Microsoft told investors on Tuesday afternoon that its license to OpenAI's technology is no longer exclusive through 2032 and that it will stop paying revenue share to OpenAI; OpenAI continues paying Microsoft through 2030, capped, and Microsoft retains roughly 27 percent equity in the OpenAI for-profit, valued at about $135 billion at the latest mark. [1] The disclosure landed on Apr. 28, the day before Microsoft's fiscal Q3 2026 earnings report. The Q3 print, released after the close on Apr. 29, showed revenue of $82.9 billion (up 18 percent), Azure growth of 40 percent, an AI run rate of $37 billion, and commercial remaining performance obligations of $627 billion — a 99 percent year-over-year jump. [1] Microsoft did not name OpenAI's share of the RPO. The Apr. 29 paper's Microsoft and Meta major on the hyperscaler capex margin test asked whether the print would pass the margin test; today's tape — META down 9.82 percent, GOOG up 5.6 percent, MSFT soft after-hours — is the answer.

The companion question the paper's Apr. 29 brief on the OpenAI share of MSFT's RPO posed is the one the contract revision answered before the print. Microsoft economically consolidated rather than separated. It is no longer paying OpenAI a revenue share — a change that improves the carrier-of-record economics for Azure-on-OpenAI workloads — but it remains the largest single equity holder in OpenAI at roughly 27 percent and still carries the right to OpenAI capacity through 2032. The trade is exclusivity for economics. What MSM coverage has soft-pedaled, and what the paper's ai-state-power thread tracks, is that Microsoft is now positioned to sell Anthropic, Mistral, or xAI capacity at scale on the same Azure regions and continue collecting on its OpenAI equity stake. Wedbush analyst Dan Ives, the most-quoted bull on the move, framed it as "more room to win." [1] Ed Zitron, the most-quoted bear, called it "an economic divorce performed at the altar."

The day's third anchor is the Apr. 29 major on Cerebras naming OpenAI a customer, a lender, and a shareholder on the same page. The same OpenAI counterparty now sits inside four simultaneous instruments — Microsoft's contract revision (here), Cerebras's S-1 disclosures (yesterday's major), the Tumbler Ridge tort docket (today's article 04), and the Florida criminal probe disclosed in the BBC's late paragraph yesterday. The paper's ai-state-power thread argued through April that the disclosure register would tighten; today the register is dockets, not disclosures.

What the print actually said

Microsoft's Q3 came in above consensus on revenue and EPS. Revenue of $82.9 billion beat the $81.27 billion consensus by $1.6 billion. [1] Azure's 40 percent growth was 200 basis points above consensus — the third consecutive quarter of acceleration. The AI services run rate of $37 billion, disclosed for the first time at this granularity, implies trailing-twelve-month AI revenue of roughly $34 billion and a sequential-quarter growth rate of 11 percent. The Intelligent Cloud segment — the bucket that contains Azure — grew to $32.9 billion, up from $29.7 billion in Q2.

Operating margin expanded by 160 basis points to 47.1 percent. Free cash flow margin, however, slipped from 9.3 percent to 7.2 percent — the capex translation Apr. 29's paper flagged. Microsoft's capex run rate is now $115 billion annualized; CFO Amy Hood guided to $130 billion for fiscal 2027 on the call. Both numbers are above the consensus going in. The market response was muted because the OpenAI exclusivity revision had already absorbed the bullish surprise.

The non-GAAP carve-out is the line that names the contract change without naming it. Microsoft disclosed an OpenAI-related non-GAAP loss of $14 million for the quarter, compared with a $583 million loss a year ago. [1] That collapse — from $583 million to $14 million on the same quarter year-over-year — is what stopping revenue share looks like in the income statement. The remaining $14 million is the residual on contract obligations Microsoft still owes through the 2030 cap; that figure should approach zero by the FY2027 prints.

The $627 billion RPO number, taken alone, is the most consequential single datum in the print. RPO, in Microsoft's disclosure framework, is contracted revenue not yet recognized — what one analyst called "the visible portion of the next four years." Up 99 percent year-over-year, the RPO is the strongest forward-revenue signal Microsoft has produced since the Q4 2024 print. The undisclosed component — what share is OpenAI's committed Azure spend — is the metric the paper has tracked for three editions. Today's print left it undisclosed.

That non-disclosure is the editorial event. SEC accounting standards do not require Microsoft to break out RPO by counterparty. They do require Microsoft to disclose any single customer that represents more than 10 percent of total revenue. Microsoft has not historically had to make that disclosure for OpenAI because OpenAI's Azure spend has run below the 10 percent line. With a $37 billion AI run rate and an OpenAI counterparty whose committed Azure spend is, on industry estimates, between $80 billion and $120 billion across multiple years, the question of when the 10 percent threshold gets crossed is not an academic one. The contract revision — stopping revenue share — accelerates the crossing because every dollar of OpenAI capacity now flows through MSFT's revenue line without a kickback.

The Wedbush read versus the Zitron read

Dan Ives at Wedbush, in a Wednesday morning note, told clients the new arrangement gives Microsoft "years of access to OpenAI's tech, lifts the friction between the two companies, and lets MSFT keep more of the Azure revenue." [1] His price target on the stock remained $625, the high end of the Street. Goldman maintained a Buy and a $600 target. The 49-analyst consensus target of $571.95 implies a 35 percent upside from the post-print close, on Yahoo Finance's compilation. [1]

The bear case, voiced most clearly by Ed Zitron and Gary Marcus on Twitter through Wednesday and Thursday, has three components. First, the contract change frees OpenAI to negotiate better deals with Amazon and Google, which weakens Azure's pricing power on AI capacity over time. Second, Microsoft is consolidating economically with a counterparty that simultaneously faces the Tumbler Ridge tort docket, a Florida criminal probe, and the Cerebras-S-1 governance questions about its multi-instrument exposure. Third, the timing — Apr. 28, the day before the print — looks like an effort to control the news cycle. The bear case does not say the contract revision is bad. It says the timing of its disclosure was managed.

The convergence the X register is tracking is that both readings can be true. Microsoft can win — operating margins expanding, AI run rate up — and Microsoft can also be diversifying away from a counterparty whose legal and governance exposure is rising. The next FQ4 print, due in late July, will say which weighting matters more. If the OpenAI carve-out approaches zero and the RPO continues to compound at 80-plus percent, the bull case wins. If Anthropic, Mistral, and xAI capacity sales begin showing up as named partners on the Azure marketplace at scale, the bear case wins. Both can occur on the same earnings call.

What the contract revision did not change

Three things remained the same. Microsoft retains roughly 27 percent equity in the OpenAI for-profit, currently marked at about $135 billion. That equity stake is the largest single line in Microsoft's marketable-securities portfolio. Second, Microsoft retains its preferred-customer access to OpenAI's frontier models through 2032 — meaning Azure customers can still buy the latest GPT capacity on Azure, just no longer with the implication that Azure is the only place to buy it. Third, OpenAI continues to pay Microsoft a capped license fee through 2030; that revenue stream survives.

What changed is the directional flow of cash. Before Apr. 28, Microsoft paid OpenAI a revenue share on Azure-on-OpenAI workloads. After Apr. 28, that flow stops. The OpenAI license fee in the other direction — capped through 2030 — survives. The net effect on FY2026 income, by the disclosed carve-out, is roughly a $570 million swing to Microsoft on a quarterly basis (the $583 million prior-year loss versus this quarter's $14 million). Annualized, that is roughly a $2 billion improvement in OpenAI-related operating income.

The two contracts that did change have not been disclosed in their full text. The 8-K filed Apr. 28 names the categories — exclusivity termination, revenue-share termination — but not the financial mechanics. The fuller text will land in the 10-Q at the end of next week. Three Wall Street analysts told the paper they expect the 10-Q disclosure to show the OpenAI revenue-share line falling to zero in this quarter and to confirm that Microsoft's pro-rata share of OpenAI's losses, which had been included in the carve-out, will continue under the equity-method accounting that governs its 27 percent stake. The latter mechanism is why MSFT has been recording an OpenAI-related loss while OpenAI itself runs at a loss.

What the next print will surface

The FQ4 print, scheduled for late July, will be the first to show a full quarter without OpenAI revenue share. If commercial RPO continues at the 99 percent-plus pace, the AI run rate clears $42 billion, and the OpenAI carve-out approaches zero, the contract revision was a winner. If RPO decelerates and Anthropic, Mistral, or xAI named partnerships begin showing up in the marketplace disclosures, the contract revision was a hedge.

The question that will not be answered before then is the one the paper has tracked for three editions: what share of MSFT's $627 billion RPO is OpenAI's committed Azure spend? The answer, when it comes, will determine whether the contract revision was an exclusivity sale at par or below.

OpenAI declined to comment beyond a Wednesday-morning statement that the contract revision "reflects the maturation of the partnership." Microsoft's investor-relations team referred questions to the 8-K. The 10-Q lands next week.

-- THEO KAPLAN, San Francisco