Microsoft

Microsoft (NASDAQ: MSFT) sells cloud infrastructure, productivity software, and Windows. Since 2019 it has also been OpenAI’s largest financial backer — an investment that by October 2025 stood at roughly $135 billion for a 27% stake in a restructured OpenAI Group PBC, giving Microsoft commercial access to a frontier AI lab’s models without building one from scratch.

DeadRisk carries Microsoft as a mag7, big-tech name inside The AI Capex Supercycle — the deepest corpus we hold on any single company, and not a stock-picking exercise. The number worth reading here isn’t the capex line every hyperscaler now reports each quarter. It’s what Microsoft’s own record says about the company it depends on most: the OpenAI contract was renegotiated three separate times between October 2025 and May 2026, Microsoft started replacing OpenAI’s models with its own inside Excel, Outlook, PowerPoint, and Bing, and by March 2026 Microsoft was weighing legal action against OpenAI itself. The paper trail runs earlier than any of it: Microsoft’s 2024 annual report had already named OpenAI a competitor, eighteen months before the coverage record caught up.

The filing came first. In its annual report for fiscal 2024, Microsoft added OpenAI as a competitor — alongside its usual search and advertising rivals — a month after OpenAI showed off its own SearchGPT prototype. The record barely reacted. The two companies kept behaving like partners for another year and a half.

Underneath, the contract was already straining on terms nobody outside it could see. By June 2025, Microsoft was pushing to remove the “AGI clause” — language that let OpenAI cut off Microsoft’s access to its most advanced models the moment OpenAI’s board declared artificial general intelligence reached. By September, the clause survived, only modified, not removed. The same month, Microsoft moved to add Anthropic’s Claude inside Office 365 after finding Claude Sonnet 4 beat GPT-5 on some tasks — a second model supplier inside Office 365, diversifying away from sole reliance on OpenAI.

The restructuring landed October 28, 2025: Microsoft’s stake settled at $135 billion, about 27% of OpenAI Group PBC, down from 32.5%, with access to OpenAI’s models through 2032 — including systems built after AGI — while OpenAI won the right to build products with other partners for the first time. Two days later the cost showed up on Microsoft’s own income statement: the stake cut $3.1 billion from net income, the equity-method loss share of a lab still burning cash. Three months on, Microsoft rolled out the Maia 200, its second in-house AI chip, on January 27, 2026. Two days later, Microsoft reported Q2 capex up 66% to $37.5 billion — a beat that still sent MSFT down 10.0%, from $481.63 to $433.50.

March 2026 compressed the pattern into a week. On the 17th, Microsoft named Jacob Andreou, a 33-year-old former Snap executive, to run a newly consolidated Copilot team, freeing Mustafa Suleyman to focus only on building models. The next day, Microsoft weighed legal action against Amazon and OpenAI over whether AWS could sell an “OpenAI Frontier” product without breaching Microsoft’s own agreement with OpenAI — a single, still-unconfirmed report pitting Microsoft against both its partner and a cloud rival over the same clause. A month later, OpenAI’s revenue chief said the Microsoft partnership had “limited our ability” to reach clients through AWS’s Bedrock, and pointed to the Amazon relationship instead.

Late April brought the second renegotiation: Microsoft agreed to stop paying OpenAI a revenue share, and OpenAI won the right to serve products through any cloud provider — Azure exclusivity, ended; the same amendment removed the AGI clause entirely, while Microsoft kept its contractual right to OpenAI’s models through 2032 regardless. Two weeks later, the third renegotiation ran the other direction: new terms capped OpenAI’s payments to Microsoft at $38 billion — down from a figure that could have reached $135 billion by 2030 if OpenAI hit its growth targets. By July, the in-house swap was live: Microsoft began replacing OpenAI and Anthropic models with its own MAI models inside Excel and Outlook to cut AI costs, then confirmed the same swap for image generation in PowerPoint and Bing, with Suleyman citing MAI models as roughly 85% cheaper to run. On July 30, fiscal Q4 closed the loop: the OpenAI stake was marked down about $600 million for the quarter, while the Anthropic stake gained $3.2 billion — the same earnings call that reported capex up 70%+ to $41 billion, even as an accounting change lowered the 2026 capex forecast to $175 billion from $190 billion with spending plans otherwise unchanged. MSFT rose 15.5% the next session, from $390.54 to $451.10.

TEXXR’s edge-arc tool tracks not just how much gets said about a company but what kind of thing gets said — and on Microsoft, the kind changed abruptly. Microsoft’s predicate-labeled coverage-graph edges carried no competitor tag across the eight quarters from 2024Q1 through 2025Q4 — zero of 1,654 — then 10 of 109 in 2026Q1, 9%, naming both OpenAI and Anthropic. Financial edges rose 22 points over the same span, from 15% to 37%. Launch edges fell 10 points, from 33% to 23%; regulatory fell 8 points and partnership fell 5. The same tool’s regime-detection pass, run on the identical quarterly window, flags 2026Q1 itself as a structural break, not a drift: semantic drift of 0.074 against those eight prior quarters’ average of 0.025, with network turnover (Jaccard 0.57) dropping Meta and Xbox Game Pass Ultimate out of Microsoft’s most-connected entities and pulling in Maia 200 and “data centers” instead. Read plainly: the graph’s most-connected list used to include a rival’s product Microsoft doesn’t make. Now it includes the chip Microsoft built to cut its own hardware costs — a different kind of independence than the model swap already underway in Excel, Outlook, PowerPoint, and Bing, and not itself a substitute for OpenAI’s models.

One thread in the graph needs a caution the Oracle dossier already put on record for a different company: extraction volume is not event count. The Financial Times report on Microsoft’s weighing legal action against Amazon and OpenAI, filed March 18, 2026, shows up as nineteen near-identical edges in TEXXR’s graph — same date, same predicate, same two paragraphs of summary reworded by the extraction pass. That is one report registering with unusual force, not nineteen separate disputes. It sits, unresolved as rumored, next to a confirmed edge from three weeks later in which OpenAI’s own revenue chief said the Microsoft relationship had limited its reach to Amazon’s customers — corroboration of the tension from the other side of the table, without confirming Microsoft ever filed anything.

Positioning: how much of MSFT is sold short, on the record brokers file twice a month. It lags by design — read it as where the crowd stood, not where it stands.

Signals in the coverage record worth tracking, not trade ideas:

• Whether the MAI swap reaches Copilot’s core reasoning model. The record so far shows Microsoft replacing OpenAI in Excel, Outlook, PowerPoint, and Bing image generation — supporting surfaces. Copilot’s primary chat and reasoning engine, the highest-usage surface, is not yet confirmed swapped; the 30 million-plus paid-seat figure Microsoft reported for Q4 FY26 is for the Microsoft 365 Copilot subscription as a whole, not confirmed usage of this engine specifically.
• Whether the $38 billion cap holds. It already replaced a formula that could have run to $135 billion by 2030. A fourth renegotiation, in either direction, would be the cleanest read on how settled the new terms actually are.
• Whether the March 2026 legal threat against Amazon and OpenAI is ever confirmed as filed litigation, or stays rumored the way it has sat since the Financial Times first reported it.
• Whether the 2026 capex accounting change shows up in the balance sheet mix — leases versus owned infrastructure — in Microsoft’s next 10-Q, or turns out to be presentational, given management’s own statement that spending plans are otherwise unchanged.
• The competitor-edge share past 9%. Whether it keeps climbing as MAI models ship into more products, or plateaus once the in-house swap is complete.

Microsoft lives or dies by the same constraint as these names — small equal-weighted groups, one constraint each. Not investment products: see all categories.