Microsoft logs $3.2B from Anthropic investment, but OpenAI was a mixed bag

Microsoft just printed $3.2 billion in quarterly earnings from an AI investment that barely existed eight months ago. Let that sink in. The company invested $5 billion in Anthropic last November. By June, that stake had supposedly generated a gain worth 64 percent of the principal. In one quarter. The math only works if you ignore the circularity: Anthropic agreed to buy $30 billion of Azure services as part of the same deal. Microsoft is effectively booking future cloud revenue as a current investment gain. That is not a return on capital. It is a revenue acceleration mechanism dressed up in venture capital clothing.

The market will cheer the headline number. Diluted EPS got a 33-cent boost. Analysts will model it forward. But the mechanism matters. Microsoft does not mark Anthropic to market every quarter — it only did so this time because the gain was too large to bury. That discretion tells you everything about the quality of the number. When the accounting serves the narrative, the narrative is suspect.

Contrast OpenAI. Microsoft owns roughly 27 percent of that lab, a stake built over years and billions. This quarter, it took a $600 million write-down. Seven cents of EPS evaporated. The financial press will frame this as a stumble. It is not. It is the most honest number Microsoft has reported on AI in months. Valuations fluctuate. The fact that Microsoft marks OpenAI quarterly — and eats the losses publicly — makes the Anthropic silence louder. You do not hide a mark-to-market process unless the marks inconveniently disagree with the story you are telling.

The full-year OpenAI gain of $5 billion looks respectable until you realize Anthropic nearly matched it in ninety days. That comparison is the real signal. Microsoft disclosed the Anthropic gain precisely because the disparity is embarrassing. One investment follows conventional accounting. The other follows a side letter that guarantees Azure consumption. The latter produces better optics. The former produces actual economics.

Revenue share from OpenAI remains a black box. Microsoft collects payments but refuses to quantify them. That omission is deliberate. If the revenue share were material, Microsoft would trumpet it. The silence suggests the cash flow is either negligible or volatile enough to undermine the "strategic partnership" framing. Meanwhile, the $30 billion Azure commitment from Anthropic is not revenue — it is a contracted minimum. Microsoft will recognize it over years as services are consumed. Booking a chunk of it now as an investment gain is aggressive accounting, not operating leverage.

Microsoft earned $35.8 billion this quarter. $90 billion in revenue. $133.7 billion for the year. The Anthropic gain represents 3.5 percent of quarterly net income. The OpenAI write-down represents 1.7 percent. Both are noise. The only reason we are discussing them is that Microsoft chose to spotlight one and downplay the other. That choice reveals the company's anxiety about its AI narrative. It needs the market to believe its cloud moat is widening because the model builders are locked in. The Anthropic deal looks like proof. The OpenAI volatility looks like the opposite.

The deeper truth: Microsoft does not need either lab to win. Azure grows because enterprises need compute, not because a specific model provider signed a purchase order. The circular deals — invest in the lab, mandate the lab buy your cloud — are financial engineering, not strategic moats. They inflate the appearance of ecosystem lock-in while transferring risk to the balance sheet. When Anthropic's $30 billion commitment rolls through the P&L over the next decade, the gains will reverse. Microsoft knows this. The market should too.

None of this is fraud. It is GAAP-compliant narrative management. But journalists should not lap it up. The $3.2 billion gain is a derivative of a procurement contract. The $600 million loss is a derivative of a fair-value estimate. One is manufactured. The other is measured. Microsoft's profitability is real. Its AI investment returns are a hall of mirrors. Read the footnotes, not the headlines.