SemiAnalysis: Anthropic Will Become the First $6 Trillion Market Cap Company
TechFlow Selected TechFlow Selected
SemiAnalysis: Anthropic Will Become the First $6 Trillion Market Cap Company
Shadow Ledger Calculates Anthropic at $6 Trillion?
Navigating Web3 tides with focused insightsBy: TechFlow Research
On June 1, Anthropic secretly filed an S-1 with the SEC.
Five weeks later, the well-known research institution SemiAnalysis released a long report, using its own models to piece together the company's complete financial statements from the bottom up by product line, pricing tiers, and customer types. Wall Street calls this kind of thing a shadow ledger, and the conclusion boils down to one sentence: Anthropic could become the first company in human history with a market cap of $6 trillion.
What does this number mean? It is more expensive than any listed company on Earth, including Nvidia. SemiAnalysis did not start as a clickbait outlet; this institution began by dissecting chip supply chains and data center costs, so where does its conclusion come from?
A Curve Steep Beyond Belief
First, let's look at the three core numbers in the report.
Annual Recurring Revenue (ARR): $9 billion at the end of 2025, currently over $60 billion, nearly sevenfold in half a year.
Monthly Net New ARR: $3 billion in January this year, jumping to $11 billion in March.
Profit: In Q3 2026, Earnings Before Interest and Taxes (EBIT) on a GAAP basis will exceed $1 billion.
For reference, Zoom, which set the growth record in the SaaS industry during the pandemic, took about four years to reach $1 billion in ARR. Anthropic took 11 months, and then added hundreds of billions more in the following 17 months.
The engine behind the steep curve has a name: Claude Code.
This AI coding tool launched in mid-2025, and the report estimates it currently accounts for more than 7% of all code submissions on GitHub. In other words, for every 14 lines of code submitted by programmers worldwide, 1 line comes from it.
The customer stickiness data is even more exaggerated. The report cites figures disclosed by Anthropic's CFO, with Net Revenue Retention (NRR) reaching 500%: the batch of customers contributing $30 billion in ARR this year contributed only $2 billion a year ago. Old customers grow on their own; this is the most expensive kind of growth in the software industry.
From Money-Burning Machine to Money Printing Press
There are many fast-growing AI companies; the real explosive point of this report lies in the income statement.
SemiAnalysis calculates that Anthropic's overall gross margin has rebounded from negative 94% in 2024 to around 65% currently, with API business gross margin exceeding 80%. In two years, this machine has transformed from the industry's most fierce money-burning furnace into a money printing press with gross margins comparable to mature software companies.
The secret lies in inference efficiency.
Measured by ARR per megawatt of computing power, this figure was $16 million nine months ago and will reach $60 million later this year. The same unit of electricity extracts nearly four times the revenue.
When Opus 4.5 was released last November, the price was cut to one-third of the previous generation. The market once thought it was a price war, but SemiAnalysis gave the opposite judgment at the time: after the price cut, Anthropic's profit margin on this product line was actually higher because the production cost per token dropped faster.
This scene has a historical counterpart.
In April 2015, Amazon disclosed AWS's financial figures separately for the first time. Wall Street suddenly discovered a high-margin business hidden inside this money-burning machine, and Amazon's valuation logic was rewritten thereafter. The day Anthropic's S-1 is revealed will likely be the AI industry's AWS moment.
OpenAI Becomes the Reverse Side in the Mirror
The sharpest part of the report is a set of mirror comparisons using OpenAI as a reference.
In Anthropic's revenue structure, 75% to 85% comes from usage-based API billing, while consumer subscriptions account for only 5%; OpenAI is exactly the opposite, with subscriptions being the bulk, backed by hundreds of millions of free users.
Free users are a moat in the narrative, but a bottomless pit on the cost sheet: the report deduces that assuming both companies reach $100 billion in ARR, OpenAI's gross profit will be about $25 billion less than Anthropic's because it has to support free users. This difference directly determines who has more bullets to train the next generation of models. By 2028, the report expects the cumulative EBIT gap between the two to widen to $250 billion.
The report uses a provocative statement: Anthropic has the ability to make OpenAI dance to its rhythm. Going public first is the first dance step, throwing the pressure of disclosing finances and accepting market testing onto the opponent unchanged. OpenAI's IPO is rumored to be postponed to 2027.
The Real Motive for Going Public: Buying Electricity and Computing Power
Why now? The report calculated a computing power bill.
By 2030, the unconstrained computing power demand of Anthropic and OpenAI combined will exceed 100 GW. The reality is that the net new computing power for the entire industry in 2025 and 2026 is only 2.5 GW and 5 GW respectively, and the computing power currently available to the two companies combined is just over 6 GW. There is a 94 GW gap between demand and supply, and there is only one material to fill it: money, and cheap money at that.
The window is closing. Alphabet has completed $84.75 billion in equity financing, and Meta is rumored to be planning financing at the hundred-billion level. No matter how big the capital market's appetite for AI infrastructure is, it cannot withstand several trillion-dollar giants drawing water in turn. Whoever goes public first locks in the ammunition first. This IPO is less a milestone and more like a margin deposit for computing power futures.
Cracks in the Shadow Ledger
After analyzing the hype report, it is time to talk about what it didn't say.
This endpoint value of $6 trillion is built on an extremely thin assumption: monthly net new ARR remains stable at $15 billion without attenuation, pushing all the way to $300 billion in ARR by the end of 2027. The acceleration from $3 billion to $11 billion did happen, but drawing a three-month slope into a two-year straight line is the oldest magic trick in sell-side research. Any quarter of slowdown, and $6 trillion will shrink to $4 trillion or $3 trillion, changing the nature of the story.
The ledger itself also has a dark seam.
Forecasting institution FutureSearch cites Sacra's data pointing out that Anthropic records terminal customer consumption resold via cloud platforms as revenue at gross amount, with platform splits recorded as expenses. If required to change to net recognition during IPO due diligence, the book ARR could shrink by 20% to 40% at once.
There is also a stance issue. SemiAnalysis itself disclosed that its annualized spending on Claude tokens once reached as high as $10.95 million; it is one of Anthropic's heaviest paying customers, using Claude to do research and then researching Claude's owner. Models can be made extremely detailed; one's position cannot determine the data, but it will determine where the spotlight shines.
Regulatory risk is no longer an assumption. In June, a U.S. government export control order caused Anthropic's newly released flagship models Fable 5 and Mythos 5 to stop services globally. The report lists regulation at the end of the risk warning section, but in reality, it fired a shot just last month.
The secret prospectus will eventually be revealed one day, most likely this autumn or winter.
Join TechFlow official community to stay tuned
Telegram:https://t.me/TechFlowDaily
X (Twitter):https://x.com/TechFlowPost
X (Twitter) EN:https://x.com/BlockFlow_News
@chaoxiangooo
