#Gate直通IPO第三期Anthropic

Founders Cash Out While Value Drops $600 Billion: Is This IPO Still a Good Deal?

Anthropic is going public, but the start is rough. On Nov 10, the roadshow opened with 825 million shares at $70 to $78, for up to $64.4 billion. The night before, the draft still showed $69.3 billion. About $5 billion vanished overnight.

Value fell from $2 trillion talked about two weeks ago to about $1.4 trillion, down about 30%. That is below SpaceX at about $1.77 trillion at its IPO.

At the same time, 7 founders will sell old shares for about $1.75 billion in total, about $250 million each.

Scale is huge, but not a record

SpaceX raised $75 billion in June, about $85.7 billion with the extra share option. Anthropic tops at $64.4 billion, about $74 billion with the option. Still the second largest IPO ever.

Private placement adds $26 billion at the IPO price. Google and Nvidia put $10 billion each. AMD, Broadcom, Micron, Microsoft and TSMC split the rest $6 billion. Together with the IPO, total is near $100 billion.

But all seven are upstream suppliers in the compute chain. Suppliers pay to own a slice of a buyer, and the buyer then pays them for compute. Money looks bigger, but it may just spin inside the chain. If end demand lags, both sides feel it.

Revenue is real, but costs bite

Revenue is real. $386 million in 2024, $4.585 billion in 2025, about $17 billion in Q3 alone. Yearly run rate hit about $81 billion by end of October. Same buyers spend about 9 times more a year later. Net revenue retention is about 900%. Cash flow from work was about $3.9 billion in the first half.

But growth is cooling. It was up 144% quarter on quarter in Q2, then 47% in Q3. Profit turned to loss. Q2 had $974 million in operating profit, Q3 had a $567 million loss. Compute cost rose from 48% of revenue to 58%. The firm had to buy about $5 billion in short-term compute at high prices. Price cuts and discounts cut Q3 revenue by about $600 million.

Another soft spot: in the first nine months, 5% to 7% of revenue, about $1.7 to $2.3 billion, came from buyers later found to have broken usage or business terms. Most came from users outside the US whose accounts were closed. Closing accounts with no refund has long been a habit. That it shows up as a steady income line is hard to read as a plus.

Paper loss is the odd part. Net loss on paper hit $104 billion in nine months. Real operating loss was only about $1.26 billion. The rest came from revaluing early funding where backers pay now and get shares later. As value rose from over $100 billion to over $1 trillion in just over a year, those future shares rose in value too, and the rise had to be booked as a loss. It is an accounting effect, not a cash burn, but it still clouds the books.

Commitments are heavy. Near $900 billion in long-term compute buy and lease deals with Google, Amazon, Broadcom and others. Cash and short-term holdings are about $132 billion. Top five buyers gave about a quarter of revenue in nine months. Over 80% of Q3 revenue came from models out in the past six months. Anthropic must keep outrunning all rivals. One gap between models could cost it buyers fast.

Price setup

825 million Class A shares. About 801 million are new. At the $74 midpoint, that part brings about $59.3 billion to the firm. The rest, 23.648 million, are old shares for the 7 founders.

There is a 30-day price shield. Underwriters can buy up to 123.75 million extra shares at the offer price if needed. If price falls, they buy to hold it up. If it rises, the firm can issue more shares. Shield lasts only 30 days. SpaceX had the same shield, yet its stock fell below offer price in mid-July.

Demand is strong. Orders filled the range at once, and big buyers showed more hunger than for SpaceX. But mood is weak. On Nov 5, US stocks fell 5.5% in one day, the worst in 19 months, and AI chip stocks fell hard. Listing into a sell-off is poor timing.

First-day selling is also a risk. Some directors, execs and staff can sell up to 10% from day one. Many hold shares from when value was very low, so selling would lock in a big gain. If many sell at once, price drops.

Price already bakes in a lot. Banks priced it off Anthropic's own 2028 revenue forecast of $190 to $200 billion. Yearly run rate is still about $81 billion. Revenue must more than double in two years for today's price to hold.

Model news

Fable 5.2 came a day before the roadshow, just two months after 5.1. It runs up to 30% faster, with a fast mode 2.5 times faster still. It costs $8 per million input tokens and $40 per million output tokens, about 20% cheaper than last gen. It scored 65 on Artificial Analysis intelligence index, first, ahead of Google Gemini 4 Carbon at 60. In FrontierCode 2.0, it scored 49.1%, up from 32.4%, and ahead of OpenAI GPT-6 Astra.

Yet Anthropic holds it back on purpose. In training, it capped Fable 5.2's skill at speeding up AI research, like chip kernel tuning and model design, at Opus 5.5 level. Racing hard on models while holding back its best model is the firm's biggest split.

Why now, when OpenAI waits?

Three firms at trillion-dollar scale lined up: SpaceX, Anthropic, OpenAI. SpaceX went in June. Anthropic follows. OpenAI stays private. On Nov 6, OpenAI closed $11 billion at $1.25 trillion value.

OpenAI filed in secret in June, a week after Anthropic. In Sept, Altman said no IPO in 2026, citing AI safety. Advisers had given two paths: wait to 2027 for a trillion-dollar listing, or cut value and go now. Altman turned down the second. Its finance chief also urged delay, saying disclosures were not yet at public-firm standard. In private, value is set by talks. Once listed, you are graded each day.

Anthropic has three reasons to go now. One, spend is huge. Near $900 billion in compute deals, against about $132 billion in cash and short-term holdings. Private rounds cannot keep up. This deal brings near $100 billion at once. Two, books look ready. Revenue is strong, and Q2 even saw operating profit. OpenAI faces more pressure, with about $278 billion in cash burn seen from 2026 to 2030. Three, first-mover edge. Much cash around the world waits for a pure AI stock. The three firms together top $4 trillion in value, but cash in the market is finite. First to list gets it first.

Verdict: yes, but with care

Yes, because the business is real. Revenue grows, buyers spend more, product scores back it up. But price already holds much of the future. Value is down from $2 trillion to $1.4 trillion, yet still high. Revenue must more than double in two years. Compute cost, circular money flow, buyer focus and $900 billion in deals make that hard. Founder sales of $250 million each also weaken the we-believe message. The sale itself is not a crime, but its timing raises a query.

So look with discipline, not hype. Keep size measured. Short term is mood, long term is results. First-day pop is not proof. What price does after the 30-day shield ends is the real test. Anthropic may be a century IPO. That does not make it a good deal at any price. A good deal needs right price, right size, right horizon. Without those three, even the strongest story will not shield a buyer.

#WCTCS9
#ShareWeekly
$SPCX