Microsoft’s AI-Lab Stakes Added $0.40 to Q4 EPS — and Showed How Circular Financing Works Inside a Hyperscaler’s Income Statement

Microsoft’s FY26 Q4 results contain something rarer than a revenue beat: a hyperscaler’s net income carrying the AI economy’s circular financing structure inside it — in the form of non-cash, mark-to-market gains on two private lab stakes.

What Happened

Buried inside Microsoft’s otherwise strong fiscal fourth-quarter 2026 results — net income of $35.77 billion, diluted EPS of $4.81, up from $3.65 a year ago — is a set of line items that say as much about the AI economy’s structure as any revenue figure. Per Microsoft’s FY26 Q4 8-K filing, the company recorded a $3.2 billion gain on its Anthropic investment (adding approximately $0.33 to diluted EPS) and a net gain of $480 million on its OpenAI investments (approximately $0.07 to EPS) in the quarter alone. Together, those two non-operating marks contributed roughly $0.40 of the quarter’s $4.81 EPS.

The full-year picture sharpens the signal further. For all of FY26, the OpenAI stake contributed approximately $4.963 billion in gains, or $0.67 per diluted share — a striking turnaround from Q4 FY25, when the same position was a net loss of $1.575 billion, or $0.21 per share. The reversal reflects the surge in OpenAI’s valuation and changes in accounting for its restructuring, not a change in Microsoft’s operating relationship with the lab. Net discrete items added approximately $0.27 to EPS on top of the investment marks.

The Anthropic context makes the structure explicit. Microsoft invested $5 billion into Anthropic in November 2025 as part of an arrangement in which Anthropic simultaneously committed to purchasing $30 billion of Azure cloud services. The rising valuation of that same lab — which is widely reported to be advancing toward a public offering — is now booking a $3.2 billion paper gain back onto Microsoft’s income statement in a single quarter.

The Structural Read

Two things live inside these numbers, and they are worth separating. The first is an earnings-quality question. Approximately $0.40 of Microsoft’s $4.81 in quarterly EPS came from non-operating investment marks on AI-lab stakes. The underlying Azure business — explored in depth in the Azure Q4 FY26 analysis — is running well. But headline EPS, as reported, is flattered by items that do not recur on a predictable schedule and do not represent cash collected from customers. That is not a criticism of the results; it is a reading instruction.

The second thing is more structurally interesting: circular AI financing has arrived on a hyperscaler’s income statement. The Anthropic position is the clearest single example yet of a pattern we have tracked across the AI infrastructure stack — the hyperscaler funds the lab, the lab buys the hyperscaler’s compute, and the lab’s rising valuation flows back as a gain to the hyperscaler’s net income. The loop is closed, and all three roles — vendor, investor, customer — sit with a single counterparty. The Nvidia–SSI dynamic and Google’s TPU backstop arrangements follow the same architecture at different points in the stack.

The year-over-year swing on OpenAI — from a $1.575 billion net loss to a $4.963 billion full-year gain — is directionally consistent with OpenAI’s reported revenue acceleration (see the OpenAI ARR analysis), but the mark is a valuation judgment on a private stake, not booked customer revenue. The durable point is the structure, not the scoreboard.

Three Implications