The IPO queue forms: OpenAI and Anthropic file, Perplexity sets 2028 as its own clock
Lead.
On 8 June 2026, OpenAI confirmed it had confidentially filed paperwork for an initial public offering in the United States, the company said in a blog post on Monday, just over a week after rival Anthropic submitted its own confidential filing. The next morning, 9 June, a third name attached itself to the front of the queue: Perplexity, the AI-search startup, told CNBC it intends to list in 2028 on its own schedule, "regardless of what happens" to OpenAI or Anthropic. Three companies, ten days, and a market structure that no one fully priced in six months ago.
The clustering is the story. The frontier-model business, until recently the most jealously private corner of technology, is now openly sequencing itself into the public capital markets, and the smaller players are signalling they will not wait for the two leaders to set the terms. What is being tested, in real time, is whether the AI sector can sustain the private valuations it printed in 2024 and 2025 once they are forced to clear a public-markets audit, and whether the order in which the listings arrive will shape investor appetite for the ones that follow.
The ten-day window.
The trigger is recent and specific. On 8 June 2026, BBC News reported that OpenAI had filed plans to go public, a development that arrived "one week after Anthropic did the same." OpenAI's own blog post, summarised the same evening by TechCrunch, said the ChatGPT-maker had filed confidentially with US regulators, without disclosing a target valuation or a timeline. TechCrunch's reporting described the filing as a direct response to Anthropic's earlier confidential submission, framing the two as an "investment race" now being waged on a public-markets track rather than in the private venture round.
The 9 June disclosure from Perplexity, first circulated via a Reuters wire summary referencing CNBC's reporting, added the twist that will most likely define the autumn of 2028. Perplexity's leadership has communicated to staff and prospective investors that the company intends to list in 2028, on its own timetable, and will not defer that decision to the pricing windows of OpenAI or Anthropic. The framing — "regardless of what happens" — was the line circulated by prediction-market feeds on 9 June and is consistent with how Reuters characterised CNBC's report. Whether that timetable holds depends on the public-readiness of its own books and on whether market windows remain open; the company has not committed to file confidentially this year.
The order matters because of how public investors digest risk. The first mover in a category sets the narrative frame; the second mover is read against it; the third has to argue it is not a derivative bet. Anthropic's filing on or around 1 June, followed by OpenAI's on 8 June, established a duopoly-shaped script. Perplexity's announcement effectively says: we are not in the duopoly, and we will not be sequenced behind it.
The counter-narrative inside the labs.
The public-filings story is being told alongside, and partly against, a second narrative now running inside the largest AI companies: that frontier development itself may need to be slowed. On 9 June 2026, OpenAI publicly said the world may need a mechanism to coordinate "slowing frontier development when needed." That language is not new for the company — it echoes language used in safety and governance discussions in 2023 and 2024 — but its reappearance on the morning after an IPO filing sharpens the tension. A company telling regulators and the public that frontier training runs may need to be throttled is also a company asking those same regulators and the same public to value, in dollars, a future in which those runs are throttled.
There is a more sceptical reading. The three companies now publicly committed to public listings have collectively raised tens of billions of dollars in private capital at valuations that have, in some cases, multiplied year-on-year. The same capital that funded the private rounds is now, in many cases, the same capital that wants a liquid exit. The decision to file confidentially — a US regulatory status that allows a company to submit draft registration statements to the Securities and Exchange Commission and keep them out of public view during review — is a legal choice as much as a strategic one. It signals, to staff and early backers, that the management has reached a confidence threshold about audited financials; it signals, to competitors, that the company is willing to be benchmarked against peers under public disclosure rules. It does not, by itself, force a near-term listing; the clock starts when the company chooses to make its S-1 public.
A third reading, common in venture and hedge-fund circles, is that the filings are defensive. With Anthropic and OpenAI now formally on the IPO track, late-stage private capital will demand clearer exit paths from every other frontier-adjacent name. Perplexity's 2028 commitment, in that frame, is less about timing and more about preserving negotiating leverage in private rounds over the next eighteen months.
The structural frame.
What is happening is the conversion of a strategic-asset bubble into a public-markets asset class. The frontier-AI sector was, until this year, valued on the basis of private marks, model-demo narratives, and forward-revenue projections drawn from non-binding enterprise agreements. Public markets price on a different ledger: audited revenue, recognised deferred revenue, customer concentration, gross margin, and a defensible path to operating cash flow. The most consequential shift of 2026 may not be the launch of any single new model, but the moment at which frontier-AI companies are required to clear the disclosure threshold of a listed balance sheet.
For incumbents — Microsoft, Alphabet, Amazon, Meta — the question is whether their strategic stakes in OpenAI, Anthropic, and the broader ecosystem survive the transition. For sovereign and quasi-sovereign investors, the question is whether they will be permitted to participate at the listed level, given the export-control, compute-location, and beneficial-ownership rules that already govern parts of the AI supply chain. For the smaller labs, the question is whether public capital will underwrite the compute and data-center commitments that frontier training requires, or whether the public market will fund them only after the private market has stopped doing so on tolerable terms.
The OpenAI statement on coordinating a slowdown in frontier development sits inside this frame too. A public company cannot make a credible commitment to slow its own training runs without that commitment appearing in risk factors, in the management discussion, and eventually in the prospectus. The market will price the commitment against the competitive cost of honouring it. The labs know that; the question is whether they have decided, individually or collectively, to bear that cost.
Precedent — the 2014 to 2018 social and cloud cycle.
The most useful precedent is not the dotcom era, but the 2011-2014 social and cloud listing cluster. When Facebook filed in 2012, Twitter followed in 2013, and the cloud-software cohort (Box, Shopify, others) priced in 2014-2015, the assumption inside the venture industry was that the public market would learn to value platform-revenue ramp on its own terms. The result was mixed: some listings traded up sharply on debut, several traded down for eighteen months, and the ones that ultimately anchored the cycle (a narrow set of cloud names) were the ones whose unit economics held up under disclosure. The cohort that listed in 2014-2015 but lacked the underlying margin profile either stalled or was absorbed.
The AI cycle will not repeat that pattern exactly. The capital intensity is higher, the customer concentration is more institutional, and the regulatory perimeter is more active. But the lesson reads: order of arrival is not destiny. The 2014 cohort taught the market that the third and fourth listings in a category often clear at lower implied multiples than the first, even when the underlying business is improving, because investor attention and underwriting bandwidth are finite. If that pattern repeats, Perplexity's decision to commit to a 2028 window — and to declare that commitment publicly now — is a way of pre-emptively distinguishing its narrative from those of the two larger labs, so that it is not priced as a residual.
Stakes — who wins, who loses, on what horizon.
The clearest winners, in the near term, are the private backers of the three named companies. Their marks, written at private-round prices, will eventually be testable against public-market quotes; a clean listing window, even at a flat debut, validates the model. Investment banks with capacity to underwrite multi-billion-dollar AI floats are also clear beneficiaries: the duopoly-shaped first wave concentrates fees among a narrow set of lead managers, and a third listing extends the earning window.
The exposed parties are mid-stage AI startups that have not yet secured a public track and that now face a more demanding private-round environment. Their private valuations will be marked against an AI-sector public benchmark, and if that benchmark trades softly in 2027-2028, their funding terms tighten. Compute providers and cloud hyperscalers have a more mixed exposure: they are suppliers to the labs, but the labs' public commitments to capital discipline will, in time, constrain the pace of compute spend growth they can extract.
The policy reader should watch three things. First, the timing of the public S-1s — the confidential filing becomes meaningful only when the document is made public. Second, the disclosure language around model-release cadence, safety commitments, and any "frontier slowdown" mechanism, since that is the language the prospectus will inherit. Third, the participation of sovereign-linked capital, which will be a leading indicator of how the export-control and national-security apparatus around AI will or will not be loosened to accommodate public listing.
What remains uncertain.
The sources confirm the filings, the timetable signals, and the public language around frontier slowdown. They do not confirm valuation ranges, target exchanges, or underwriting lineups. They do not specify what "slowing frontier development when needed" would mean in operational terms, nor who would adjudicate the need. The Reuters wire summarises CNBC's reporting on Perplexity and the BBC and TechCrunch summaries describe the OpenAI and Anthropic filings, but the financial substance — the size of the offerings, the percentages of the companies being sold, the use of proceeds — is not yet on the public record. The most consequential decisions of the next 24 months will be made inside the documents that have not yet been published, and the most accurate reading of this cluster is that the queue is forming, not that it has cleared.
Desk note: Monexus reads the 8-9 June cluster as a sequencing event, not a pricing event. The wire coverage, led by BBC and TechCrunch on the OpenAI filing and by Reuters summarising CNBC on Perplexity, is consistent; the speculation about valuation, exchange, and use of proceeds that will fill the next several weeks of financial press has not yet been confirmed by primary documents. The OpenAI statement on coordinating frontier slowdown has been reported as a position, not yet as a mechanism.
Wire provenance
This editorial synthesis draws on the following public wire/social posts:
- http://reut.rs/441gjms
- https://x.com/polymarket/status/perplexity-2028-ipo
- https://x.com/polymarket/status/openai-frontier-slowdown
- https://en.wikipedia.org/wiki/Confidential_filing
- https://en.wikipedia.org/wiki/OpenAI