Alphabet reported a record $62.6 billion in profit. Amazon’s quarterly earnings climbed 38%. Both were treated as proof that the AI bet is paying off, and both also hid a quieter truth in the fine print: a large slice of those gains came from the rising value of their stakes in Anthropic, not from ads, cloud contracts, or product sales.

That is the heart of the story in Big Tech profits inflated by startup stakes. Fortune reported this week that roughly half of Alphabet’s total profit, about $28.7 billion, flowed from its Anthropic position. Fortune also reported this week that Amazon’s $8 billion investment in the startup now carries a book value above $70 billion. Those are unrealized gains, which means no cash changed hands, no customer was won, and no product was shipped.

The figures are large enough to change the story. When Big Tech posts blowout AI profits, how much of the number comes from the business itself, and how much comes from private-market marks that investors cannot see, test, or trade? The answer sits in the accounting, but the accounting is not what most people think they are reading.

The mechanism starts in private markets, not in product divisions. When a startup raises money at a higher valuation, the value of existing stakes can be marked up too. Under current accounting rules, that revaluation flows into reported income. No asset is sold. No operating step has improved. Someone simply agreed to pay more for a newer slice of the same company.

That is what happened with Anthropic. Business Insider reported in April 2025 that Anthropic was valued at about $41 billion at the end of 2024 and at $61.5 billion after a March funding round. In the same report, Business Insider noted that Google’s quarterly earnings release included “an $8.0 billion unrealized gain on our non-marketable equity securities related to our investment in a private company.” The company did not name the startup in that line, but the math pointed to Anthropic.

Amazon’s version of the same story showed up in nonoperating income. Bloomberg reported last week that Amazon’s third-quarter profit climbed 38%, helped by a $9.5 billion pretax gain from its investment in Anthropic. Bloomberg also reported that the higher value was reflected in Amazon’s nonoperating income for the period. Alphabet, meanwhile, said profit included “net gains on equity securities of $10.7 billion,” in part from a private company, and Bloomberg reported that people familiar with the matter identified that company as Anthropic.

None of this is improper. It is disclosed, and the accounting treatment is allowed. But it does mean investors have to read the profit line with more care than the headline invites. Operating income tells you how the business is doing. Nonoperating gains can tell you what happened to a portfolio stake after a funding round. Those are not the same thing, even when they land in the same quarter.

A simple test helps separate the two. If a gain is described as unrealized, tied to equity securities, and triggered by a private financing event, it is a valuation event. It may matter a great deal, but it is not the same as selling more ads or moving more cloud capacity.

Alphabet is the cleanest example because the numbers were so large. Business Insider reported in April 2025 that the company’s net income rose to $35.5 billion from $23.7 billion a year earlier, and that the increase was materially shaped by the $8 billion unrealized gain on a private company stake. That comparison looks like a sharp improvement in operating momentum. Part of it was. Part of it was a startup mark.

That distinction matters because Alphabet’s core business is still doing the heavy lifting. Reuters reported last week that investors were especially pleased with Alphabet’s ability to fund its AI spending from operating cash flow. Reuters said Alphabet’s capital expenditure in the September quarter was $23.95 billion, equal to 49% of cash generated from operations, compared with 64.6% for Meta and 77.5% for Microsoft. Alphabet’s shares rose as much as 6% before closing about 3% higher.

That is the useful part of the story. Alphabet’s operating business is strong enough to support a very expensive AI buildout without leaning as hard on the balance sheet as some rivals. But that operating strength arrived wrapped around a separate and much less durable source of profit. One is cash generation. The other is a valuation change in a private company.

Amazon’s case is similar, though the accounting path is slightly different. Bloomberg reported last week that the company’s $8 billion investment in Anthropic had become worth more than $70 billion. Fortune reported this week the same figure in the same range. Either way, the scale of the paper gain is obvious. What matters for readers is that this kind of increase can lift earnings even when the underlying business is doing something far more ordinary.

The point is not that investors should ignore these gains. It is that they should put them in the right bucket. Operating profit reflects sales, margins, and demand. Startup-stake gains reflect financing rounds, market appetite, and the judgment of a small group of private investors. The first can be modeled. The second can disappear.

There is another reason to be cautious. Private valuations in AI are not being set in a clean, independent market. They are being shaped inside a network of companies that often invest in the same startups, buy from the same startups, and sell to the same startups.

Bloomberg mapped that circularity earlier this year and warned that these deals can create skewed incentives and magnify losses if AI demand falls short of the hype. Bloomberg’s own description is blunt: a company whose supplier is also a major shareholder may keep buying even when it no longer makes commercial sense. If demand cools, the same company can get hit twice, once because it buys less, and again because its stake falls in value.

That matters here because the Anthropic markups sat inside that same ecosystem. Reuters reported in March that Nvidia had finalized a $30 billion investment in OpenAI and committed $10 billion to Anthropic. Reuters also reported that Nvidia and OpenAI had announced a $100 billion deal in September last year. The lines between supplier, customer, and investor are getting harder to draw, which is not usually a sign of pristine price discovery.

Bloomberg also noted that the circularity is most risky when a handful of buyers account for a large share of the market, which is exactly how AI spending looks right now. That does not make the gains fake. It makes them fragile. If AI spending slows, the fallout would not stay neatly inside one line item.

The right way to read these earnings is to split the story into two questions. First, is the company actually earning more from operations? Second, is part of the reported profit coming from a private investment that was revalued after a funding round? If the answer to the second question is yes, the headline needs a lot more context than the market usually gives it.

There is a decent reason investors have been willing to squint at the fine print. Reuters reported last week that Alphabet’s ability to fund rising AI spending from cash flow helped lift the stock even as other tech names came under pressure. The market is not blind to the difference between cash and paper. It just tends to notice after the earnings call, which is a charmingly inconvenient habit.

The bigger test is still ahead. If Alphabet and Amazon can keep posting strong operating results even as startup-stake gains fade or stop altogether, that will show the AI buildout has real earnings power behind it. If profits weaken sharply once the valuation marks cool off, then the headline numbers were doing more work than the businesses themselves.

That is the line readers should keep in mind. Startup-stake gains can flatter earnings for a while, but they are not a substitute for operating performance. The hard part is separating the two before the market does it for you.