I kept staring at the number the way you stare at a restaurant bill that cannot possibly be yours. Two trillion dollars. Not for a chipmaker already printing historic profits. For a company whose last reported year, if the leaked figures hold, produced a few billion in revenue and a loss large enough to swallow several mature software firms whole. That is the strange place public markets may be walking into if the Anthropic IPO lands anywhere near the valuation being whispered about.
An independent research shop, the sort that builds a career on spoiling the party, put it in blunt language this week. They called the coming listing the most ridiculous IPO of 2026. They also pegged a sober value closer to $150 billion and warned that Wall Street is about to sit an exam in investor gullibility. I do not treat every bear note as gospel. I have watched enough of these calls age badly. Still, the gap between the private dream and the public arithmetic is wide enough that ignoring it would be lazy.
Why The Anthropic IPO Has Become A Market Stress Test
Listings are supposed to do two jobs at once. They raise permanent capital, and they let early backers cash a portion of the bet. When those jobs stay in balance, ordinary buyers can live with a rich multiple. When the second job swallows the first, the prospectus starts to read like an exit ramp with better lighting. That is the charge being leveled here, and it is worth taking apart without the usual cheerleading.
Anthropic, led by Dario Amodei, has become one of the names people reach for when they talk about frontier artificial intelligence. The models are serious. The enterprise contracts are real. The brand sits in a small club with a handful of closed-model labs. None of that automatically justifies a market value that would place the firm among the most expensive companies on earth before it has shown it can keep the lights on without fresh outside money.
Perhaps the most interesting aspect is how quickly the conversation jumped from product quality to ticker fantasy. Product quality matters. So does the bill for training runs, inference, talent, and the legal weather around training data. A great model with a broken unit economic story is still a great model. It is not automatically a great stock.
The Number That Refuses To Sit Still
Reported coverage, citing a leaked prospectus, put 2025 revenue near $4.6 billion and the net loss near $42 billion. Sit with that pair for a second. Revenue is not trivial. A loss of that size is not a rounding error either. It implies a business still spending far ahead of what customers are paying, which can be rational in a land grab and reckless if the land never pays rent.
The company has also talked, in public remarks, about a sharp jump in annualized revenue run rate. By late July the figure being cited was about $65 billion, up roughly sevenfold from a year earlier. Separate reporting suggested a path toward $100 billion in annualized revenue by the end of 2026. Run rate is not the same thing as collected, audited, full-year revenue. I have found that investors blur those terms right when the marketing is loudest.
A run rate is a speedometer reading. It is not the distance already driven, and it is definitely not the fuel left in the tank.
Market observer, paraphrased from common listing-season caution
If the higher run-rate figures prove durable, the bearish snapshot from 2025 will look dated fast. If they prove promotional, the people who bought the story at a two-trillion tag will be the ones explaining the chart. Both outcomes are still live. That is why this is a debate, not a verdict.
What A Two Trillion Tag Quietly Demands
The research note made a comparison that sticks. To deserve the valuation being sought, Anthropic would need something on the order of double the trailing profit Nvidia has been posting. Nvidia’s net income over the past four quarters has been described as topping $190 billion. Double that, and you are asking a loss-making lab to eventually earn like a company that already owns the picks and shovels of the boom.
Is that impossible? No. Is it the base case a careful buyer should underwrite on day one? I doubt it. Nvidia sells scarce hardware into a frenzy with margins public investors can already audit. A model lab sells access, seats, and API calls into a market where substitutes keep arriving, some of them free to inspect. Those are different machines.
You can love the technology and still hate the entry price. Those two feelings are allowed to share a portfolio meeting.
A Research Firm With A Long Memory For Hype
David Trainer, founder of New Constructs, has made a habit of tagging certain offerings as the most ridiculous IPO of their year. The label is theatrical. The track record is mixed, which is exactly why it is useful rather than sacred.
In 2019 the firm aimed the phrase at WeWork, then privately valued around $47 billion. Within about six weeks the offering was pulled. Weak demand met a financial story that did not survive daylight. The company later filed for bankruptcy in 2023. That call aged well, and it still gets cited whenever a private valuation looks decorative.
The 2021 shoe brand Allbirds was another bearish target. It opened on the Nasdaq at a valuation near $4.1 billion. Earlier this year its assets were reported sold for an estimated $39 million, with the remaining story pivoting toward artificial intelligence. A brutal round trip, whatever you think of the original product.
Then there is the miss. DoorDash wore the same ridiculous tag in 2020 and was compared, unflatteringly, with WeWork. The stock jumped on day one to a market value above $60 billion and later sat nearer $83 billion. Trainer said in a later interview that crazy stuff happens and that he cannot let the misses rewrite the method. Fair. A process that is sometimes early and sometimes wrong can still be a useful irritant.
In my experience, the value of these notes is not the headline adjective. It is the forced question they ask: what has to be true for this price to be sane? If you cannot answer that in plain sentences, you are renting someone else’s optimism.
WeWork Was A Warning, Not A Template
The research firm was careful on one point, and I think they were right to be. Anthropic offers more to society than a flex-office brand ever did. Better tools for coding, research, customer operations, and analysis are not beanbags and free beer. Confusing social usefulness with shareholder usefulness is how people overpay with a clean conscience.
Their sharper line was about scale of damage. A failed office IPO bruised a corner of the market. A listing pitched near $2 trillion, if it were to reprice violently, would tug on indexes, on funds that must own size, and on the story retail investors have been sold about artificial intelligence as a one-way trade. Bigger stage, bigger spill if the floor gives way.
Useful technology and a fair stock price are not the same purchase. You can admire the first and still refuse the second.
The firm also argued that the offering’s real purpose looks more like liquidity for Wall Street backers than wealth creation for the public float. That claim is an interpretation, not a document I have read. It is still the right question to ask of any late-stage listing: who is the customer for these shares, the company or the cap table?
The Prospectus Most Of Us Have Not Seen
Here is the awkward part. Anthropic has not made a full prospectus public in the ordinary way. The research note itself admitted it had not been among the few shown a selective S-1. So the critique rests on leaked figures and press reporting, not on a line-by-line filing anyone can download. That cuts both ways.
Bears can overfit a leak. Bulls can hide behind the absence of the real document. Neither side gets to pretend the missing pages are automatically friendly. When a filing finally lands, the items that matter will be dull on purpose: revenue quality, gross margin after inference cost, customer concentration, remaining performance obligations, related-party compute deals, stock-based pay, and how much cash the business actually burns once the growth slogans are stripped out.
- Revenue that renews versus revenue that was pulled forward to decorate a quarter
- Gross margin after the true cost of serving each query, not before it
- How many customers would hurt if they left, and how fast they could leave
- Compute commitments that behave like debt even when the footnote calls them something softer
- Dilution from stock pay, warrants, and any late private rounds
Until those lines exist in public, every valuation argument is a sketch. Sketches can still be honest. They just should not be priced like blueprints.
Open Source Is The Competitor That Does Not Need A Roadshow
The note’s central business claim is harsh: since open-source models arrived in force, closed models have had a harder time turning usage into durable profit. I would not sign that sentence in permanent ink. Closed labs still hold advantages in reliability, safety tooling, enterprise support, and the sheer polish of a managed product. Plenty of companies will pay to avoid running their own stack.
But the pressure is real, and it is not theoretical. When a capable model can be inspected, modified, and hosted by someone else, the closed vendor has to justify a toll. That toll can be convenience. It can be indemnity. It can be a feature lead that lasts a season. It is rarely a permanent moat just because the brand is famous.
I’ve found that software investors underestimate how fast a “good enough” substitute resets pricing. It happened in databases, in productivity suites, in parts of cloud infrastructure. Artificial intelligence will not be immune just because the demos look like magic. Magic with a falling price is still a commodity in waiting if you cannot own distribution or a scarce input.
Anthropic’s edge, if it holds, will be less about secrecy and more about trust, evals, and workflow lock-in. Those are earnable. They are also copyable. A $2 trillion story needs the first outcome and quietly assumes the second will not arrive on schedule.
Losses Can Be A Strategy Until They Become A Habit
Spending ahead of revenue is not a moral failure. Amazon did it. Plenty of software firms did it. The test is whether each extra dollar of loss buys a position that later throws off cash at a rate the entry price can survive. A $42 billion net loss against $4.6 billion of revenue is an extreme version of that bet. Even if 2026 run-rate talk is directionally right, the market is being asked to believe the cost curve bends before the multiple does.
Training and inference are not like old software, where the marginal copy was nearly free. Every serious answer has a power bill, a chip reservation, and a team that can leave for a rival on a Friday. If prices per token fall faster than volume rises, growth can look spectacular while cash stays ugly. That is the trap hidden inside “usage is exploding.”
A simple listing filter I keep on a notepad: Revenue quality first Cost of serving second Dilution third Narrative last If the order flips, the price is usually doing the selling.
There is also the talent auction. Frontier labs bid against each other, and against the richest platforms, for a small pool of researchers. Stock-based pay flatters cash burn and punishes future shareholders. Public investors often meet that bill only after the insiders have already been topped up.
The Safety Argument Cuts In Two Directions
Anthropic has been unusually direct about the chance that advanced systems could pose catastrophic or even existential risk. The research note treated that stance as another reason to skip the IPO. I think that reading is incomplete, though not empty.
On one side, a company that takes downside seriously may build controls enterprises actually want. Regulation, if it arrives in a serious form, could favor labs that already speak that language. On the other side, a business that publicly frames its own product as a possible civilizational hazard is asking investors to underwrite both the upside of adoption and the policy risk of restriction. You do not get to cash the sermon and ignore the footnote.
Public shareholders are not a safety board. They are a claim on residual cash. If governance, export rules, or liability law cap the very uses that justify a mega-multiple, the multiple was a story about permission as much as about product. Permission can be revoked.
Who Actually Gets Paid If The Bell Rings
Late private rounds in artificial intelligence have been enormous. Strategic investors, venture funds, and cloud partners have reasons to want a public mark that validates the last check. A listing near the top of the whispered range would do that job beautifully for anyone who bought earlier and can sell into strength. It would do a poorer job for the buyer who shows up on the opening print because a headline said the future had a ticker.
This is not a conspiracy. It is how venture math works. Funds have lives. Employees have options. Partners have strategic goals that are not identical to your retirement account. Liquidity is a legitimate purpose of a public market. It is a bad purpose if it is the only one, and if the price required to deliver it assumes profits that have not been earned.
The firm said the IPO is positioned to be a far bigger rip-off of U.S. capital markets than WeWork, precisely because the number is so much larger. “Rip-off” is a loaded word. I would rather say mispricing risk. Mispricing does not require bad faith. It only requires a crowd that would rather not do the division.
A Plain Comparison Of The Stories On Offer
It helps to put the competing pictures side by side, without pretending either one is audited. The table below is a reading aid, not a model.
| Claim | Bullish reading | Skeptical reading |
| Valuation target | Scarcity of frontier labs justifies a premium seat | $2 trillion assumes profits not yet visible |
| 2025 snapshot | Early scale, investment phase, look through the loss | $4.6 billion revenue, about $42 billion net loss |
| Run rate talk | $65 billion, maybe $100 billion annualized, proves demand | Run rate is not profit, and not always repeatable |
| Competition | Enterprise trust keeps closed models priced | Open models cap pricing power |
| Social value | Tools that raise productivity deserve capital | Usefulness is not a multiple |
| Past bear calls | DoorDash shows the label can be wrong | WeWork and Allbirds show hype can break |
Notice what is missing. There is no column for “the chart will go up because everyone is excited.” That column has funded a lot of opening days. It has also funded a lot of quiet second years.
What Nvidia’s Profits Actually Illustrate
Using Nvidia as the yardstick is provocative, and a little unfair, which is why it works as rhetoric. The chip company sits at a bottleneck. Customers order years ahead. Margins are already in the financial statements. Anthropic, by contrast, is closer to the application and model layer, where the bottleneck can move. If compute stays scarce, model labs pay up. If compute loosens, or if rivals match quality, the tollbooth shifts.
A buyer of the IPO is therefore underwriting at least three bets at once. Demand for paid model access keeps compounding. Anthropic keeps a quality or trust lead long enough to charge for it. Costs fall faster than prices. Miss one leg and the two-trillion frame creaks. Miss two and the independent firm’s $150 billion sketch starts to look less like a stunt.
Could the firm be worth more than $150 billion if execution is clean? Sure. Private markets have already argued for far higher marks. Public markets have a habit of arguing back once the lockups and the quarterly calls begin. The argument is the product. The shares are just the ticket.
Index Funds Will Not Save A Bad Entry
A Nasdaq listing of this size would not stay a niche toy. Benchmarks notice mass. Funds that track size would own a slice whether the portfolio manager likes the gross margin or not. That is fine for the market’s plumbing. It is a weak reason for an individual to chase the open.
Passive ownership can support a price. It does not repair a cash-flow hole. If the story later disappoints, the same mechanical buying that felt like a floor becomes a crowd trying to exit through a door sized for normal days. I have watched that movie in other growth waves. The credits are always longer than the trailer.
- Separate the product you admire from the security you can price
- Translate run rate into a full-year guess, then haircut it
- Ask what gross margin looks like after serving costs, not before
- Count dilution as a real expense, because it is one
- Decide in advance what evidence would make you wrong
That last step is the one most listing-season commentary skips. If nothing could change your mind, you are not analyzing. You are cheering.
A Closer Look At The Gullibility Charge
“Unprecedented test of investor gullibility” is a phrase built for a headline. It also smuggles in a sneer. Public investors are not a single gullible blob. Some will fade the open. Some will trade the volatility and leave. Some will buy because their mandate says own the new giant. Calling all of that gullibility flattens a market into a morality play.
Still, the sneer lands somewhere. There is a pattern in hot listings where the questions that matter get postponed until after the first green day. People ask whether they “missed it” before they ask what they would be missing. That reversal is how rich prices become normal prices in group chats.
Perhaps I am too plain about this. Excitement is not evidence. A conference keynote, however polished, is not a cash flow statement. Amodei speaking on a big stage tells you the company can hold a room. It does not tell you the room’s owners should pay any particular multiple.
How Prior “Ridiculous” Calls Should Temper Both Sides
If you only remember WeWork, you will treat every skeptical note as prophecy. If you only remember DoorDash, you will treat every skeptical note as a fade to buy. Both memories are incomplete, and both are available to whoever wants a simple script.
WeWork failed a basic smell test: the unit economics and the governance did not match the private mark. Allbirds discovered that a pleasant brand is not a moat once growth cooled. DoorDash kept a network that customers and restaurants still needed, and the stock found a life after the joke headlines faded. The lesson is not “bears are wise” or “bears are fools.” The lesson is that ridicule is not a model. Cash conversion is.
Anthropic is closer to DoorDash than to a shoe brand in one respect: the product is already woven into workflows for a lot of paying teams. It is closer to the WeWork caution in another respect: the private mark and the public proof are not yet speaking the same language. Holding both thoughts at once is uncomfortable. It is also the job.
What Would Make The Bull Case Boring And Therefore Strong
I would take the optimistic case more seriously if it got duller. Not a vision of universal intelligence. A few years of reports where revenue growth stays high while the loss rate shrinks in a way you can reconcile to the cash flow statement. Customer cohorts that expand without heroic discounts. A gross margin that survives price cuts from open alternatives. Guidance that is missed small, not waved away as “investment.”
Boring excellence is how mega-caps actually get earned. The companies people now treat as inevitable spent years being merely good at the unfashionable parts: billing, support, supply, and saying no to projects that did not pay. Frontier labs are allowed a wilder early chapter. They are not allowed to skip the chapter where the numbers agree with the speech.
Fair entry, roughly: durable revenue x credible margin x years of lead - dilution - policy haircut. If any term is a shrug, the product is not the price.That snippet is not a formula a banker would stamp. It is a reminder that every missing term gets filled by hope, and hope is expensive at this altitude.
Enterprise Demand Is Real, And Still Not A Blank Check
It would be dishonest to write this as if nobody is paying. Teams are buying seats, API access, and bundled tools. Coding assistants alone have changed how some shops ship. Support desks, research groups, and internal search projects are not imaginary. A sevenfold jump in run rate, if the company’s own late-July claim holds, is the sort of growth software investors used to dream about in quieter cycles.
The catch is concentration and switching. If a handful of large contracts, or a handful of cloud distribution deals, carry the run rate, the story is narrower than the headline. If customers can move workloads to a cheaper model with a weekend of engineering, retention is a quarterly event, not a marriage. I want to see net revenue retention the way a lender wants to see collateral, not the way a fan wants to see a trailer.
There is also the quiet competition from the platforms that already own the customer relationship. A lab can be best in class and still get margin-squeezed by the storefront. Distribution has humbled better technologists than this cycle will admit.
Policy, Lawsuits, And The Cost Of Being Early
Any honest pass at this listing has to leave room for rules that do not exist yet. Training-data disputes, copyright claims, safety mandates, and export controls can all change the cost of being a frontier lab. Some of those outcomes would entrench the well-capitalized players. Some would cap the uses that currently juice the revenue forecast. Pricing a stock as if the friendly version is the only version is how people discover “regulatory overhang” after they already own it.
The company’s own language about catastrophic risk makes this harder to hand-wave. You cannot market caution to policymakers and market inevitability to buyers without someone eventually asking which speech was the real one. Maybe both are sincere. Markets still have to pick a discount rate.
How A Careful Buyer Might Actually Behave
None of this is a recommendation to short a story that can gap on a single contract announcement, or to buy it because the logo is everywhere. It is a case for sequence. Read the filing when it exists. Compare the leaked loss picture with the audited one. Watch the first two reports for the gap between run rate and recognized revenue. Notice whether management talks about customers or about destiny.
Position size is the part amateurs skip. A company can be important to the economy and still be a small line in a portfolio if the price already assumes perfection. Importance is not allocation. I would rather own a smaller claim on a fair price than a heroic claim on a story that needs every quarter to go right.
There is also time. Missing the first week is not the same as missing the business. DoorDash’s life as a public company was not decided on the opening print. WeWork’s fate was not saved by the private mark. The second and third year are where these arguments get settled, usually in quieter type.
The Liquidity Story Versus The Ownership Story
Strip the adjectives away and two narratives remain. Narrative one: a rare lab is about to let the public own a piece of a technology shift, and the price looks high because the shift is high. Narrative two: a cap table that has been marked up in private is looking for a public bid large enough to ratify those marks, and the operating losses are being treated as a temporary costume.
Both can contain truth. Early owners deserve a path to liquidity if the company is ready for public scrutiny. Public buyers deserve a price that does not require them to out-earn Nvidia just to break even on the dream. When those deserts conflict, the listing still happens. It just happens at someone’s expense. The research note thinks it knows whose. I think we will know only after a few boring quarters, which is an unsatisfying answer and also the accurate one.
The most expensive words in a hot market are not “I missed it.” They are “it has to grow into the valuation.”
Growth into a valuation is possible. It is also the phrase people use when they do not want to say they paid for years that have not happened yet. At $150 billion, the independent firm is saying the years required are already too many. At $2 trillion, the years required start to look like a different company.
What I Will Be Watching When The Papers Finally Drop
First, the bridge from the leaked 2025 loss to whatever the company claims is happening now. If the loss rate is collapsing as revenue scales, the bear note will need a rewrite. If the loss rate is merely being redescribed, the rewrite belongs to the bulls.
Second, related-party economics with cloud and chip suppliers. Preferential compute can be a genuine advantage or a subsidy that flatters margin. The footnote will tell you which, if you read it twice.
Third, the use of proceeds versus the selling-shareholder table. A company raising money to fund a plan is a different animal from a company mostly letting insiders out. Both can be legal. Only one looks like a long-term invitation.
Fourth, language. Destiny, safety of humanity, and unprecedented demand can all appear in a filing. So can retention, backlog, and cancellation terms. I trust the second set more on a Tuesday morning.
A Note On Being Wrong In Public
Trainer’s own line about not letting the misses bother him is healthier than the usual victory lap. Anyone writing about this IPO, including me, should borrow it. Anthropic could post the margins skeptics say are unavailable. Open models could stall at the enterprise door. A safety brand could become a pricing power brand. If that happens, the “ridiculous” label will join the DoorDash file, and the people who faded the open will have a scar.
The opposite path is just as available. Run rate could cool. A price war could arrive earlier than the models in the pitch book. Losses could stay large enough that even a successful product looks like a poor stock at the whispered tag. In that world the WeWork comparison, which the firm itself softened, will get dragged back out by people who enjoy being right more than they enjoy being precise.
I would rather be early to the question than late to the loss. Questions are cheap. Entry prices near the top of a private fantasy are not.
Pulling The Threads Without Pretending To Have The Filing
So where does that leave a reader who just wanted to know whether the loud note is noise? It leaves you with a tension worth keeping. The technology is more serious than the last generation of ridiculous-IPO targets. The valuation being discussed is more serious too, in the dangerous sense. A company can be good for the world and still be a bad buy at the wrong mark. Those sentences do not cancel each other.
The leaked picture, $4.6 billion of revenue and a $42 billion net loss, is the anchor until a real prospectus replaces it. The run-rate claims, $65 billion and a possible path toward $100 billion annualized, are the rebuttal until audited numbers confirm them. Nvidia’s profit pile is a yardstick, not a destiny. Open-source pressure is a ceiling on lazy pricing, not proof that closed labs cannot earn. Past calls from the same research firm are a reminder to stay humble, not a reason to outsource your judgment.
If the Anthropic IPO arrives near $2 trillion, it will not be ridiculous because a research note used the word. It will be risky because the price requires a profit trajectory few businesses in history have delivered, on a timeline the current loss rate has not yet earned. That is a narrower claim than the headline. It is also the one I can defend without pretending I have seen documents the public has not.
Markets can stay enthusiastic longer than a skeptical model stays comfortable. They can also reprice a story in a week once the costume comes off. Between those moods, the only durable edge is knowing what you own and what you paid for the privilege. Everything else is atmosphere.
