The IPO That Refused to Wait: Why OpenAI's 2027 Tilt Just Broke the AI Capital Cycle's Most Important Trade

John Ian Lau — Technology & AI infrastructure desk

For most of June the AI capital-markets calendar had a near-perfect shape. SpaceX had cleared the largest tech IPO in history on June 12, drawing $250 billion of order book demand. Anthropic filed its confidential S-1 with the SEC on June 1 at a $965 billion post-money valuation. OpenAI followed on June 8 with its own confidential filing, with Goldman Sachs and Morgan Stanley booked as lead bookrunners on both deals and an autumn pricing window in the cards. Goldman strategists had already guided the buy-side to expect $160 billion to $225 billion in 2026 US IPO proceeds, the strongest pipeline in nearly a decade. By Wednesday this week the trade was simple — long the underwriting franchises, long SoftBank for its 32% economic interest in OpenAI, long the AI infrastructure complex into a year-end re-rating catalyst. Then The New York Times published a Thursday-evening story citing three people involved in OpenAI's deliberations, and by Friday's open the entire structure had inverted.

OpenAI is now leaning toward holding off its IPO until 2027 rather than pricing this autumn, the Times reported. The company's bankers had presented Chief Executive Sam Altman with two paths: list later this year at a valuation below the trillion-dollar target the company has held since January, or wait until 2027 and price into a market more likely to award that number. Altman, per the reporting, treated any compromise on the trillion-dollar valuation as a non-starter. Chief Financial Officer Sarah Friar has been internally advocating the 2027 timeline. The market read both pieces of intelligence as confirmation that the autumn window is closed.

Why the price action mattered more than the headline

SoftBank Group shares fell as much as 13% in Tokyo overnight — the biggest single-session drop since August 2024 — wiping out roughly $24 billion of market value on a company whose investment case has been progressively rebuilt around its 32% economic interest in OpenAI. Morgan Stanley closed down 3.9% in New York; Goldman Sachs fell 4.1%, both touching their lowest levels since June 12. Among Asian semis, Advantest dropped almost 10%, Tokyo Electron lost 3% and South Korea's tech-heavy KOSPI fell 8% intraday, briefly triggering a 20-minute circuit-breaker halt. The Nasdaq slipped 0.46% in the cash session, but the index is now on track for its biggest monthly decline since March 2025. The Philadelphia Semiconductor Index is down 6% week-on-week.

The contrarian read on the day's price action is that this is not really about an IPO timeline. It is about the entire structure of capital pre-positioned around a confirmed autumn catalyst. When the catalyst slips by a year and the leadership at the most-watched private company in the world tells the market it will not bend on valuation regardless of conditions, every position built on the assumption of a defining 2026 listing has to be re-marked. The unwinding showed up most cleanly in SoftBank because its discount to net asset value had already compressed to roughly 15% on OpenAI excitement — the cheapest equity-implied exposure to OpenAI's pre-IPO value. Friday's move puts that discount back near 30%, and tells you something about how much hot money had stacked into the IPO trade through the Anthropic and OpenAI filings.

The analyst dissent

Sell-side reaction to the news split sharply along time horizons, and the divergence is the interesting part. Dan Ives, principal analyst at Wedbush Securities, published a note Friday calling the AI complex a "Twilight Zone market" — a phrase calibrated specifically to the gap between fundamentals and price action, not to a bearish call.

"This is year three of a 10-year AI buildout," Ives wrote, comparing the current spend phase to "the construction of the Las Vegas Strip in the 1950s, when enormous capital went in long before the economic returns became visible." Ives's earlier framing of Anthropic's S-1 filing as "an opening of the floodgates for the IPO market" — a market that has been "relatively dormant for a few years" — has not been retracted. The position is the same: the IPO pipeline is opening, but on a longer fuse than the consensus had priced.

That is the most important distinction running through the post-print analysis. Ives is not arguing the deals will not happen. He is arguing that the bridge between AI capex and AI cash flow is being financed in real time by capital that wanted a 2026 exit and may now have to underwrite a 2027 one. For the underwriting banks that translates into a fee-revenue calendar that has just been deferred by twelve to fifteen months — a meaningful change in the discount rate the market should apply to current Morgan Stanley and Goldman Sachs multiples. Brian Finneran, a Truist Securities analyst, framed it almost identically in a Friday client note flagged across European trading desks: the delay raises legitimate questions about whether the broadly bullish capital-markets narrative built into bank multiples is cooling, but a staggered IPO calendar is itself a positive signal for 2027 numbers. "It de-stresses the schedule," Finneran wrote, "and could lift 2027 estimates."

The bullish counterpoint, articulated overnight on Bloomberg Television, is that Anthropic's October target now becomes the defining 2026 print. Bankers have privately told both Anthropic and OpenAI that whoever lists first will set the index for the entire AI category, according to Wall Street Journal reporting earlier this month. If OpenAI steps back, Anthropic by default inherits the role of price-setter at a $965 billion valuation and a roughly $60 billion raise — already the largest AI listing in history if it prices on the October target. The information barriers the banks were operating between the two deal teams suddenly look less like prudence and more like an option on which deal proceeds first.

What the delay actually solves

Read the OpenAI decision through the lens of capital allocation rather than market timing and a different picture emerges. The company carries roughly $600 billion of compute infrastructure commitments through 2030, and its current cash burn cannot be plausibly underwritten at a sub-trillion-dollar listing without diluting late-stage private holders who paid up at the $852 billion March 2026 mark. Friar's preference for 2027 reflects an internal view that the company will look meaningfully different — and more clearly worth $1 trillion — after another twelve months of enterprise contract growth, advertising-revenue ramp, and the regulatory clarity the Trump administration's staggered-release framework for GPT-5.6 has introduced. Altman's refusal to lower the price tag is consistent with a CEO who has spent two years building investor expectations around a specific number and understands that anchoring at the wrong level on day one of public trading is a far more expensive mistake than waiting another year.

The risk to the 2027 thesis is not that the company changes its mind. It is that the comparable transaction prices below trillion-dollar levels in the meantime. If Anthropic prices its October listing at $850 billion rather than $965 billion — a meaningful possibility given the current Nasdaq tape — the trillion-dollar anchor for OpenAI becomes harder to defend even with another year of execution. The pricing of the Anthropic book in the next four months therefore matters disproportionately for what OpenAI's eventual 2027 valuation looks like, even though the two deals are walled off from each other on the same syndicate.

Our view

The cleanest cross-asset read of Friday's price action is that the IPO trade has not died but has been deferred and re-routed. Owning the underwriting franchises into a 2027 calendar is structurally different from owning them into a 2026 calendar — fee timing pushes out, capital is tied up longer on bridge financing, and the option value on which AI deal lists first re-prices entirely to Anthropic. The SoftBank trade looks asymmetric on the long side at current spreads to NAV, provided you can tolerate a year of carry on what is now an option premium rather than a known catalyst. The harder trade is in the AI infrastructure complex itself — Micron, Nvidia, AMD, Broadcom — where the implicit assumption was that a successful OpenAI listing this year would re-rate the entire compute supply chain on multiple expansion alone. That re-rating now waits twelve months. Position accordingly: long Anthropic-linked exposure where you can find it, long SoftBank into NAV discount widening, neutral the banks, and patient on the chip names that benefited most from the catalyst calendar that just got re-written. The floodgates Dan Ives described are still opening. They are opening more slowly than the market wanted, and what comes through them first is no longer the company most investors had been positioning for.

Solomon Grey Capital research notes are for informational purposes only and do not constitute investment advice. The authors and affiliated entities may hold positions in securities or instruments mentioned. Quotations from named sources are sourced from publicly available reporting cited in context and have been independently verified at the time of writing.