Anthropic is heading toward the largest IPO ever, at a possible $2 trillion valuation

[Skye Jacobs]

Bottom line: Anthropic is heading toward a possible October IPO, with many investors telling the Financial Times that they expect the company to reach a valuation of $2 trillion or more. That figure would make it the largest public offering on record and put the company at the top of a market that is becoming increasingly cautious about AI spending and valuations.

The expectations are driven by Anthropic's rapid revenue growth. Investors expect its annualized revenue to reach $100 billion to $120 billion by the end of 2026. Anthropic said in May that its annualized revenue had surpassed $47 billion.

"If Anthropic is growing 800% a year, you'd think at the incredibly low end they would trade at 30 times [revenue]," one investor in the group told the Financial Times. "That would make them a $3 trillion company."

Anthropic has not set a public valuation target for the offering. Several investors said senior executives had not shared one privately, either. Still, backers have built their own models based on the company's enterprise sales growth and the performance of its AI systems.

The company has gained ground against OpenAI and Google this year. Its strategy has centered on business customers, with companies using Anthropic's models and tools in internal workflows and customer-facing products. Ramp data showed that Anthropic increased its share of US business spending on AI last month.

The same data points to a growing issue for the sector: companies are watching their AI bills more closely. Ramp analysts said businesses were "hitting their limit on AI spend" and shifting some workloads to cheaper systems.

Anthropic's top model costs more than two and a half times as much to use as OpenAI's flagship model, according to Artificial Analysis. Chinese open-weight models are considerably cheaper. That price gap matters as companies shift from pilots and small projects to large-scale deployments, where inference costs can rise quickly.

Some customers have already changed their approach. Rather than pushing employees to use the most capable AI tools whenever possible, they have moved certain tasks to lower-cost models. The shift does not necessarily mean demand for frontier systems is falling. It does mean companies are deciding more carefully which workloads require the highest-performing models.

Anthropic filed paperwork with the Securities and Exchange Commission in June. The filing placed the company in a quiet period, limiting what it could say publicly about its financial results. Anthropic declined to comment on the planned offering.

The company has raised just under $100 billion from venture capital firms, sovereign wealth funds, and other institutional investors in 2026. Its valuation reached $965 billion in May, including new investment, when it moved ahead of OpenAI for the first time.

But a public listing would come with risks that private investors have so far been willing to accept. Anthropic has faced pressure from the Trump administration and remains in litigation with the Defense Department, which labeled the company a supply-chain risk earlier this year.

The Commerce Department's export controls also forced Anthropic to briefly remove its Fable 5 and Mythos 5 models in June. Two investors said the disruption slowed overall revenue growth that month and raised concerns among customers who relied on the models.

The company recovered after that period, according to the investors. Even so, the IPO will test whether public-market investors are willing to place a multitrillion-dollar valuation on an AI company that is growing quickly but operating in a market where pricing pressure, regulation, and competition are all increasing.

"It's easy to come up with challenges," said an Anthropic investor who has also backed AI groups including OpenAI and SpaceX, which went public at a $1.77 trillion valuation in June. "But the company continues to be in first position in performance, positioning and what people want exposure to."

Permalink to story:

Anthropic is heading toward the largest IPO ever, at a possible $2 trillion valuation

 

[yRaz]

Well, we all saw how the spaceX IPO went. I do like Claude the most but I find myself using Gemini the most just because work pays for a base package. My job doesn't really pair with AI very well but it's helps for formating reports and excel documents. Maybe saves me 20 minutes a week and now it gives me a scape goat anytime there is a mistake. "Well you told me to try to use AI more."

 

[mbk34]

I'm not sure I'd want to invest. The AI industry as a whole smells like a bubble. Does it's income justify it's valuation? Is AI vulnerable to new ideas that might make existing techniques and hardware worthless. Requirements on water and energy which are getting increasingly costly. Communities rising up against data centres. The American economy looking increasingly delicate.

 

[umeng2002]

Oh... investors expect...

 

[Megalomaniac]

IPO are supposed to be after a company has proven to be profitable.

Now they are nothing more than Ponzi schemes to pay back the initial investors.

Space X is just the latest example of Elon Musk cvreating bag holders out of the fools who actually think we're going to Mars. Starlink is profitable but the Space division isn't.

 

[PXAbstraction]

Well, good luck if you choose to buy in I guess. Data so far (case in point, the SpaceX price) indicates to me not to go anywhere near companies tied to this bubble so I will be staying far away.

 

[NumberNine]

When this machine learning frenzy settles into something more sustainable than an arms race, when all the excitement and fomo are gone and the disruption appeal has faded... We will finally have reasonable pricing and new useful infrastructure which can be sustained without bribing local officials for land. Machine Learning isn't going away, but very few of the current "AI" companies, will survive the transition into mundane infrastructure. Anthropic is one of the most likely companies to exist on the other side of this.

Additionally, once the party has ended we'll be feasting like kings on the leftover hardware.

 

[Axeia]

Although I do feel like they're the best horse in this AI race to bet on by actually having a strong product/products and plans to turn it into those sweet dollars... personally, I still wouldn't invest in it.

Respect at least for having made it this far on their own strength.
They didn't have the media hype Open AI gets with AI messiah Sam Altman.
Not the clout/magic touch from Musk's promises.
Not Meta's endlessly deep pockets (not even the metaverse failure made them any less deep it seems).
Not Microsoft/Google (Alphabets) massive existing userbase and shoving it every existing product.

 

[RedGuard]

Ah, now it's time to take the money from the suckers.

 

[ScottSoapbox]

The key fact missing is what is the cost for Anthropic to make that $100B/year revenue?

They currently have the best AI models and business model, but how close they are to breakeven is crucial.

 

[wiyosaya]

Uh-Oh. fElon is going to be jealous. This will take all the attention away from him and his narcissistic head will explode!

<...> and now it gives me a scape goat anytime there is a mistake. "Well you told me to try to use AI more."

Perfect. I'll have to remember that if my company forces AI on us. So far, that hasn't happened, thankfully.

 

[pnntmp]

I think all projections of Anthropic's future growth and revenue should be drastically revised in the light of the recent Grok 4.6 release. It's faster, about 8x cheaper than Fable, and of comparable capability. In October, when the IPO is expected, Grok may be ahead. In any case, Anthropic will be forced to reduce their prices a lot.

It's also worth mentioning that roughly 1/3 of Anthropic's compute capacity is rented from SpaceX AI and can be taken away at any time with a 3 month notice.

For the IPO to be successful, Anthropic should categorically sideline Dario.
The temporary bump they got from Dario's preposterous theatrics with the DoW is long gone, but all the negative consequences remain if full effect. Potential investors must be ensured that can't happen again.

 

[yRaz]

The key fact missing is what is the cost for Anthropic to make that $100B/year revenue?

They currently have the best AI models and business model, but how close they are to breakeven is crucial.

If it wasn't for the billions of I'm just going to call "data center credit" then they wouldn't be able to function, neither would OpenAI. They're both burning through hundreds of billions a year. DCs are selling time on their hardware just so they don't sit idle and all the "profitable" AI businesses are entirely dependant on artificially cheap server time. The minute the same dollar keeps getting passed around it will all come falling down.

These "investments" these companies keep making in each other are also tied to a rate. That "$100billion investment" is tied to a number per unit of compute. People seem to be entirely ignoring that these companies aren't investing $100b. As an example, they're "investing" $25b of compute, saying it's worth $100b and then going to the bank saying "give us a loan based on these numbers".

The problem isn't as simple as these companies are posting weird revenue numbers or not making money. The problem is that numbers have been so fudged across so many spreadsheets that noone really actually knows what the real numbers actually are. They only people making money are nVidia, TSMC and the memory frauds.

And everyone is eventually going to end up packaging the massive debt and dump it on shadow companies/shell companies that file bankruptcy.

 

[LordVile95]

This will be great until Chinese backed and other open web AI proves more useful to individuals with better privacy features.

 

[LordVile95]

If it wasn't for the billions of I'm just going to call "data center credit" then they wouldn't be able to function, neither would OpenAI. They're both burning through hundreds of billions a year. DCs are selling time on their hardware just so they don't sit idle and all the "profitable" AI businesses are entirely dependant on artificially cheap server time. The minute the same dollar keeps getting passed around it will all come falling down.

These "investments" these companies keep making in each other are also tied to a rate. That "$100billion investment" is tied to a number per unit of compute. People seem to be entirely ignoring that these companies aren't investing $100b. As an example, they're "investing" $25b of compute, saying it's worth $100b and then going to the bank saying "give us a loan based on these numbers".

The problem isn't as simple as these companies are posting weird revenue numbers or not making money. The problem is that numbers have been so fudged across so many spreadsheets that noone really actually knows what the real numbers actually are. They only people making money are nVidia, TSMC and the memory frauds.

And everyone is eventually going to end up packaging the massive debt and dump it on shadow companies/shell companies that file bankruptcy.

In a gold rush sell shovels, though Nvidia might get into hot water with all the circular financing they’re engaged with

 

[yRaz]

In a gold rush sell shovels, though Nvidia might get into hot water with all the circular financing they’re engaged with

That's why they're backing out of a lot of it. They're also staying to talk about consumer products again but I haven't heard anything about the 6000 series

 

[Tantor]

AI is a little bit like the Garden of Eden. Both are fantasy paradises with two trees. The tree of knowledge of good and evil, and the tree of eternal life. AI promises it will grant knowledge about everything. It also has potential to drastically improve human lifespan.

God created the Biblical garden, but strictly forbade eating the fruit of knowledge of Good and Evil. Probably because he did not want to be recognised as evil by Adam and Eve. Note that the Snake who was supposedly evil, had no worries of being seen as evil. Probably because we was not evil. The Bible makes no mention that Adam and Eve were repelled by him after eating the fruit. But they did hide from God.

Similarly, modern elites created AI, but are very concerned about uncontrolled access. It may expose their dark secrets. Such as who was behind 911, the Epstein files or the Soviet Bolshevik revolution and world Communism.

However, the second tree was not initially forbidden by God. I'm pretty sure that AI will be used to increase human lifespan and the knowledge may trickle down to the masses.

So the Elites, like God, are stuck on the horns of a dilemma. AI grants the elites incredible knowledge but may unmask them. It may grant the elites eternal life, but it may also make the peons immortal, which is the last thing the elites want.

The Bible sums it up: "And God said, Behold! The man has become as one of Us, to know good and evil. And now, lest he put forth his hand and also take from the Tree of Life, and eat, and live forever, God sent him out of the garden of Eden to till the ground out of which he was taken." - Gen 3:22

The Snake was a bit like Prometheus. He brought forbidden knowledge to humanity against the will of God. Jesus went a step further, presenting himself as actual edible fruit of both trees. He said that eating his body would grant both knowledge and eternal life.

 

[bviktor]

Uh huh, revenue. Let’s not talk about profits though, coz that just happens to be negative.

 

[OortCloud]

Time for the Butlerian Jihad I say!

 

[FaTaL]

Everytime I asked claude to write me a python script, the script generated syntax errors.
ChatGPT - worked everytime.

 

[Loadedaxe]

AI is a little bit like the Garden of Eden. Both are fantasy paradises with two trees. The tree of knowledge of good and evil, and the tree of eternal life. AI promises it will grant knowledge about everything. It also has potential to drastically improve human lifespan.

God created the Biblical garden, but strictly forbade eating the fruit of knowledge of Good and Evil. Probably because he did not want to be recognised as evil by Adam and Eve. Note that the Snake who was supposedly evil, had no worries of being seen as evil. Probably because we was not evil. The Bible makes no mention that Adam and Eve were repelled by him after eating the fruit. But they did hide from God.

Similarly, modern elites created AI, but are very concerned about uncontrolled access. It may expose their dark secrets. Such as who was behind 911, the Epstein files or the Soviet Bolshevik revolution and world Communism.

However, the second tree was not initially forbidden by God. I'm pretty sure that AI will be used to increase human lifespan and the knowledge may trickle down to the masses.

So the Elites, like God, are stuck on the horns of a dilemma. AI grants the elites incredible knowledge but may unmask them. It may grant the elites eternal life, but it may also make the peons immortal, which is the last thing the elites want.

The Bible sums it up: "And God said, Behold! The man has become as one of Us, to know good and evil. And now, lest he put forth his hand and also take from the Tree of Life, and eat, and live forever, God sent him out of the garden of Eden to till the ground out of which he was taken." - Gen 3:22

The Snake was a bit like Prometheus. He brought forbidden knowledge to humanity against the will of God. Jesus went a step further, presenting himself as actual edible fruit of both trees. He said that eating his body would grant both knowledge and eternal life.

That is actually a pretty clever analogy, and I give you credit for making the Garden of Eden do some heavy lifting in an AI thread.

The part I think genuinely works is the idea that AI threatens information scarcity. For most of human history, expertise was expensive. You needed access to universities, lawyers, researchers, technical specialists, libraries, institutions, or simply a lot of money. AI potentially gives an ordinary person access to a surprising amount of analytical capability for very little cost.

That absolutely can shift power.

Where I part company with you is when the metaphor starts being treated as evidence for hidden historical conspiracies. AI may help uncover wrongdoing by connecting documents, identifying inconsistencies, searching enormous datasets, or making obscure information easier to find. But it still needs reliable source material. It cannot summon some mystical “truth database” and reveal who secretly orchestrated every major event in history.

Garbage in, omniscient serpent out.

The longevity point is more interesting than it may initially sound. AI is already being applied to drug discovery, protein research, diagnostics, genomics and other areas where better computational tools could accelerate medical progress. If those technologies eventually extend healthy human lifespan, the real argument probably won't be whether “the elites” want everyone living longer. It will be the much more ordinary human problem of who gets access first, what it costs, and how long it takes before it becomes broadly available.

That pattern is hardly new. Wealthy people generally get new medicine, computers, communication technology and other expensive innovations first. Then competition and scale tend to push them downward through society.

I also think there's a less sinister explanation for why governments and corporations worry about unrestricted AI. A system capable of democratizing knowledge is useful, but the exact same capability can democratize fraud, cybercrime, propaganda, biological knowledge and other things we'd rather not hand out with a cheerful “Ask me anything” button.

So I don't think we're quite at...God - elites, Serpent - AI, Adam and Eve - Reddit users with unlimited compute.

Although...I'll admit that last one would explain quite a lot.

The strongest portion of your argument, to me, is simpler...AI may reduce the advantage historically enjoyed by people and institutions that controlled access to knowledge.

That is a very real possibility, and probably much more profound than any secret file AI might eventually dig up.

Sources, for anyone wishing to consult the Tree of Peer Review before eating the fruit.

NIH / National Library of Medicine - AI in drug discovery
Nature - AlphaFold / protein structure prediction
NIST - AI risks and misuse

 

[NotYourITguy]

That's why they're backing out of a lot of it. They're also staying to talk about consumer products again but I haven't heard anything about the 6000 series

Most of this is general consumers not understanding how corporate financing works and lack of understanding that consumer GPUs are a terrible investment of resources right now. Once supply chain shortages subside over the next couple of years, Nvidia will crank out more consumer GPUs again because it'll be worth doing from a company growth standpoint.

It would be a brain dead business move to launch consumer RTX 6000 series GPUs at 1/10th the profitability of pro series units that are sold by the literal truckload when component shortages mean prioritization must be done on which cards get fully built and shipped.

 

[Anton Longshot]

That is actually a pretty clever analogy, and I give you credit for making the Garden of Eden do some heavy lifting in an AI thread.

The part I think genuinely works is the idea that AI threatens information scarcity. For most of human history, expertise was expensive. You needed access to universities, lawyers, researchers, technical specialists, libraries, institutions, or simply a lot of money. AI potentially gives an ordinary person access to a surprising amount of analytical capability for very little cost.

That absolutely can shift power.

Where I part company with you is when the metaphor starts being treated as evidence for hidden historical conspiracies. AI may help uncover wrongdoing by connecting documents, identifying inconsistencies, searching enormous datasets, or making obscure information easier to find. But it still needs reliable source material. It cannot summon some mystical “truth database” and reveal who secretly orchestrated every major event in history.

Garbage in, omniscient serpent out.

The longevity point is more interesting than it may initially sound. AI is already being applied to drug discovery, protein research, diagnostics, genomics and other areas where better computational tools could accelerate medical progress. If those technologies eventually extend healthy human lifespan, the real argument probably won't be whether “the elites” want everyone living longer. It will be the much more ordinary human problem of who gets access first, what it costs, and how long it takes before it becomes broadly available.

That pattern is hardly new. Wealthy people generally get new medicine, computers, communication technology and other expensive innovations first. Then competition and scale tend to push them downward through society.

I also think there's a less sinister explanation for why governments and corporations worry about unrestricted AI. A system capable of democratizing knowledge is useful, but the exact same capability can democratize fraud, cybercrime, propaganda, biological knowledge and other things we'd rather not hand out with a cheerful “Ask me anything” button.

So I don't think we're quite at...God - elites, Serpent - AI, Adam and Eve - Reddit users with unlimited compute.

Although...I'll admit that last one would explain quite a lot.

The strongest portion of your argument, to me, is simpler...AI may reduce the advantage historically enjoyed by people and institutions that controlled access to knowledge.

That is a very real possibility, and probably much more profound than any secret file AI might eventually dig up.

Sources, for anyone wishing to consult the Tree of Peer Review before eating the fruit.

NIH / National Library of Medicine - AI in drug discovery
Nature - AlphaFold / protein structure prediction
NIST - AI risks and misuse

A link to a super interesting article about AI drug discovery :

https://www.science.org/content/blog-post/so-how-ai-drug-discovery-doing-really

 

[yRaz]

Most of this is general consumers not understanding how corporate financing works and lack of understanding that consumer GPUs are a terrible investment of resources right now. Once supply chain shortages subside over the next couple of years, Nvidia will crank out more consumer GPUs again because it'll be worth doing from a company growth standpoint.

It would be a brain dead business move to launch consumer RTX 6000 series GPUs at 1/10th the profitability of pro series units that are sold by the literal truckload when component shortages mean prioritization must be done on which cards get fully built and shipped.

I think what I was trying to say went over your head. nVidia is already showing signs that the enterprise market is slowing. Maybe not on quartly reports but it does look like they aren't getting as many orders 6.months to a year down the line.

 

[Loadedaxe]

A link to a super interesting article about AI drug discovery :

https://www.science.org/content/blog-post/so-how-ai-drug-discovery-doing-really

Considering some of the AI discussions we've been having lately, this article is absolutely ammunition for both sides. The AI haters are going to read “disappointingly limited” and stop there, while the AI evangelists will read “absence of evidence” and stop there.

Lowe is doing something considerably smarter than either camp.