Anthropic Hit a $965 Billion Valuation on the Fastest Revenue Ramp in Software History

The Growth Numbers Are Unlike Anything in Software History

Anthropic was founded in 2021 by Dario and Daniela Amodei and five other former OpenAI researchers who raised concerns about the company's direction and AI security priorities. Five years later the company they built has grown from $1 billion in annualized revenue at the end of 2024 to about $47 billion in annualized run-rate revenue in May 2026 a 47-fold increase in 17 months. NoneAnthropic went from $1 billion to $47 billion in less than two years. The company's revenue growth rate has been described as 10x annually for three consecutive years a trajectory that has no useful historical analogy in the B2B software industry

On June 1 2026 Anthropic confidentially filed a draft S-1 registration statement with the SEC setting the stage for a possible public listing as early as October 2026. Goldman Sachs JPMorgan and Morgan Stanley are leading an offering that is expected to raise more than $60 billion which would make it the second-largest IPO in history trailing only SpaceX's $75 billion raising two weeks earlier. The latterThe company's private valuation was $965 billion set in a $65 billion Series H-1 round in May 2026. Investment bankers working on the deal consider a debut above $1 trillion to be the base case assuming markets cooperate

What Anthropic Actually Is

Anthropic's core product is Claude a family of AI assistants with large language models that power both consumer-facing interfaces and enterprise API applications. The company's revenue model has two main streams: API usage fees charged to developers and businesses that integrate Claude into their own products and enterprise subscription contracts with large enterprises that deploy Claude in their organizations. Approximately 80% of Anthropic's revenue comes from enterprise customers a concentration that provides visibility ofrevenue but creates a risk of customer concentration if key accounts churn or reduce usage

The flagship product of 2026 has been Claude Code an AI coding assistant that is integrated into developers' workflows and has achieved $2.5 billion in annualized recurring revenue as a standalone product. Eight of the Fortune 10 companies are enterprise customers of Anthropic deploying Claude in functions ranging from financial analysis and legal review to software development and customer service automation. The company has also been building specific vertical applications: Claude for Finance Claude for Healthcare and Claudefor Legal each designed with domain-specific training and compliance features that enterprise buyers require

A Worked Example: What 965 Billion Dollars Is Assuming

It's impossible to have intuitions about a valuation of this size so the only honest way to evaluate it is to run it backwards and see what it requires

Start with the manifold. $965 billion versus $47 billion annualized revenue is about 20.5 times revenue. Historically mature high-quality enterprise software companies have traded at 6 to 10 times revenue. So the price already contains a huge amount of future growth before this happens

Now examine the margin bridge which is the most difficult problem. Gross margin today is around 40 percent compared to the internal goal of 77 percent by 2028 with revenues of $70 billion. Those two numbers look like a normal improvement plan. Solve them and they are not

With 47 billion in revenue and a gross margin of 40 percent the cost of revenue is about 28.2 billion of which about 19 billion is calculated.The gross profit is about 18.8 billion

If the target of 70 billion revenue and a gross margin of 77 percent by 2028 is achieved the cost of revenue should be 23 percent of 70 billion or about $16.1 billion

Read the cost of revenue row. Serving 49 percent more revenue has to cost 43 percent less in absolute dollars. Per dollar of revenue delivered the cost has to fall about 62 percent in about two years

That's the whole bet and it's a bet on the price of inference crashing rather than selling more. Obviously not bad. The cost per unit of model capacity has fallen dramatically and repeatedly. But it means that the difference between a company worth a trillion dollars and one worth a fraction of that amount is decided by a hardware and efficiency curve and not by something in the sales process

Finally run the valuation backwards to a normal multiple. Assuming this will eventually grow into a mature enterprise software business trading at 8 times revenue then $965 billion implies revenue of about $121 billion. That's about 1.7 times the internal target for 2028 and about 2.6 times current revenue

So the price does not ask whether Anthropic meets its 2028 plan. It assumes that the plan is met and then substantially exceeded and that the margin transformation occurs as planned. Neither assumption is unreasonable given the sequence of growth in the description above. That both things are necessary simultaneously is what the phrase no margin for error actually means

These figures use rounded inputs and a single assumed terminal multiple and moving that multiple moves the answer a lot which is itself the point

The Competitive Landscape

Anthropic and OpenAI filed confidential IPO documents within days of each other in early June setting up what will likely be the most watched IPO competitive dynamics since Google and Yahoo in the early 2000s. The two companies have been alternating advantages in frontier model capabilities throughout 2025 and 2026 OpenAI's GPT-5 series Anthropic's Claude 3.7 Sonnet and Claude 4 Opus and Gemini 2.5.Google's Ultra have taken turns being considered the market-leading model for different benchmarks and use cases. Anthropic's positioning around "AI safety" - building models that are more reliable more interpretable and less likely to produce harmful results - has become commercially significant as companies become more cautious about deploying AI in regulated industries. That safety positioning has translated into regulatory credibility in the EU and the UK where Anthropic haswon government contracts that OpenAI has not obtained

The competitive comparison based on figures as of the filing date: Anthropic's annualized revenue of $47 billion exceeded OpenAI's $25 billion ARR estimate. Anthropic's $965 billion valuation exceeded OpenAI's $852 billion. Anthropic's 10x/year revenue growth rate exceeded OpenAI's estimated 3.4x.is in the lead the opposite of the situation 18 months ago when OpenAI was widely seen as the dominant commercial player and Anthropic was the security-focused academic upstart

The Investment Case, and the Real Risks

The bullish case for an anthropic IPO is based on the revenue trajectory and business moat. If the company achieves $70 billion in revenue and 77% gross margins by 2028 internal targets cited in IG's coverage of the confidential filing the long-term earnings power justifies a valuation well above $1 trillion on normal enterprise software multiples. The $70 billion revenue target implies continued growthfrom $47 billion requiring the enterprise AI adoption curve to continue without a major downward inflection

The risks are real and significant. First Anthropic is not yet profitable and spends approximately $19 billion a year on computing alone. Gross margin is currently around 40% well below the 77% target and well below the 70-80% margins that justify enterprise software multiples. Compute cost is the main barrier to profitability and is declining as the efficiency of the model improves but "decline" and"Achieved profitability" are not the same thing. Second Anthropic is currently locked in a legal battle with the US government after the Pentagon declared it a supply chain risk a designation typically reserved for companies with foreign ownership concerns. Third the IPO's valuation of $965 billion implies roughly 20 times future earnings a premium that essentially leaves no room for error if revenue growth slows relative toits current pace.Fourth computing costs are shared with Amazon and Google who also compete directly with Anthropic in artificial intelligence services a structural conflict of interest that will need to be disclosed in the public S-1 and will face intense investor scrutiny

Case Study: Snowflake and the Price of a Perfect Business

There is recent precedent for a great fast-growing enterprise software company to be purchased at a historically unprecedented multiple and it is worth knowing the outcome before anyone forms an opinion on this offering

Snowflake went public in September 2020 in what at the time was the largest software IPO in history. The offering was priced at $120 per share. It opened trading around 245 and closed around 254 valuing the company at about $70 billion against product revenues ranging from $500 million to $600 million a year. That's a higher multiple than100 times revenue

The company was not overvalued in any operational sense. It was truly excellent. Revenue grew several times over in the years that followed net income retention was among the best ever recorded in enterprise software and it won the workloads for which it competed. Almost all of the operating assumptions in the bull case came true

The stock spent years below its first-day close. The investors who bought on the first day owned a company that performed brilliantly and delivered nothing to them because multiples compressed faster than revenue grew. Quadrupling revenue while the multiple drops by more than four leaves you worse off and that's arithmetic rather than misfortune

Apply the lesson instead of the specific numbers. At 20.5 times revenue Anthropic is priced much more conservatively than Snowflake which is a real and significant difference. But the mechanism that hurt Snowflake investors is fully present here: When a company's price is based on a multiple that assumes exceptional growth persists growth that comes as expected is the neutral case not the good one. Performance comes from whether the multiple holds up andThe multiple is set by other people's willingness to pay which is the only variable that no company controls

The uncomfortable corollary is that being right about Anthropic as a business and wrong about the entry price produces the same result as being wrong about the business. Snowflake holders learned that with a company that basically did everything right

Where the Bull Case Is Weakest

Four issues that I think deserve more scrutiny than the headline growth numbers are currently receiving

The annualized run rate is a weak measure for usage-based revenue. Taking a strong month and multiplying it by twelve is standard practice and assumes that the month repeats. Contract subscription revenue does this largely. Consumption revenue where customers pay for what they use and can reduce it at any time without canceling anything does not carry the same commitment. A step function from 30 billion to 47 billion in a single month as in the sequence above is impressive and is also exactly the pattern that most favors annualization

Income has a circularity problem. Amazon and Google are simultaneously major investors in Anthropic their largest computing providers and competitors in artificial intelligence services. Capital flowing from a supplier to a customer who spends it back on that supplier's product is a structure that requires very careful reading and is one of the reasons why the public S-1 will be more informative than anything published so far

Customer focus goes against the visibility argument. Eighty percent of enterprise revenue with eight of the Fortune 10 as clients is presented as a strength and it is until a small number of very large accounts renegotiate simultaneously. The article above points this out. It's worth saying more clearly: enterprise concentration produces excellent visibility and terrible tail risk and that is the same fact

The Salesforce comparison is not fair to either company. Salesforce spent eight years reaching $1 billion because it had to create a marketplace sell a new deployment model to skeptical buyers and build the underlying infrastructure. Anthropic sells into a market with enormous pre-existing budget urgency and a distribution mechanism the API that didn't exist in 2004. The growth is real and comparison inflates it

My honest position is that the quality of the business appears genuine and the price incorporates a specific technical forecast on inference costs that most people citing revenue figures have not examined. That is an opinion not advice and I have no position on it

How I Would Read the Public S-1

The confidential filing means there are still interesting revelations ahead. When the public version arrives this is the order I would work in

I would go to the revenue recognition and breakdown notes first before the growth narrative to see how much revenue is contracted subscription and how much is consumption. That single division determines whether the annualized figure means what the headline implies

Second I would look into customer concentration disclosure which registrants must provide when a single customer exceeds ten percent of revenue. With eight of the Fortune 10 companies as customers whether any of them cross that line is one of the most valuable facts the filing will contain

Third I would go to the related party transactions section for the Amazon and Google deals. The investment terms calculation commitments and any minimum spend obligations belong there and reading them together is the only way to honestly size up the circularity issue

Fourth I would track the cost of revenue per unit of revenue over the periods presented because that ratio is the margin bridge in the worked example above. Two or three data points showing a sharp decline would support the 2028 target. A fixed ratio would invalidate most of the valuation

Fifth I would carefully read the risk factors in the Pentagon's designation rather than dismiss them as boilerplate since a supply chain risk determination affects government revenue and can influence procurement decisions far beyond the agency that made it

This is how I would approach the document. It is a description of the method and not a recommendation on the offer

The Bottom Line

Anthropic grew from $1 billion in annualized revenue in December 2024 to about $47 billion in May 2026 filed confidentially on June 1 and has a latest private valuation of $965 billion with bankers considering a $1 trillion debut as a base case. There is nothing comparable in the history of enterprise software

The price arithmetic is where the argument really is. 965 billion versus 47 billion revenue is about 20.5 times versus 6 to 10 times for mature enterprise software. Reaching the internal 2028 target of 70 billion revenue with a 77 percent gross margin means the cost of revenue must fall from about 28.2 billion to about 16.1 billion an absolute reduction of 43percent while revenue grows 49 percent or roughly a 62 percent cut in cost per dollar of revenue. And keeping the valuation at a normal 8x multiple implies about $121 billion in revenue about 1.7 times the 2028 plan

The snowflake is the case to hold next to it. It went public in 2020 with over a hundred times revenue executed almost flawlessly increased revenue several times over and left first day buyers underwater for years because the multiple fell faster than the business grew. The IPO debate is going to be spectacular and the number to argue about is not the growth rate. It is what the price already assumes over the cost of inference in twoyears