The comically codependent AI capital circle
The Wall Street Journal reports that Anthropic is pacing toward its first profitable quarter on a non-GAAP basis, projecting billions in revenue. However, a deeper dive into recent tech financial reports reveals a "comically codependent" accounting loop: hyperscalers are pumping massive capital into AI labs, marking up their own private equity stakes to book billions in "other income," and watching that exact same cash cycle straight back into their cloud infrastructure buckets as compute fees.
Wall Street Journal readers woke up to a dazzling headline: “Mind-Blowing Growth Is About to Propel Anthropic Into Its First Profitable Quarter.” According to leaked investor decks, Anthropic is projecting a massive $10.9 billion in revenue for Q2 2026, pulling in a tidy $559 million in "operating profit." It sounds like the ultimate vindication for the frontier AI labs. The unit economics are scaling, the enterprise demand for Claude Code is rampant, and sustainability has finally arrived.
Then you read the literal last two paragraphs of the WSJ report:
It is unclear what accounting methods Anthropic has used to book revenue and costs, as the company isn’t yet required to follow the financial reporting requirements of a public company... Anthropic counts sales of its technology through cloud partners as revenue, while OpenAI doesn’t.
Ah. There it is.
When you peer past the press release formatting, the numbers start to look a lot less like a tech miracle and a lot more like creative financial engineering. Word on the street is that Anthropic is claiming non-GAAP EBITDA profitability for this quarter largely because SpaceX heavily discounted its compute costs for May and June. When you look at the actual foundational economics, they remain stubbornly unprofitable. Combine that with a CFO affidavit from March that makes reported revenues incredibly difficult to reconcile, and the narrative cracks.
Fortunately, the Financial Times did some actual digging into the hyperscalers' Q1 2026 numbers, and they found a glaring anomaly that should be getting far more scrutiny than the headline capex figures.
Per a quoted Goldman Sachs note, Alphabet and Amazon generated a staggering $53 billion in "other income" in Q1 2026 alone. That accounted for nearly 60% of those two companies' net income for the quarter, and 34% of the collective $155 billion pocketed by the five largest hyperscalers. It represents the group's largest share of earnings attributable to "other income" in over a decade.
Where did that magical $53 billion come from? $49 billion of it was explicitly due to equity markups in private companies.
In other words: more than half of Alphabet's net income last quarter did not come from selling targeted search ads or provisioning enterprise cloud infrastructure. It was an accounting markup on its Anthropic stake. Amazon's situation was nearly identical. As Anthropic's paper valuation skyrocketed from $183 billion last September, to $380 billion in February , and now flirts with an astronomical $850 billion, the hyperscalers get to book those implied valuation gains straight to their public bottom lines.
The circle is now fully closed:
- Hyperscalers pump tens of billions of dollars into private AI labs.
- Those massive investments artificially inflate the labs' private valuations.
- The hyperscalers write up the value of their equity stakes, booking billions in "other income" to look highly profitable to public stock investors.
- The AI labs turn right around and hand that exact same investment cash back to Google Cloud, Azure, and AWS to pay for raw compute.
The Financial Times labeled this dynamic “comically codependent.” That feels incredibly generous. It looks less like comedy and more like a high-stakes valuation carousel.
Remember, the "operating income" being paraded around is strictly Non-GAAP, a standard that conveniently ignores the absolute elephant in the room: massive capital expenditure. The broader market still hasn't fully registered that these AI labs are booking their foundational model training as capex rather than opex—effectively hiding the single most expensive part of their entire business model from the operating expense line.
If these frontier labs were genuinely anywhere near true commercial profitability, why are they perpetually pitching massive new multi-billion dollar funding rounds, issuing fresh equity, and diluting their current shareholders into oblivion?
There are plenty of lean technology companies with a fraction of these revenue figures and zero operating income that successfully complete traditional IPOs. So why hasn’t Anthropic pulled the trigger on a public listing? Unless, of course, their actual GAAP bookings look a whole lot different than the pitch decks they hand out to private investors.