Private Markets Update & Deal Flow

It happened. SpaceX priced at USD 135 per share on Thursday June 12, raising USD 75B — the largest IPO in history, surpassing Saudi Aramco's 2019 record by more than 2.5x. The stock opened at USD 150, ran to an intraday high of USD 176.52, and closed day one at USD 160.95 — a 19% pop on its first session, giving the company a market cap of approximately USD 2.1T at close. Gwynne Shotwell rang the opening bell from Starbase, Texas. Audible cheers were heard outside the Nasdaq MarketSite in New York. The book had come in north of USD 250B of demand — 3.5x oversubscribed — driven by long-only funds writing large tickets and Musk himself dropping into investor Zoom calls to close the room.

The S-1 confirmed the business is Starlink first, everything else a distant second. Connectivity generated USD 11.4B in 2025 revenue at a 63% EBITDA margin on 10.3 million subscribers — the only segment actually profitable. The xAI division absorbed a USD 6.36B operating loss last year. Total company revenue was USD 18.7B in 2025, with a net loss of USD 8.7B between January 2025 and March 2026. The valuation is a USD 2T bet on what Starlink becomes, not what exists today. MSCI announced SpaceX would be eligible for early inclusion starting today (June 13), meaning structural index-fund demand kicks in on day two — a tailwind that has already pushed after-hours to USD 166.

The question is no longer whether this IPO worked. It did. The question is what it tells us about what comes next. Anthropic and OpenAI are filing into a market that just confirmed it will pay frontier-technology multiples at scale. That is the most important data point of the week for anyone positioned in pre-IPO AI.

The result: a 19% first-day pop on the largest IPO in history is the market declaring that SpaceX occupies a category of one. The S-1 made clear this is a Starlink story at its core — USD 11.4B in revenue, 63% EBITDA margins, 10.3M subscribers — with everything else still a future option. MSCI inclusion begins today, injecting structural index-fund demand into a stock with only a 4% float. The setup for continued upward pressure is there. More importantly for pre-IPO clients: the market just told you what it thinks frontier-technology conviction is worth. Anthropic and OpenAI file next.

Anthropic is releasing Claude Fable 5 — a new Mythos-class model made broadly available to the public for the first time. Described as twice as fast as Opus, Fable 5 gives general users access to the calibre of reasoning that Anthropic had until now reserved for a small circle of vetted organisations under Project Glasswing. The capability gap between what Mythos could do and what the public could access has been a defining feature of Anthropic's product architecture; that gap is now narrowing deliberately.

The release comes with a structural concession to safety. Anthropic has added strict routing guardrails: if a user prompts Fable 5 on sensitive topics — bioweapons, cybersecurity exploits, software vulnerabilities — the system does not engage through the new model. Those queries are automatically rerouted to the older, more tightly controlled Claude Opus 4.8. The architecture is a careful hedge: release the capability, constrain the attack surface, and monitor where the two populations diverge. Anthropic also confirmed that Claude can now search users' Gmail inboxes natively, the latest move in an ongoing push toward deep enterprise integration.

The commercial backdrop makes the timing clear. Anthropic has a confidential IPO filing in process, run-rate revenue crossing USD 47B, and a USD 965B valuation that the market needs to see justified by product momentum. Fable 5 is part of that justification. The question investors will be asking — whether Mythos-class capability can be meaningfully monetised without the liability exposure that came with restricting it — is now live in production.

Why it matters for pre-IPO investors: every public deployment of Mythos-class capability is a data point on Anthropic's addressable enterprise market. Fable 5's guardrail architecture shows the company is serious about monetising safely — the combination that public markets will need to see before they accept a USD 1T valuation. Our pre-IPO allocation remains open.

Jeff Bezos's physical AI startup has raised USD 12B in a Series B round, valuing Prometheus at approximately USD 41B. Bezos personally invested in the round and confirmed on CNBC — speaking live from the Blue Origin Rocket Park — that Prometheus now takes up the "bulk" of his time. The company is building what it calls an "artificial general engineer": AI tooling that compresses the distance between design intent and manufactured object, operating across aerospace, computing, and automotive engineering. Not a chatbot. Not a robot. Something more consequential than either, if the product story holds.

The team of 120 to 150 people spans San Francisco, London, and Zurich, drawn from OpenAI, Google DeepMind, and Nvidia. Co-CEO Vik Bajaj — Stanford professor and Verily co-founder — leads alongside Bezos, who is described as hands-on in a way he was not at Amazon or Blue Origin in their later stages. The company is reportedly considering a Berkshire Hathaway-style holding structure that would allow it to acquire manufacturers benefiting from its tools — turning a software company into a physical economy play at scale. Amazon and AWS remain a natural downstream partner, though Bezos has insisted the two entities will operate at arm's length.

The conversation was live on CNBC. Bezos also addressed the May 28 New Glenn explosion directly — a rare moment of public acknowledgement from the founder whose company has otherwise stayed tightly controlled on the accident. The juxtaposition was pointed: Blue Origin stumbling on one stage, Prometheus accelerating on another.

The signal: USD 41B is early-stage pricing for a company with no shipped product and everything to prove. What it buys is a founder with unmatched hardware and logistics credibility, a team assembled from the best AI labs on earth, and a thesis — collapsing engineering timelines with AI — that nobody has successfully commercialised at scale yet. If Prometheus delivers, USD 41B will look modest. If it doesn't, the cap table will look like the world's most expensive proof-of-concept.

Mistral AI is in advanced discussions to raise approximately €3B at a €20B valuation — nearly doubling its worth from just nine months ago. The Paris-based startup, founded in 2023 by former Meta and DeepMind researchers, has emerged as Europe's most credible answer to OpenAI and Anthropic. It has crossed USD 400M in annualised revenue, and its roster of enterprise clients — ASML, TotalEnergies, HSBC, and several European governments — gives the business a structural anchor that few AI companies at this stage can claim. Roughly 60% of revenue comes from European clients, which is precisely the point.

The backstory matters. Mistral raised a €1.7B Series C at €11.7B in September 2025, led by ASML — a semiconductor equipment giant whose 11% stake signals something beyond a financial bet. The chip supply chain and the model layer are converging, and ASML's investment is a vote of confidence that Mistral sits at the intersection. Nvidia is also on the cap table. Beyond equity, the company secured €830M in debt financing in March 2026 from a seven-bank consortium to fund its own AI infrastructure buildout — 13,800 Nvidia GPUs at a data centre outside Paris, and a major Swedish facility planned for 2027. Mistral is building the compute stack, not just renting it.

The strategic logic of a €20B raise is straightforward: the window to build a credible European frontier model company is narrow, the geopolitical tailwind is real, and Mistral needs the firepower to stay competitive with labs spending ten times as much. The discussions are reportedly at an early stage and terms could change. But the direction of travel is clear — Europe is serious, the capital is there, and Mistral is the vehicle everyone is backing to prove it.

The investor angle: Mistral is not a consolation prize for investors who missed OpenAI and Anthropic. It is a structurally different bet — on sovereign AI infrastructure, on European enterprise demand, and on the proposition that the EU's regulatory regime is a moat rather than a headwind. At €20B it is still a fraction of its American rivals' valuations. If the European enterprise market adopts AI at anything like the pace of the US market, the gap narrows quickly. The round is not yet closed; we are watching it closely.

OpenAI has filed confidentially for a public listing, with Goldman Sachs and Morgan Stanley in place on a deal that could land as early as this autumn. The official line from Sam Altman's camp is cautiously non-committal — no firm timing, plenty they'd rather do while still private — but the filing buys the option to move fast if the window opens. Translation: they want to be ready, and the SpaceX book this week tells them the window may already be open. ChatGPT has just crossed a billion monthly users, the fastest app to reach that mark in history, faster than Google Maps. The growth story is there.

The complication is that OpenAI chose exactly this moment to declare war on token pricing. Altman has publicly called AI costs a "huge issue" for customers and signalled the company expects to move in lockstep with Anthropic on price reductions. A price war right before two public listings is a simultaneous stress test of both business models. Every dollar off the token price squeezes a margin that was never comfortable to begin with — and enterprise customers who've started doing their own arithmetic are already testing open-source alternatives at a tenth of the price. The real fight is in enterprise: Anthropic's Claude Code has built momentum with software engineers, and OpenAI is countering with Codex while trying to hold its corporate accounts.

The why is simple and enormous. OpenAI has told investors it plans to spend roughly USD 600B on AI infrastructure by 2030. Public markets are the deepest capital pool available. The question is whether it wants that scrutiny while the revenue story is still mid-edit.

The positioning question: OpenAI and Anthropic are heading to public markets simultaneously, from different strategic postures. OpenAI leads on consumer reach; Anthropic leads on enterprise safety narrative and, increasingly, enterprise revenue. The price war is the wild card — it could rationalise the market or it could compress both companies' paths to profitability at the worst possible moment. Watch the enterprise retention numbers, not the headline user counts.

A story that did not get the attention it deserved: the ECB quietly stepped in last year to restrict Revolut's European operations, temporarily limiting its ability to launch new products in the EEA until it addressed what regulators called "deficiencies" in its product approval process. The ECB also ordered a third-party review of Revolut's risk, compliance, and legal functions. The restrictions extended outside Europe — Revolut's European arm was reportedly prevented from making acquisitions or onboarding new customers in certain markets — during the same period the company was preparing a share sale and publicly projecting long-term ambitions toward a USD 200B valuation.

The root issue is structural: CEO Nik Storonsky has described Revolut employees as "self-guided missiles" — expected to move fast with minimal supervision. That culture drove Revolut to 75 million customers and the title of Europe's most valuable fintech, but it created exactly the kind of tension that puts regulators on the phone. The company says it has strengthened its internal review processes since then and remains in "regular and constructive dialogue" with regulators. It is unclear whether all ECB restrictions have been fully lifted.

None of this has visibly slowed the growth story. Revolut is now reportedly valued at USD 115B in a new secondary share transaction, with EQT, BlackRock, and Schroders Capital among participants — which would make it larger than several major listed European banks. Pre-tax profits rose 57% to GBP 1.7B on GBP 4.5B of revenue. The company has since obtained a banking licence in Mexico, filed for a US banking charter, and launched mortgages, teen accounts, and branch locations in Europe.

The investor read: Revolut's regulatory history is not a disqualifying fact — it's a pricing input. The company survived ECB restrictions that would have stopped a less operationally robust business, emerged with its growth intact, and is now valued above most listed European banks. The path to a potential IPO at USD 200B+ depends on the US banking charter landing and the ECB relationship normalising fully. Both are watchable, neither is fatal.

Perplexity CEO Aravind Srinivas told CNBC this week that the company is targeting a public listing in 2028 — and that this timeline holds regardless of how the upcoming Anthropic and OpenAI IPOs perform. "Agnostic of these two companies, we were planning for something in 2028 so that still remains the case." He did not sugarcoat the risk: poor performance from either deal could create ripple effects across the AI IPO cohort. He flagged the SpaceX listing as the leading indicator for how Anthropic and OpenAI will be received, and his warning to the frontier model makers was blunt — if either goes six months without a meaningful capability advance, the valuation multiples become a liability rather than an asset. Perplexity's own commercial trajectory, with annualised revenue tripling in 2026, has bought Srinivas the credibility to be this direct.

Perplexity is building toward a public listing on a thesis that the value in AI accretes to the orchestration layer, not the model. With revenue tripling and a CEO willing to call out competitors directly, it is positioning itself as the measured, infrastructure play while the frontier labs race to market. 2028 gives it time to be right.

Kalshi — the regulated prediction market platform — is introducing the first formal insider-trading guardrail in the prediction market space. Following recommendations from its advisory committee, the company will require users on selected markets to disclose their employer before placing trades. The measure is aimed at participants who may hold material nonpublic information: company employees, political insiders, regulatory staff. The disclosure system rolls out in coming weeks. As prediction markets mature and handle increasingly significant financial and political questions, the regulatory infrastructure around them is catching up. Kalshi's move is a marker of that transition — and a signal that the asset class is being taken seriously enough to attract the kind of oversight that previously applied only to listed securities.

Kalshi is maturing prediction markets into something closer to a regulated financial product. The employer disclosure requirement is the kind of institutional infrastructure that separates a novelty from an asset class. When prediction markets attract the same oversight framework as listed securities, it is a signal that they are being taken seriously as a price discovery mechanism — not just a betting platform.

Vinted's marketplace CEO Adam Jay made a pointed claim this week: secondhand shopping is no longer a temporary habit driven by inflation or cost-of-living pressure. Consumers are forming lasting routines around buying and selling used items because it is cheaper, simpler, and more sustainable — and the numbers back him up. In 2025, Vinted's GMV rose 47% to EUR 10.8B, revenue increased 38% to EUR 1.1B, and the platform was recently valued at more than USD 9B following a EUR 880M secondary share transaction involving EQT, BlackRock, and Schroders Capital.

The company remains predominantly European — strongest in France and the UK — but has clear ambitions to expand into the US and move beyond fashion into electronics and furniture. High shipping costs are the primary friction point in the American market. Meanwhile, Vinted is investing heavily in its own logistics and payments infrastructure through Vinted Go and Vinted Pay. Those investments are not free: despite strong revenue growth, net profit fell 19% in 2025. The platform is scaling infrastructure faster than it is monetising it — a familiar trade-off for marketplaces targeting global reach.

There is active market speculation about a future IPO, but Vinted is in no rush. The company is cash-positive, can still access private capital, and says it is focused on the long-term goal of making secondhand shopping a "first choice" worldwide — not a discount fallback. That positioning, if it holds at scale in the US, makes the EUR 880M secondary look like early access to something considerably larger.

The investor read: Vinted is a rare consumer platform that combines genuine unit economics (cash-positive, profitable before infrastructure investment) with a structural tailwind — the secular shift toward reuse that no recession created and no recovery will reverse. The US expansion is the key variable. If shipping costs can be solved, the addressable market is multiples of what Europe alone provides. The EUR 880M secondary at USD 9B may look very different in three years.