On April 23, 2026, Forge Global's CEO Kelly Rodriques confirmed to multiple outlets that Anthropic shares were clearing prices on his platform that imply a $1 trillion company. Three months earlier, Anthropic closed a Series G that priced the same shares at $380 billion. Nothing has changed at the company in that time except the queue of people who would like to own a piece of it.

That is the story. Not the trillion. The 2.6× gap between what Forge buyers are paying and what the last actual fundraise priced in.

What the number actually means

Forge Global is one of the larger secondary marketplaces for private-company shares. It is where employees and early investors who already hold stock can sell to outside buyers without waiting for an IPO. The price on Forge is set by whoever is willing to pay the most for a small slice of someone else's holding — illiquid, with no board rights, no ability to force a sale, and no guarantee that any future IPO prices anywhere near where you bought.

So a $1 trillion print on Forge is not what Anthropic is "worth." It is what one buyer paid, on one platform, for one packet of shares, on one day. It is closer in spirit to a single Christie's hammer price than to a market cap.

That said, the number is still doing real work, because three things are simultaneously true:

• Anthropic's last primary round, reported across the trade press at $380 billion post-money in February 2026, was itself a record.
• Anthropic's annualised revenue went from roughly $9 billion at the end of 2025 to $30 billion in March 2026 — a 233% jump in a single quarter, according to figures cited by Decrypt and corroborated in the Yahoo Finance write-up.
• OpenAI is currently trading on the same Forge platform at around $880 billion, below the $852 billion it raised at earlier in the year, because in Q1 it had more sellers than buyers.

So the secondary market is paying Anthropic a 163% premium over its primary round, and OpenAI a 3% premium over its. Two of the three frontier US labs trade in the same forum, on the same day, against the same pool of capital, and the spread between them is enormous.

Why anyone pays the premium

Two reasons keep getting cited, and they are different reasons.

The boring one: revenue is real and growing. A 233% sequential jump in annualised run rate, if the figures hold, is the kind of curve that justifies extreme multiples on a forward-looking basis. If you believe Claude's enterprise pull and the Claude Code franchise will keep compounding, $30 billion in March becomes plausibly $50–60 billion by year end, and the trillion starts to look like a 17–20× revenue multiple — high, but not unhinged for a hyperscaling category leader.

The less boring one: there are almost no shares to buy. Anthropic employees and early investors have had very few liquidity windows. When buyers pile in and almost nobody is selling, prices move violently. Glen Anderson, the CEO of Rainmaker Securities, told Entrepreneur that "it's almost less about the return than being able to say they're an Anthropic investor". He also told Decrypt that a $960 billion bid which would have been "unthinkable a month earlier" was being snapped up within hours by competing buyers.

That is not a market in which prices reflect cash flows. That is a market in which prices reflect access. Which, fine — it is also a market that has cleared at a trillion. Caplight, which tracks private-market interest, says investor interest in Anthropic shares spiked 650% over the past twelve months. One shareholder reportedly tried to list at $1.15 trillion. On X, Jesse Leimgruber posted that he would buy at a $1.05T mark. None of that should be confused with valuation in the textbook sense.

What the spread tells you about OpenAI

The OpenAI line in this story is the quiet one, but it is the more diagnostic one.

OpenAI raised its most recent primary round at $852 billion. On the same secondary marketplace where Anthropic is clearing 2.6× its last primary, OpenAI is clearing roughly flat to slightly up on its. Decrypt reports that in Q1 2026, OpenAI on Forge had more sellers than buyers — meaning insiders who had access to the market were, on net, taking chips off the table.

That is not the same as bad news for OpenAI. Insiders sell for a hundred reasons that have nothing to do with conviction in the company. But it does tell you something about the part of the trade that is FOMO. When insiders are net sellers and the price is barely budging above primary, the secondary mark is mostly mechanical. When insiders are scarce and outsiders are climbing over each other, the secondary mark is mostly emotional.

Anthropic right now is the second. OpenAI right now is the first.

The IPO number nobody quotes

The interesting hidden figure in the Decrypt coverage is this: bankers reportedly model an Anthropic IPO at $400–500 billion. That is the number a public market — not a private secondary — would pay for the same business.

Pause on the spread. The same shares of the same company are simultaneously priced at:

• $380 billion in the most recent primary round
• $400–500 billion in IPO modeling
• $1 trillion on Forge Global secondary

You can pick any pair of those and they disagree by a factor of 2 to 3. There is no single "Anthropic valuation" right now. There is a primary-market valuation, an IPO-readiness valuation, and a secondary-market valuation, and they are not the same activity.

The honest read is that the trillion is the noisiest of the three and the IPO model is probably the cleanest. The headline writers prefer the loudest number, which is how we ended up here.

How worried should you actually be

The "is this a bubble" question is the wrong one. The right question is: which of the three numbers ages well?

If Anthropic's revenue keeps doubling each quarter, the primary round looks like the bargain of the decade and the trillion looks merely aggressive. If revenue growth slows to a more normal 30–50% annual cadence, the primary round looks fairly priced and the trillion looks like the kind of number people will be embarrassed to have written about. The current revenue print is plausibly compatible with the first scenario — Claude Code adoption, enterprise traction, and the Amazon Trainium expansion all give the curve room to keep compounding. It is plausibly compatible with the second too. We will know in two quarters.

The thing to actually worry about is concentration. Anthropic, OpenAI, and a handful of CoreWeave/Lambda/Together-class providers absorb a wildly disproportionate share of the late-stage AI capital flowing through secondary venues. When the top of the table moves on a single Forge print, that is the market telling you it has nowhere else to put the money.

That is bubble-shaped behaviour. It is not yet a bubble. Bubbles are a retroactive diagnosis.

What to watch

1. Whether Anthropic prices a primary round inside the trillion this year. That is the only test that converts a Forge mark into a real one. If a primary round in late 2026 prices at $700 billion–$1 trillion with serious institutional anchors, the secondary was right. If the next primary prices at $500 billion, the secondary was a mood.
2. Whether OpenAI's secondary catches up. If Anthropic stays at $1T and OpenAI drifts back toward primary, the market is making a directional bet on Anthropic specifically, not a generic AI bet. That has implications for everything from cloud commitments to chip allocation.
3. Whether the revenue print is repeated. $9B → $30B annualised in one quarter is the single load-bearing fact in this story. If Q2 prints another step like that, the trillion is defensible. If Q2 prints flat, the trillion is the part of the article that ages worst.
4. Whether anyone outside Forge confirms the clearing prices. Forge is one venue. EquityZen, Caplight, Hiive, and the bilateral broker market all see the same shares, often at different prices on the same day. If the trillion holds across venues, it is real demand. If it is a Forge-only phenomenon, it is an order-book artefact.

The temptation in stories like this is to pick a side: bubble vs. justified, hype vs. fundamentals. The honest answer is that the trillion is partially earned and partially a forced print on a thin venue, and the gap between the two halves is exactly as wide as the gap between primary and secondary. Treat the headline as roughly half real, and you will be calibrated about right.