 2-3 years), but warned about the "hashtag Anthropic" problem, where employees signed to sell, then backed out after their stock quietly 15x'd. His rule: "Make sure that the shares are all inside the SPV," not merely a right to buy them. As more ordinary investors get pulled into private AI names, the plumbing is where they'll get hurt.
The AI-bubble machinery kept whirring, mostly on debt. We've covered this thread for weeks, so just the fresh bits: the big new worry is that the buildout is increasingly financed by borrowing and off-balance-sheet structures. A Wall Street Journal analysis (discussed on Tech Brew Ride Home, Aug 17) pegged roughly $3 trillion in off-balance-sheet AI commitments across nine top tech firms, nearly five times their ~$600 billion in reported capex. Nvidia's move to backstop up to ~$500 billion of GPU financing earned it the nickname "the Bank of Mum and Dad," and on Limitless (Aug 13), even BlackRock's Larry Fink compared the structure to the mortgage-backed securities of 2008 (the hosts pushed back that GPU demand is real and supply-constrained, unlike houses in 2007). Meanwhile the economics of the models themselves are deflating fast: per-unit token prices have fallen about 43% (from ~$2.04 to ~$1.16) as buyers pile into cheaper Chinese open-weight models, Databricks' Ghodsi noted his customers went from "please, no Chinese models" to "give us Kimi, give us GLM" once their CFOs saw the compute bills. And Alphabet is now free-cash-flow negative, having raised $50B+ in debt and issued ~$85B of new shares to fund the buildout. The bulls aren't wrong that demand is real; the question is who's holding the paper when the depreciation schedule catches up.
Quote of the Week
Michael Dempsey, Compound, on The Delphi Podcast (Aug 19):
"In venture, there's no liquidity. And so when the bid starts to soften, it doesn't soften 10 or 20 percent. It softens 100 percent."
The single sentence that explains why "the window is open, so go" and "wait for calm water" are really the same argument seen from two ends. In public markets you get a slow-motion exit. In private markets, when sentiment turns, there's no exit at all, which is exactly why the smartest operators this week were so obsessed with when to walk through the door.