The numbers don't add up. When Alphabet reported a jaw-dropping $62.6 billion in quarterly profit last week, the market cheered. Google shares hit an all-time high. Analysts upgraded targets. The narrative was simple: AI is delivering. The AI bet is paying off. The trillion-dollar infrastructure buildout is generating returns.
But buried deep in the earnings report — and amplified by a Fortune investigation published just hours later — was a revelation that should make every investor pause: nearly half of that $62.6 billion profit, some $28.7 billion, came not from search ads, not from cloud revenue, not from YouTube subscriptions... but from revaluing Alphabet's stake in Anthropic. A private company. A paper gain. A number that exists only because someone decided Anthropic is worth more today than it was yesterday.
This is not profit. This is accounting theater. And Alphabet is not alone.
The Anthropic Markup Scam
Fortune's reporting, based on analysis of Big Tech's latest filings, reveals a pattern that stretches across the AI sector's most valuable companies. Amazon, which also holds a significant Anthropic stake, booked a $16.8 billion gain from the same revaluation. That single line item represented more than half of Amazon's entire pre-tax profit for the quarter. Remove the Anthropic markup, and Amazon's "AI-powered earnings beat" evaporates into a mediocre result.
The mechanism is straightforward. Alphabet invested billions in Anthropic over multiple rounds. Amazon poured in billions more. As Anthropic's valuation skyrocketed — from a few billion to $100 billion to a reported $1 trillion on secondary markets — those early stakes became worth dramatically more on paper. Under accounting rules, these investments must be "marked to market." When the market (or a funding round) says Anthropic is worth more, the investment's value rises. And that rise flows straight into the investor's profit and loss statement.
The problem is obvious: Anthropic's valuation is not revenue. It is not profit. It is not cash flow. It is a bet. A very large bet, backed by some of the world's most sophisticated technology, but still a bet. And Big Tech is treating that bet as if it has already paid out.
The circularity is almost comical. Anthropic's valuation is rising because Big Tech keeps investing in it. Big Tech's profits are rising because Anthropic's valuation keeps rising. The money goes around in a circle, and every quarter the numbers get bigger, and nobody asks where the actual value is being created.
How We Got Here
The Anthropic investment cascade began in earnest in late 2023, when Google committed $2 billion to the AI startup founded by former OpenAI researchers. Amazon followed with a $4 billion commitment. Over subsequent funding rounds, both companies increased their stakes. By early 2026, Alphabet and Amazon had collectively invested more than $25 billion in Anthropic, making them the company's largest outside shareholders.
At the time, these investments were framed as strategic partnerships. Anthropic's Claude models would power features in Google Cloud and Amazon's AWS. The companies would collaborate on AI safety research. The startups would gain compute resources; the tech giants would gain cutting-edge AI capabilities. It was a classic Big Tech venture strategy: invest in the disruptor to avoid being disrupted.
But as Anthropic's valuation exploded — driven by the broader AI boom, competitive pressure from OpenAI, and the company's own technical advances — those strategic investments became something else entirely. They became speculative assets. And speculative assets, when they rise in value, create accounting profits that have nothing to do with the underlying business.
The accounting treatment is technically correct. If you own 10% of a company that was worth $10 billion and is now worth $100 billion, your stake is worth $10 billion instead of $1 billion. That $9 billion increase is a real economic gain, at least on paper. But it is not the same as earning $9 billion from selling products or services. It is not the same as $9 billion in cash hitting your bank account. It is a valuation change, and valuation changes can reverse just as quickly as they advance.
The Valuation Problem
Anthropic's valuation trajectory should concern anyone paying attention. The company has gone from a $4 billion valuation in 2022 to a reported $1 trillion on secondary markets in early 2026. That is a 250x increase in four years. For context, Google itself took nearly two decades to reach a $1 trillion valuation. Anthropic did it in less time than most startups take to reach profitability.
The valuation is driven by a small number of very large funding rounds, each led by the same Big Tech companies that already hold stakes. When Google leads a round at a higher valuation, Alphabet's existing stake gets marked up. When Amazon participates, the same thing happens. The companies are, in effect, setting the price for their own assets. It is the corporate equivalent of appraising your own house and then claiming the appraisal as income.
Secondary market transactions — where early employees and investors sell shares to new buyers — provide some external validation. But these markets are thin, illiquid, and prone to selection bias. The buyers are typically other insiders, venture funds, or wealthy individuals who already believe in the AI thesis. The prices they pay reflect enthusiasm more than fundamentals. And when those prices are used to value the entire company, the result is a number that may have little relationship to what Anthropic would be worth in a real sale or public offering.
The company is reportedly now raising a new round at a $900 billion valuation, which would officially surpass OpenAI's $852 billion valuation. If that round closes, Alphabet and Amazon will book additional billions in paper gains. The cycle will continue. The profits will look even more impressive. And the underlying reality — that these are unrealized, illiquid, speculative gains — will remain hidden in plain sight.
What This Means for the AI Bubble
The Anthropic profit illusion is not an isolated accounting quirk. It is a symptom of a broader dynamic that is inflating the entire AI sector. When Big Tech companies invest in AI startups and then book the resulting valuation increases as profit, they create a self-reinforcing cycle. The profits justify more investment. The investment drives higher valuations. The higher valuations generate more profits. The cycle continues until something breaks.
That something could be a valuation correction. If Anthropic's next funding round comes in below expectations — say, $500 billion instead of $900 billion — the mark-to-market adjustment will flow in reverse. Alphabet and Amazon will be forced to write down their investments. Billions in "profit" will disappear. The quarterly earnings that Wall Street celebrated will be restated. The AI growth narrative will take a serious hit.
It could also be a fundamental setback. If Anthropic's models fail to keep pace with OpenAI, or if a safety incident damages the company's reputation, or if regulatory pressure forces changes to its business model, the valuation could collapse regardless of funding round dynamics. Private company valuations are fragile. They depend on continued optimism and continued capital inflows. Remove either, and the number can fall fast.
The broader risk is that investors are using these inflated profits to justify even more AI spending. Big Tech is on track to spend roughly $700 billion on AI infrastructure in 2026. Much of that spending is justified by the "returns" from AI — returns that, as we now know, are largely paper gains from private company valuations. If those paper gains prove ephemeral, the infrastructure spending will look like a massive misallocation of capital. The data centers, the chips, the power plants — all built for demand that may not materialize at the scale expected.
🔥 Our Hot Take
Here is the uncomfortable truth that nobody on Wall Street wants to say out loud: Big Tech's AI profits are a house of cards. The $62.6 billion Alphabet profit is really a $34 billion operating profit plus a $28.7 billion valuation fantasy. The Amazon beat is really a middling quarter dressed up with a $16.8 billion accounting trick. The entire narrative of "AI is generating returns" rests on a circular valuation game that could reverse overnight.
The Anthropic bubble is particularly dangerous because it is hidden in plain sight. These are not secret transactions. The investments are disclosed. The accounting treatment is standard. The valuation increases are reported. But the market has chosen to ignore the distinction between operating profit and valuation gains, because the headline numbers are so large and so impressive. Nobody wants to be the analyst who says the emperor has no clothes when the stock is hitting all-time highs.
But the emperor does have no clothes. And the moment of recognition is coming. It might be triggered by a disappointing Anthropic funding round. It might be triggered by a regulatory crackdown on mark-to-market accounting for private investments. It might be triggered by a broader tech selloff that forces investors to look more carefully at what they are actually buying. Whenever it comes, the correction will be sharp, because the gap between perception and reality is enormous.
The AI revolution is real. The technology is transformative. The infrastructure buildout is necessary. But the financial narrative surrounding AI has gotten ahead of the actual economics. Big Tech is not making $700 billion in AI profits. It is making billions in real operating profits and hundreds of billions in paper gains, and calling the total "AI earnings." That distinction matters. And when the market finally recognizes it, the AI bubble will look very different indeed.
For now, the music is still playing. The valuations are still rising. The profits are still being booked. But anyone who understands how this works knows that the current numbers are not sustainable. They are not even real. They are a valuation exercise dressed up as a profit statement. And when the exercise ends, the dressed-up numbers will vanish.