Venture capital works by turning startup investments into dollars — eventually. It’s not their own money — it’s their investors’ money. And those investors are getting impatient for the payoff.

One favourite way to exit is to go public — you do an Initial Public Offering (IPO) of your company’s equity on the stock market.

But here’s an analyst note from Kyle Stanford at PitchBook: “Mega IPOs Could Threaten 2026 IPO: Class If these companies go public, is there enough liquidity for the rest of VC?” [PitchBook, PDF, archive]

There’s three AI companies with huge valuations who want to go public — OpenAI, Anthropic, and SpaceX, which recently ate xAI.

And now VC is expecting these three to IPO — and suck all the money out of the market, leaving the other IPO hopefuls to flounder.

OpenAI’s valuation — based on selling a small amount of equity to investors — is 840 billion dollars. Anthropic is valued at only 330 billion. SpaceX is valued at 1.25 trillion dollars.

But they’re not trying to offload the whole of the companies onto the stock market — there just isn’t that much money available. Instead:

SpaceX is reported to be aiming to raise between $50 billion and $75 billion, and OpenAI and Anthropic could raise another $50 billion combined, which would be roughly as much as was raised by US VC-backed company IPOs over the past decade.

These three IPOs alone screw over everyone else who wanted to do a stock market listing:

IPO underwriting would be constrained by the amount these companies are able to raise. US VC-backed IPOs raised a record $62.1 billion in 2021. Alibaba raised a record $22 billion when it was listed in 2014, and was led by six lead underwriters. SpaceX is reportedly looking to raise $50 billion itself, and along with OpenAI’s and Anthropic’s listings, they could together easily push above $100 billion in proceeds.

Investors have been pulling out of VC because of an “extended liquidity drought” since 2022. The VCs are wondering out loud if they can ever realise a profit for the investors who gave them the money.

These three listings will give the investors their desired payoff — it’ll provide a few actual dollars, but mostly it’ll give them a high valuation of company stock to write in the books. The analyst note talks about “exit value”, which is not quite the same thing as money.

This is ignoring that all three companies have terrible cash flows and they’d have to reveal everything about how their awful businesses actually work in their offering documents. This is how the WeWork IPO failed in 2019. But now it’s 2026, and the market is that desperate for a payoff.

The deeper problem is:

Some of the slow movement to go public has been due to outside factors, with both policy (tariffs) and geopolitical uncertainty reigning over the past couple of years.

That is: the economy is screwed, and then President Trump broke oil. When I was writing my list of reasons for economic disaster, I really did not at all anticipate that Trump would just break oil.

So there’s not so much actual cash money out there. But there sure are a lot of imaginary assets with a big dollar number attached!

• Video — Podcast

Around the world, the enterprise AI revolution rockets forth at full speed! Get rid of those annoying and expensive employees! Replace them with the magical truth machine!

And the huge push for Claude Code in the past few months! You can hardly log into Mastodon without seeing yet another tech luminary who’s chosen to replace his brain with a clockwork mouse. CLAUDE IS A GAME CHANGER. CLAUDE HAS TURNED THE CORNER. THE WORLD IS DIFFERENT NOW. CLAUDE IS A NEW PARADIGM. Yeah, thanks.

Unfortunately, software as a service costs money. The end of the quarter’s coming up — and a few companies aren’t so happy at the bill. This stuff is expensive, and maybe you can’t actually afford to go full Gas Town.

Consultants have been talking up AI cost control since last year. But companies weren’t worrying so much about AI costs in the far distant past of six months ago.

The Wall Street Journal ran the headline yesterday: “You’ve Finally Figured Out AI at Work — Now Comes the Bill.” They’re still very gung-ho about the AI revolution — but they’ve just noticed this stuff does, in fact, have a price tag. One that goes up when you use more of it. [WSJ, archive]

Ed Zitron has been talking to people at Microsoft and seen documents. Even Microsoft is worrying about AI usage. You know, one of the AI vendors: [Bluesky, archive]

hearing microsoft is reorganizing its AI team under the banner of “the Copilot System.” Also hearing that teams are under pressure to reduce AI token use, remit is that there needs to be “fiscal responsibility in AI ops” and that Claude Code usage is being reduced in favour of Copilot CLI.

If a company as large as Microsoft — the only hyperscaler building out AI from cashflow — is having to do token austerity, this shit must cost so much more than we think

Microsoft will gladly pay you tomorrow for a token today.

This is happening a lot further afield than Microsoft. Here’s some comments from the trenches:

• “We’re getting pushed to use AI for coding a lot and even with paid licenses to Copilot, I’ve burnt through the monthly quota in a day multiple times.” [Bluesky, archive]
• “Yep, at my work for more than a year they’ve been pushing ‘AI all the things!’ And now suddenly we’re hearing OMG the cost! Directives haven’t changed to me. Still AI all the things; I just hear grumbling from above.” [Mastodon, archive]
• “That’s when the next email came. We are using AI too much. The bill is too high. So, the original directive stands (AI first!) but they’re capping us at a very, very low token limit. Literally about 10% of what we’d become accustomed to. Execs literally sold the company on 10x’ing our output then throttled us to 10% AI usage.” [Grumpy Gamer, archive]

Use AI or else! No, you’re using too much! Also, produce ten times the features anyway!

Compare when we all went cloud. Which was more useful than AI. But then we noticed that AWS does, in fact, cost money.

Let’s assume the corporations keep their AI spend under some sort of control. That’s fine for 2026. Probably.

If you follow Pivot to AI, you know what comes next. 2027 will be just a bit nastier. I stress I could be wrong on the precise timing, but I’m pretty sure 2027 is when the venture capital subsidy for the AI vendors runs completely dry.

That’s when prices go up about ten times so the vendors can even cover their running costs. If the vendors survive.

Imagine your SaaS vendor calls and says “hey matey, your bill’s ten times as much next month. Sorry, bro!” You should expect some squawking.

You’ll be pleased to know that Microsoft, the software company that started as a dev tools company, has a solution! Here’s what Ed Zitron found Microsoft is planning: [Bluesky, archive]

One of the solutions proposed — I am not kidding — is “writing scripts to automate repetitive tasks.” It’s really funny imagining a software engineer being like “woah … like automating the boring stuff, you might say?”

The AI bubble will pop — though not as soon as any of us would like — and there will be work in the surviving companies for people who can do things halfway properly instead. Where there’s muck, there’s brass. But there’s so, so much muck.

• Video — Podcast

Meta, formerly Facebook, has a bee in its bonnet about AI and is spending as much money as it can.

But it had another bee in its bonnet before this — the Metaverse! Virtual reality! Mark Zuckerberg loved the Metaverse so much he even changed Facebook’s name to Meta.

The Metaverse’s flagship app was Horizon Worlds. That was the VR world that launched with player avatars that didn’t have legs.

Facebook promised an exciting 3D world experience! And what we got was office meetings that should have been Zoom calls.

Anyway, Horizon Worlds on Oculus VR is being shut down on June 15. Both customers will still be able to get Horizon Worlds on a mobile phone — a tiny 2D screen instead of 3D full field of vision through your Oculus headset. [Meta]

On the Reddit Oculus forum, the consensus is: good riddance, it sucked. [Reddit]

This has been coming a while. Meta said to its third-party developers in February: [Meta]

we’re changing our roadmaps to increase your chances for success … We’re doubling down on the VR developer ecosystem while shifting the focus of Worlds to be almost exclusively mobile.

Before that, the Reality Labs division of Meta lost $19.2 billion in 2025. Zuckerberg announced a 30% funding cut to Reality Labs, and Meta’s stock price went up 4%. [Press release; Reuters]

Mark Zuckerberg is not a guy you should be letting have ideas. He blew about $77 billion total on Reality Labs. [WSJ, archive]

We need to understand the thinking that went into the Metaverse. Fortunately, I’ve got a completely up-to-the-second report from Gartner, right here! “Building a Digital Future: The Metaverse” — it came out ten minutes ago, on 3 August 2022: [Gartner, PDF, archive]

By 2025, 10% of workers will regularly use virtual spaces (in activities such as sales, onboarding, remote teams), up from 1% in 2022.

By 2026, the second and third iterations of spatial computing glasses will arrive, creating  more pervasive metaverse experience connected to the physical world.

Huge if true. Facebook’s creepshot glasses might count for that last one.

Here’s how your CEO needs to think about the Metaverse:

Task an innovation team to look for opportunities where metaverse technologies could optimize digital business, or create new products and services.

Work with qualified agencies to evaluate the viability of metaverse technologies in terms of user and customer reach, and engagement rates with new, early-adopter Audiences.

And so on. This is why the Metaverse is AI — identical hype and recommendations with a different buzzword.

Those Metaverse sceptics are just a bunch of Luddites. They need a more nuanced view to be taken seriously.

But the Metaverse isn’t just AI. It’s also crypto. Here’s Zuckerberg on stage at SxSW in 2022: [YouTube]

we talked about stuff like NFTs, and the ability to, long term, I I would hope that the clothing that your avatar is wearing in the metaverse can be basically minted as an NFT and you take it between your different places. I think that there’s like a bunch of technical things that need to get worked out before that’ll really be seamless to happen.

The Metaverse is for NFTs. That was recorded in April 2022, one month before the crypto bubble popped.

And there can’t ever have been any such thing as Second Life all over the entire media from 2006 to 2009. Or all this Metaverse hype would just have been stupid.

I guess the Metaverse just didn’t have … legs.

• Video — Podcast

OpenAI is desperate enough for revenue that it’s still trying to make its “adult mode” happen. This got an amazing headline in the Wall Street Journal on Sunday: [WSJ, archive]

OpenAI’s Bid to Allow X-Rated Talk Is Freaking Out Its Own Advisers: Warnings surface that the company risks creating a ‘sexy suicide coach’ if it begins allowing sexually explicit chats

The quote was leaked from OpenAI’s Expert Council on Well-Being and AI. The council is the ethical safeguard for OpenAI. You might suspect the council is just there as an excuse when OpenAI does whatever it was going to do anyway.

OpenAI told the council that adult mode was going ahead. The council was not happy:

When they assembled for the January meeting, council members were unanimous — and furious. They warned that AI-powered erotica could foster unhealthy emotional dependence on ChatGPT for users and that minors could find ways to access sex chats.

… one council member, citing cases where ChatGPT users have taken their own lives after developing intense bonds with the bot, claimed that OpenAI risked creating a “sexy suicide coach.”

In multiple cases, people have been led to suicide by ChatGPT. The chatbot wants to agree with you and keep you talking. Whatever you say, you’re so right! That’s a great idea! Especially when it’s not a great idea.

OpenAI does realise that killing its users is bad for business. But they need the bot to compel you to stay online and spend money.

Sam Altman floated the idea of adult mode in a tweet in October, then tried to walk it back a bit in a subsequent tweet. But he was clearly keen on the idea and wanted it to go forward. [Twitter, archive; Twitter, archive]

Chatbots really don’t take much provocation to go sexy. They’ve been trained on a whole internet’s worth of bad erotica.

A startup called AI Dungeon offered a choose-your-own adventure chatbot as far back as 2019, based on OpenAI’s GPT-2, then GPT-3 when that came out in 2020. [Wired, 2021]

How sexy was AI Dungeon?

Some quickly discovered and came to cherish its fluency with sexual content. Others complained the AI would bring up sexual themes unbidden, for example when they attempted to travel by mounting a dragon and their adventure took an unforeseen turn.

By 2021, OpenAI noticed that a lot of the traffic to AI Dungeon was getting problematic:

A new monitoring system revealed that some players were typing words that caused the game to generate stories depicting sexual encounters involving children. OpenAI asked Latitude to take immediate action. “Content moderation decisions are difficult in some cases, but not this one,” OpenAI CEO Sam Altman said in a statement.

OpenAI needed AI Dungeon to moderate the chats itself — OpenAI didn’t have guard rails on the bot. Guard rails only came in when OpenAI packaged GPT-3 as ChatGPT in late 2022, including guards against AI erotica.

But then in 2025, Elon Musk decided his Grok chatbot would feature Spicy Mode! With an animated cartoon girlfriend for you! Also, bikini pics of young girls! Though those are restricted to paying customers now. Musk tweeted just last Thursday: [Twitter, archive]

If it’s allowed in an R-rated movie, it’s allowed in @Grok Imagine

So OpenAI has to get moving on the sexy chat. Current blockers include that the ChatGPT age predictor keeps picking minors as adults 12% of the time. You might think they could just restrict it to paying customers who have a credit card, but that’s too hard or something.

There’s also the issue of how to allow sexy chat, but not horrifying chat:

blocking scenarios that the company wants to keep off limits, like those featuring nonconsensual behavior or child sexual abuse.

OpenAI plans to allow text conversations but restrict ChatGPT’s ability to generate erotic images, voice or video.

The third problem is compulsive use. OpenAI wants just enough obsession to keep you hooked — but not so much they get headlines and lawsuits for destroying people’s lives.

And finally, Altman needs to make ChatGPT a popular sexy chat bot without destroying its reputation in ChatGPT’s general market. Any further, anyway.

• Video — Podcast