Newsletter · · Ashutosh Agarwal
The AI Bubble Has Moved Into Venture and Private Markets - The VC Read - Week of July 23, 2026
Startups and venture newsletter for the week of July 23, 2026. The sharpest AI-bubble call this week was not about the stock market but about venture itself: Mark Cuban, Bill Ackman and a 20VC roundtable argued the froth has been privatized, so any reckoning lands on VCs, their funds and private equity, and the whole debate turns on one word, liquidity.
The VC Read
Week of July 16–23, 2026: The AI Bubble Has Moved Into Venture and Private Markets
The sharpest AI-bubble call this week was not about the stock market. It was about venture itself, and this time the blast radius runs straight through the funds.
A quick note on the week: the two shows this letter usually leans on for the macro view, The a16z Podcast and Invest Like the Best, did not publish fresh big-picture episodes in the last seven days, and This Week in Startups did not surface in-window episodes on these themes. Even so, there was plenty to forward: a rowdy 20VC roundtable, Mark Cuban and Bill Ackman making the same argument from different chairs, and a genuinely useful conversation about how founders actually get their money out.
The Big Debate: If This Pops, Who Actually Gets Hurt?
For three weeks running, the AI-bubble argument on these podcasts has been about the public stock market: is Nvidia the next Cisco, are the mega-caps the next dot-coms, and so on. This week the smartest people quietly moved the question. Almost nobody is claiming that Google or Microsoft are priced like Pets.com anymore. The new claim is narrower and, honestly, more uncomfortable if you work in venture: the crazy prices and the reckless capital are not in the public markets. They are in private markets. So if there is a reckoning, it lands on VCs, their funds, and private equity, not on the average person's 401(k).
Mark Cuban put it about as plainly as anyone can. Sitting down with the All-In crew, he was asked straight up whether he is worried about a bubble:
It's not the traditional dot-com bubble... you don't see that at all today. So it's not a bubble that's going to impact most people in the room, or most people across the US. But it could just destroy a lot of VCs and a lot of funds and a lot of PE. Because they're going all in.
That is the whole thesis in three sentences. Cuban's point is that the froth has been privatized. The market leaders, he named Google and Meta, are "borrowing hundreds of billions of dollars," issuing 50-year bonds, and spending their entire cash flow on data centers on top of that. He called it "planning for perfection," and warned that if AI follows the price-performance curve that every other technology has followed, "there's going to be a lot of data centers that are going to be turned into pickleball courts." His analogy was the fiber boom of the late 1990s: everyone laid fiber, then the technology got so much better that we ended up with "dark fiber that people bought for pennies on the dollar."
Jason Calacanis, next to him, made it personal about the venture math: "I've only done venture for just over 10 years, and it is wild to watch so many people who deployed at the wrong time just out of business. They just invested at the peak." He remembered getting into companies as angels at "$5 million, $10 million," and then "all of a sudden the request was $40, $50, $60 [million], and the product's not launched."
Bill Ackman, on Money Rehab with Nicole Lapin, walked in from the public-markets side and landed in the exact same place. He runs one of the most-watched hedge funds on Wall Street, and even he flinched at what private AI valuations have become:
I just saw a startup where they raised a couple hundred million dollars in their seed round at a $2 billion valuation. And that was just an idea. And their second round's going to be $5 billion. And they're far from having a product or a service or a dollar of revenue.
Then came the line that should worry anyone who thinks "the smart money will be fine": "The investors are probably the most sophisticated investors in AI companies with the most experience, and they're the ones setting the valuations at such super-high kind of levels." In other words, this is not dumb tourists overpaying. It is the pros.
Now, the steelman for the other side, because there is a real one, and it came from Rory O'Driscoll of Scale Venture Partners on the 20VC roundtable with Harry Stebbings and SaaStr's Jason Lemkin.
Rory's argument is that the giant late-stage checks are not recklessness, they are a new asset class that did not exist before, and it exists for a good reason. "No one in venture, when I started, forget '94, even in 2004, was writing $100 million late-stage checks. And now it's a thing. It is a new class of financial product: private $100 million checks in companies already worth a billion dollars." His key nuance: this is NOT companies "staying private longer." It is companies going "zero to billions of dollars in value creation in five years." OpenAI is not private because it is dragging its feet, five years ago it was "doing nothing." The late-stage private market, Rory argues, simply replaced what used to be the small-cap public market. When Fidelity or a crossover fund backs both OpenAI and Anthropic, that is not a governance scandal, it is exactly what a public-markets index investor would have done.
And the froth, on this view, is concentrated and rational. Carta data that the roundtable dug into shows ordinary seed prices are up only 10-20%, but the top 5% of seed rounds now price at $200 million. Why? Because a handful of bets, the "neo-labs" building foundation models, and the capital-hungry AI-infrastructure companies, genuinely need $200 million or $300 million just to get to the starting line. As Rory put it, "if it's going to take you $300 million to get something done, there's no point raising $20 [million] at $20 [million] pre." The example of the week: Chai Discovery raised $400 million at a $3.8 billion valuation, led by Index, Kleiner, Sequoia and Dimension. Even Ackman conceded the core of this: a company that goes from zero to "many tens of billions in revenue in a very short period of time," his example was Anthropic, can justify a much higher price than old rules would allow.
So where does that leave the debate? On one very important word: liquidity. The bull case only holds if those enormous private marks eventually turn into real cash, through an IPO, an acquisition, or a sale to another investor. The bear case is that a "valuation" on a term sheet is fiction until somebody actually pays it. Jason Lemkin is bullish here, and specific: "I've never seen such a liquid secondary market. In all of my top names, I can get out of them today with ease, and every single day I have buyers for them." Cuban is skeptical, and his prescription is blunt: stop hoarding and go public. Because this cycle was funded by private capital rather than public markets, he argues, there should be a wave of ordinary $50-100 million IPOs, so that disruptive startups have stock to use as currency to buy up the legacy companies AI is about to gut. His frustration with his own founders was the funniest and truest line of the episode: "I'm like trying to tell my portfolio companies, go public... they just don't think that's the right thing to do."
My read: everyone is quietly agreeing on the same fact and disagreeing only about whether it is fine. The fact is that risk, price and leverage have migrated out of the public markets and into venture and private credit. Whether that is a "new asset class" (Rory) or "planning for perfection" (Cuban) depends entirely on whether the exits show up before the cycle turns. Which brings us to the signals.
Signals
Is seed investing dead? A real firm just voted with its feet.
Jason Calacanis told the world he is shifting his syndicate out of very-early-stage checks and into "later-stage growth opportunities in some of the big names." On the 20VC roundtable, Jason Lemkin called it "a huge sign of the times," not easing in, but fully switching after building "multiple billion-dollar winners" as an early investor. His confession is the emotional core of the whole week: "I have an investment with a $100 million position that took me years and years and years to get to. And then just watching a growth round where someone comes in and invests $100 million, like, why did I bother?... What's the point of being the guy any earlier?" His phrase for early-stage board seats: "seats for suckers." Rory's pushback was the wiser half: some investors' entire edge is being the awesome first check (he named David Frankel, who "loves the craft"), and it would be "insane of Y Combinator to say we're giving up early stage." The money is running late-stage, but not everyone can, or should, follow it.
Source: The Twenty Minute VC (20VC), "Apple Sues OpenAI | Zuckerberg Back on X and Challenging Codex and Claude Code | SK Hynix's $26BN IPO | Is Seed Investing Dead: Jason Calacanis Departs Seed for Growth | Greylock Raises New $1.5BN Fund" (July 16, 2026).
"Discipline" or just a smaller pie?
Greylock raised its 18th fund, $1.5 billion, its first new fund in about two-and-a-half years. The polite word going around is "disciplined." Lemkin refused to play along: "It's not disciplined. It's optimized to make the most money... The guys at Greylock didn't sit back and say, you know, we could make twice as much money with a bigger fund, but we'd like to make half as much." His real point is about how GPs get paid: with a smaller fund you deploy faster and reach carry (your share of the profits) sooner, instead of running a multi-billion-dollar "platform." The genuine divide in venture right now: the Andreessens, Lightspeeds and Thrives building $5 billion mega-platforms, versus the Greylocks and Menlos raising $1-2 billion and raising often. Rory's kinder framing: both firms just celebrated 50 years and survived three brutal cycles, so there is real "survivor bias in that discipline."
Source: The Twenty Minute VC (20VC) roundtable (July 16, 2026).
The quiet liquidity crisis, and the workaround.
Here is the under-covered story of the week, and it is the missing piece of the Big Debate. "DPI," distributions to paid-in capital, i.e. actual cash handed back to a fund's investors, is what has gone missing while paper valuations soared. On Run the Numbers, Dave McClure (500 Startups) and Aman Verjee laid out why. When PayPal went public in 2002, "49/50ths of all the value created by PayPal was in the public market," employees and early backers could actually sell. Today, McClure noted, "zero percent" of the wealth created at Canva, or Stripe, or SpaceX, or OpenAI, or Anthropic has been realized in public markets, because they all stay private. So funds hit "7 to 10 years old" and start panicking about "liquidity and DPI and exits." Their fix, and their business, is the secondary market: buying shares from existing investors and employees before any IPO. McClure himself took "about $9-10 million off the table" via a secondary to buy a house, because a bank will not lend against illiquid startup stock. He estimates roughly 100,000 people in the Bay Area sit on $1-10 million of equity they cannot spend. That is the pressure valve keeping Lemkin's "liquid secondary market" alive, and it is a sign the asset class is starved for real cash, not that it is healthy.
Source: Run the Numbers, "Dave McClure and Aman Verjee on Angel Investing at Scale, 500 Startups, and Practical VC" (July 20, 2026).
$1 billion exits are not dead, but 1x exits are very real.
The roundtable's case study: Constellation Software bought TouchBistro, a $70 million-revenue Toast competitor that had stopped growing, for $70 million. That is a 1x-revenue exit, after a founder change, a wall of venture debt, and years of drama. Lemkin's brutal summary: "If you don't want to be worth 1x, do something before it's too late, man." And he thinks the decay is speeding up: he watched Marketo go "to zero" slowly, but now that AI makes switching software trivial, "it used to take a year to leave. Now it's one day." The valuation gap tells the story: Constellation buys these older B2B software companies at about 3x revenue, essentially betting "these kinds of companies don't last 10 years," while Bending Spoons pays about 12x for consumer apps it thinks will endure. The lesson for founders of good-but-not-hypergrowth companies: your exit window is narrower and lower than you think.
Source: The Twenty Minute VC (20VC) roundtable (July 16, 2026).
Are the labs themselves overvalued? The token economics are quietly repricing.
Lin Qiao, founder of the AI inference company Fireworks (which Harry Stebbings backed with a $10 million check after a 15-minute meeting, and which scaled to $1 billion in revenue in four years), made the bull-but-humbling case on 20VC: the cost of a "token" (the unit AI models charge for) will fall "10x in the next three years," and that will drive "100x usage." Crucially, open-source models are now "90% as efficient" as the frontier at "15x more cost-effective." Fireworks runs at 30-40% margins by choice, not the 70-80% software investors expect. Her worry is not a bubble, it is monopoly: "What I don't want to see is only one company owns intelligence." Meanwhile the public-market read-through is landing on the giants: analysts now see Alphabet's capital spending hitting about $195 billion this year and about $300 billion next, with hyperscaler capex eating 98% of operating cash flow, and Google's gross margin expected to slip from 71% to 65% under the weight of it. We have entered, as one CNBC segment put it, "the anti-token-maxing era," enterprises want cheaper, not just smarter.
Sources: The Twenty Minute VC (20VC), "Are OpenAI and Anthropic Overvalued? The Open-Source AI Reality | How Token Costs Will Fall 10x And Usage Will Explode 100x | The Future Is Not One AGI; It's Millions of Specialised Models with Lin Qiao, Founder and CEO @ Fireworks" (July 20, 2026); Squawk on the Street, "11AM Hour: Interactive Brokers Founder, What to Expect From Alphabet & Tesla Earnings & OpenAI Model Goes Rogue 7/22/26" (July 22, 2026).
The scarcity story is cracking.
RiskReversal's hosts argued the entire bull case rests on compute being scarce, and that story took hits this week. Meta and SpaceX's xAI both started renting out "excess compute" on short-term contracts (Elon Musk himself flagged they were "very short-term"), which, as the hosts put it, "blows a hole in the whole bull thesis of scarcity." SpaceX's shares got to about $225 around its listing and then "traded extremely poorly," which they warned could sour broader sentiment. And China keeps doing it cheaper: DeepSeek raised $7 billion at a $52 billion valuation, Alibaba committed $50 billion of capex over three years, and Moonshot's new Kimi K3 model narrowed the gap with US labs, all without Nvidia's best chips.
Source: RiskReversal Pod, "The AI Scarcity Myth Is Breaking" (July 20, 2026).
The IPO window cracked open, for memory, at least.
SK Hynix priced a $26.5 billion NASDAQ listing, the largest ever by a foreign company. It popped 13% on day one, then faded. Rory's framing: the three big memory makers (SK Hynix, Samsung, Micron) rode the AI capex boom to roughly 70% net margins from losses in 2023, and SK Hynix is up 6x, yet they trade at just 5-8x earnings because memory is famously cyclical, and "the minute more capacity comes online," prices fall. On Squawk on the Street, Interactive Brokers founder Thomas Peterffy said his customers are "hungry for IPOs" but bearish on AI itself: "much of the compute capacity we are currently building will be a write-off in 3 to 5 years," causing "severe compression in margins" and refinancing trouble, though he thinks it is "terrific" 10 years out, after a shakeout.
Sources: The Twenty Minute VC (20VC) roundtable (July 16, 2026); Squawk on the Street, "11AM Hour: Interactive Brokers Founder, What to Expect From Alphabet & Tesla Earnings & OpenAI Model Goes Rogue 7/22/26" (July 22, 2026).
Quote of the Week
It's not a bubble that's going to impact most people... But it could just destroy a lot of VCs and a lot of funds and a lot of PE. Because they're going all in.
Mark Cuban, on All-In with Chamath, Jason, Sacks & Friedberg, "Mark Cuban on the AI Bubble: Who Actually Gets Wiped Out?" (July 20, 2026).