Anthropic Kicks Off IPO Roadshow at $1.4 Trillion Valuation

Original title: Anthropic启动IPO路演,估值已缩水6000亿美元
78AI Score

Anthropic has officially launched its IPO roadshow, planning to offer 825 million shares at $70-78 each for a deal of up to roughly $64.35 billion, with trading on Nasdaq under ticker ANT expected November 19. Amid a broad AI sector selloff, its potential valuation has fallen about 30% from roughly $2 trillion to about $1.4 trillion, while early shareholders could cash out $22.1-24.6 billion on day one.

Full Report

Contents3 sections

The whole world is hitting the brakes, while Dario alone steps on the gas

henry, from the Non-Office

QuantumBit | WeChat Official Account QbitAI

Even if it means catching a falling knife bare-handed, he's determined to land a big result!

Honestly, at the start, probably no one could have imagined that A-Corp's once-largest-ever IPO could turn out this brutal, or that Dario Amodei's head could be this hard.

Just now, the IPO roadshow for Anthropic, the parent company of Claude, officially kicked off, dispelling earlier rumors of withdrawal and postponement.

As planned, Anthropic intends to issue 825 million shares at a price of $70–78 per share, with an IPO transaction size of up to approximately $64.35 billion, and plans to begin trading on Nasdaq on November 19 under the ticker symbol ANT.

At that time, non-founder directors, executives, current and former employees, and consultants may, from the first trading day, sell up to 10% of their prescribed baseline holdings, corresponding to up to approximately 316 million shares.

In other words, on the first day of listing, A-Corp's early shareholders will have the opportunity to cash out $22.1 billion–$24.6 billion (around RMB 150 billion).

Likewise, investors who want to buy A-Corp that day will also face the same amount of selling pressure.

TBH, in the eyes of ordinary people like us, these $20-odd billion are already enough for quite a few people to settle down and live comfortably in the Bay Area, but measured against A-Corp's previous expectations, we have to say this is a severely shrunken, middling result.

Why? First, some background:

On November 2, Anthropic updated its prospectus. Although the third quarter still surged 47%, the growth rate had already slowed by 67% compared with the second quarter.

On November 5, the Nasdaq Composite Index fell about 5.5% in a single day, the largest one-day drop since April 2025.

On the same day, bonds of AI cloud service provider CoreWeave were sold off, with yields on some bonds spiking above 17%.

...

It's fair to say that right as A-Corp approached its big listing day, the global AI sector had already fallen into collective panic: valuations that needed to be cut were cut, blowups that needed to happen happened, and debts that needed repaying were repaid.

Back and forth, A-Corp's potential IPO valuation also shrank from the $2 trillion previously discussed in the market to about $1.4 trillion, a drop of roughly 30%.

Even more, on a fully diluted basis, this figure is already below the roughly $1.77 trillion market value of SpaceX at the time of its listing.

Quite a few employees also took advantage of this period, not waiting for Dario to ring the bell on Nasdaq, and fled from A-Corp to OpenAI overnight.

Hence, there were also rumors that A-Corp would postpone or suspend its listing...

However, with today's roadshow, we can only say that the lone, aloof ranger Dario still withstood the pressure.

Peripheral falling knives can be caught, its own valuation can be lowered, employees can leave, and shareholders can line up waiting to cash out, but this trillion-dollar IPO's big result—

I, Dario, must take it down hard.

The whole world is hitting the brakes, while Dario alone steps on the gas

Although the valuation has shrunk severely, objectively speaking, Anthropic's growth rate is already enough to go down in the history of technology.

Anthropic was founded in January 2021 by former OpenAI VP of Research Dario Amodei, his sister Daniela Amodei, and several former OpenAI researchers.

According to the roadshow PPT, since it officially began commercialization in the first quarter of 2023, by the third quarter of 2026, Anthropic's quarterly revenue alone had already approached $17 billion.

Interestingly, according to the pie A-Corp is drawing, this $17 billion is only the midpoint of the exponential curve.

Further up, there is knowledge work estimated to be worth more than $40 trillion in economic value each year, as well as more fields such as life sciences and robotics.

Although the picture is drawn a bit too starry-eyed, one has to admit that in less than six years, from an AI startup to what was once the largest IPO in history, A-Corp has indeed delivered growth close to exponential.

Based on the upper limit of the offering price and the post-issuance basic share capital, A-Corp's equity value is about $1 trillion, equivalent to about four Alibabas.

On a fully diluted basis including potential shares such as options and restricted stock, the valuation reaches about $1.4 trillion, second only to the largest IPO in history that combines xAI and SpaceX.

According to the roadshow materials, the 825 million shares offered in A-Corp's IPO include about 801 million newly issued by the company and about 23.65 million sold by existing shareholders.

What is truly worth mentioning here is the $26 billion in private funding that a group of compute backers are preparing to put in simultaneously.

Among them, Google and Nvidia each contribute $10 billion, AMD $2.5 billion, Broadcom, Micron, and Microsoft $1 billion each, and TSMC's investment entity $500 million.

In other words, the IPO itself raises up to $64.35 billion, plus this $26 billion in strategic private placement, bringing the overall financing scale to nearly $90.4 billion at most.

Of course, as for how this money will ultimately be used, and at what price it will enter the pockets of the backers, that still depends on Dario's subsequent moves.

On the operating data side, as we said at the beginning, A-Corp's revenue growth remains exaggerated, but compute spending is still expanding further, and the growth curve has therefore been slowed.

In the third quarter, Anthropic's revenue was $16.956 billion, 12.27 times that of the same period last year, up about 47% quarter over quarter, but clearly slower than the second quarter.

Looking at the full year, A-Corp's cumulative revenue for the first nine months was $33.242 billion, and the annualized revenue run rate disclosed on October 31 had already reached about $81 billion.

At the same time, customers contributing at least $100,000 in revenue over the past 12 months have exceeded 11,900. Compared with more than 950 at the end of September 2025, that is an increase of more than 12 times in one year.

Among them, more than 1,900 customers contribute $1 million or more in annual revenue, and about 250 customers contribute more than $10 million.

Even more exaggerated, by A-Corp's own measure, the net revenue retention rate of existing customers has reached about 900%.

This also means that Anthropic's growth does not rely entirely on continuously acquiring new customers, but on continuously tapping the payment potential of existing enterprise customers.

From this perspective, Dario has indeed delivered a rather impressive commercialization report card.

But that is precisely where the problem lies.

As the panic on Wall Street over the past two days reflects, the market is no longer really doubting whether AI is useful; instead, it has begun seriously crunching a different set of numbers:

Can the revenue generated by AI actually cover the enormous compute expenses and data center investments behind it? Does this business have sufficiently good economic returns?

For Anthropic, expensive compute procurement is making this nearly exponential growth curve come at an increasingly high cost.

In the third quarter, Anthropic's compute and infrastructure expenses reached $9.863 billion, rising from 48% of revenue in the previous quarter to 58%.

The company explained that the newly delivered compute in the quarter included about $5 billion of short-term resources, priced significantly higher than comparable resources procured through multi-year agreements.

In plain terms, because demand grew too fast, Anthropic had to pay a premium for short-term compute that could become combat-ready immediately.

As a result, Anthropic's adjusted operating profit rose from $2.235 billion in the previous quarter to $2.463 billion, but its adjusted operating margin fell from 19% to 15%.

Under US GAAP, the quarter's operating loss was $567 million.

And beyond the current quarter's costs, Anthropic and Dario also have an enormous bill stretching out over the next decade.

It is reported that to support upcoming growth, Anthropic has already signed approximately $900 billion in related procurement and leasing commitments.

Of that, about $451.6 billion is expected to occur before the end of 2030, with most of the remainder scheduled between 2031 and 2045; about $70.7 billion needs to be paid within the 12 months after the end of September 2026.

In other words, while businesses beyond Coding that can sustainably generate cash still need further validation, Anthropic has already locked in data center and chip resources for the next decade or more.

Is this borrowing from the future, or continuing exponential scaling? At least as of today, that remains a question mark.

How to reduce costs and improve efficiency?

During this roadshow, Anthropic also presented its cost-reduction and efficiency-improvement plan.

On one hand, through a combination of chips including AWS Trainium, Google TPU, NVIDIA and AMD GPUs, it matches more suitable compute to each task;

on the other hand, through Anthropic Direct, it leases chips directly and deploys data centers to lower long-term compute costs.

Meanwhile, according to roadshow disclosures, the amount of code contributed daily by Anthropic engineers has reached 45 times the 2024 average level.

Previously, when adapting to AMD GPUs, the company even had Mythos complete the porting and writing of nearly all low-level compute kernels within three days.

It is fair to say that when it comes to getting AI involved in an enterprise's own R&D and production processes, Dario has already charged ahead —

using AI to lower training and R&D costs on one side, while also making AI generate more revenue on the other.

Anthropic is also further replicating the business model already validated by Coding into more knowledge-work scenarios.

For example, just today, Claude released a series of office products, including Claude Sheets, Claude Whiteboards, and Claude Data.

Combined with integrations with third-party applications such as Adobe and Notion, Claude is trying to connect work such as spreadsheet processing, data analysis, collaboration, and content production.

In the short term, office AI will be the core scenario Anthropic bets on, while further out there are even larger markets such as life sciences and robotics.

However, now that the story has reached this point, Wall Street still seems to have its own views.

In the comment section of CNBC's report on this IPO roadshow, the screen was full of short sellers:

Dario is still doubling down on the future of AI, while the shorts have already set their sights on this IPO.

Whether next week brings more danger than promise, or whether the clouds will finally part, will become clear in due course.

By the way, according to Bloomberg, Dario also plans to hold a live conversation with JPMorgan CEO Jamie Dimon tomorrow. Those interested can tune in.

Reference links [1]https://ipo.anthropic.com/us/#founders-letter

One-click triple support:****"Like" "Repost" "Little Heart"

Welcome to leave your thoughts in the comments!

— End —

🌟 Light up the star 🌟

See frontier tech progress every day

Open in WeChat