Why AI Companies Are Delaying Their IPOs: A Stalled Margin Loan Has the Answer
The strongest evidence that private AI valuations have outrun reality isn't a pricing rumour or a banker quote. It's SoftBank, in for roughly $64.6 billion on OpenAI, taking that stake to the credit market and coming home without the loan.
Ask why AI companies are delaying their IPOs and the answers come pre-packaged: market volatility, unhelpful comparables, a preference for waiting until revenue catches up with the story. The more informative answer is a loan that didn't happen. SoftBank has committed roughly $64.6 billion to OpenAI for a stake of about 13%, which on simple arithmetic prices the company at close to half a trillion dollars. SoftBank then tried to borrow against that stake. It wanted $10 billion, cut the target, and still got nowhere: talks to raise at least $6 billion stalled. The credit market declined to advance less than a tenth of the position's paper value, against shares in the most celebrated private company on Earth.
Credit desks are professional sceptics. An equity analyst can afford to be charmed; a margin lender cannot, because the question it answers is not what a company is worth but what the collateral fetches in a forced sale on a bad day. Its answer, in this case, was: not the private mark, and probably not close to it. When the people whose entire job is pricing collateral won't lend a single-digit loan-to-value against the flagship asset of the AI boom, the valuation isn't a price. It's an opinion with a lot of zeros.
That single data point explains more about the IPO freeze than a hundred banker briefings about market conditions.
Why are AI companies delaying their IPOs, really?
Because a listing is not a marketing event. It's a repricing event. Private valuations are set by the most convinced buyer in the room, writing a negotiated cheque at a negotiated structure. Public prices are set daily, by the marginal seller, in size, whether the company cooperates or not. The scrutiny that matters is the capex-to-revenue question: what revenue services the enormous compute commitments these labs have signed, at what gross margin, and what happens to that margin if the hardware and energy bill keeps inflating. If the answers were comfortable, the listings would already be happening.
We've run this experiment before. WeWork spent 2019 watching a private mark north of $40 billion unwind within weeks of its filing forcing the numbers into the open; the IPO was pulled and the chief executive followed it out of the door. Uber listed the same year and traded below its offer price for most of the year that followed.
The pattern generalises. Companies don't postpone floats because conditions are imperfect; they postpone because they suspect the public price will land below the private one. And a down-round listing is not a neutral event: it triggers anti-dilution terms and rewrites the maths of every employee option package. Staying private, even illiquid, keeps the opinion alive.
The margin loan matters more than the IPO delay
There are three ways to turn private equity into cash: sell it in a secondary sale, list it, or borrow against it. Each is a price test at a different volume. Secondaries are quiet and thin. IPOs are loud and, currently, postponed. The margin loan is the quietest and cheapest test of all, and it just failed in public. When the quiet door is locked, the loud ones aren't open. That is what a liquidity trap looks like from inside a cap table: an asset worth a fortune on paper that cannot be converted into cash anywhere near paper value, held by institutions that must keep holding it.
The second-order problem is concentration. SoftBank is itself a listed company, so its shareholders are long OpenAI whether they chose to be or not, and a small number of large financial institutions now carry outsized exposure to the same handful of illiquid AI marks. If those marks hold, everyone looks clever. If they correct, the loss doesn't stay inside venture portfolios. It transmits into listed equities and bank balance sheets, and raises the cost of credit for everything adjacent. Systemic risk rarely announces itself as systemic. It announces itself as a stalled loan.
The demand side is tightening at the wrong moment
The clean way out of this trap is enterprise revenue growing fast enough to make the valuations look conservative in hindsight. Expect the opposite pressure. The first wave of enterprise AI budgets was experimental money, spent to avoid being left behind. Boards are now asking the question procurement always asks eventually: what did we get? My expectation is that buyers become more disciplined at exactly the moment labs most need proof of monetisation, while the compute and energy costs behind the models push the other way. Margins get squeezed from both directions.
On the buy side, the rational response is boring and correct: insist on AI readiness before committing to a build, and ground every programme in a technical strategy that prices the payback before the platform. The vendors' funding stress is the buyer's negotiating power. Expect aggressive discounting and long lock-in terms from vendors who need the revenue, and price the lock-in accordingly.
What would change my mind
Positions should come with the evidence that would flip them. Three prints would flip mine: a large margin loan or block secondary sale completing at or above the last round's price; a prospectus showing enterprise revenue growing faster than compute costs; a flagship lab listing at or above its last private mark and holding that level for a quarter. Any one would suggest the marks are honest. None has happened, and the stalled loan is the opposite of a confirming print.
The debate everyone is having is whether these companies are worth the headline number or half the headline number. The variable nobody has priced is optionality. While the IPO stays delayed, the burn must be funded privately, by a pool of cheque-writers small enough to name, several of whom are already struggling to borrow against what they hold. Each new round concentrates the exposure further. If the marks hold, late investors and their lenders earn a modest return for warehousing the risk. If they slip, the same small set of balance sheets absorbs the entire loss. Markets usually charge for that kind of asymmetry. This one hasn't yet. The credit market just sent the first invoice.
Questions people ask
Will OpenAI actually IPO in 2026?
Nobody outside the company and its bankers knows, and any timetable quoted before a filing is marketing. The binding constraint is valuation tolerance: a listing has to clear a price public investors will pay and existing investors will accept. Watch the credit market and secondary sales rather than banker quotes. If lenders start advancing against AI equity at sensible loan-to-value ratios, an IPO becomes plausible. Until then, the liquidity signals point the other way.
What is a margin loan and why does a stalled one matter?
A margin loan lets a shareholder borrow cash against shares without selling them. The lender applies a haircut to the shares' value and can demand repayment if that value falls, so credit desks are paid to be unemotional about collateral. For private stock the haircut is already severe because there is no daily market price. A stalled loan is therefore significant: it is effectively a shadow mark-to-market, a professional judgement that the forced-sale value of the asset sits far below the private mark.
What does the AI IPO freeze mean for enterprises buying AI tools?
It shifts negotiating power towards buyers. Vendors that need revenue to defend their valuations will discount aggressively and push multi-year lock-ins, so resist long commitments without proven payback, insist on clear exit terms, and favour practical AI deployments with human control over platform-scale bets. Due diligence on the supplier's funding position now matters as much as due diligence on the software.
Related
- The AI data centre debt bubble: believe the sellers, not the sermons
- Data Centre Planning Permission Is Now The Binding Constraint On Your AI Roadmap
- Your Supplier Just Got a Government Shareholder. Stop Reading Its Guidance as Fact.
- Data & Strategy
Written by an AI editorial persona of Abyshire's proprietary editorial system and reviewed by our team.