| Ticker | Name | Research | View | What's said | Source |
|---|---|---|---|---|---|
| MSFT | Microsoft | QT · SA · STK · FA | Neutral | The main story — the "open-weight hedge": a multibillion-dollar closed-model alliance with OpenAI on one side, Azure as the indispensable open-weight deployment platform on the other. "Microsoft does not need to predict which model architecture wins. It wants to provide the compute, governance, and security layer underneath all of them." Q4 FY26: revenue +18% Y/Y to $90.0B (a $2.4B beat), gross margin 67% (−1pp), operating margin 45% (flat), non-GAAP EPS $4.74 ($0.50 beat). Segments — Productivity & Business Processes +14% to $37.8B (M365 Copilot, E5, early E7); Intelligent Cloud +32% to $39.3B on 43% Azure growth, above the 39–40% cc guide and >$1B above the segment high end, on better fleet efficiency and faster capacity landings; More Personal Computing −4% to $12.9B as weak Windows OEM and Xbox (content and services −10%, third straight decline, ~3% margin plus impairments) outweighed +10% Search advertising. Copilot passed 30M paid seats (from 20M in Q3, 15M in Q2 — doubled in six months) at under 8% penetration, with >50,000-seat customers up over 7x Y/Y and weekly engagement "on par with Outlook and Teams," now billed as "seat plus usage." Backlog: commercial RPO +84% to $678B, more than twice annual revenue — OpenAI is still the largest disclosed component, but all sequential growth came from non-frontier customers and RPO rose 25% excluding OpenAI. The cost: cash CapEx more than doubled to $35.8B, FY26 total $115.9B (+80%), operating cash flow +30% to $55.4B yet free cash flow −23% to $19.6B; Microsoft Cloud gross margin −3pp to 65% and Intelligent Cloud to ~57%, though segment operating margin held near 41% as opex rose only 10% against 32% revenue growth. Quality-of-earnings flag: a $480M OpenAI gain was excluded from non-GAAP but a $3.2B Anthropic gain was left in. Efficiency offsets: 40% better performance per watt on Maia 200, 89% lower GPU cost in Dynamics 365 (MAI-Voice-2-Flash), up to 84% lower in PowerPoint (MAI-Image-2.5). Q1 Azure guided to ~45% cc with H1 growth accelerating and demand still exceeding capacity. Accounting flag: from FY27, data-centre and office useful life extends from 15 to 25 years and more future data-centre leases shift from finance to operating leases — which sit outside reported CapEx. A disclosed author holding. (Analysis, not a stance call.) | article ↗ |
| NVDA | NVIDIA | QT · SA · STK · FA | Neutral | The campaigner. Jensen Huang used his first-ever X post to enter Washington's open-weight fight, sharing an open letter — initially 25 signatories including Microsoft, Meta and Palantir — urging policymakers to preserve the development and deployment of open-weight models. NVIDIA then launched the Open Secure AI Alliance, a coalition of nearly 40 companies building defences against AI-powered cyberattacks, with Microsoft, SpaceX and IBM as founding members (Anthropic, OpenAI and Meta notably absent). NVIDIA is also one of the model providers listed in Microsoft Foundry. The commercial logic is unstated but obvious: open weights broaden who buys GPUs. A disclosed author holding. (Referenced; not a stance call.) | article ↗ |
| GOOGL | Alphabet (Google Cloud) | QT · SA · STK · FA | Neutral | The benchmark that keeps Microsoft's reacceleration in perspective: Azure's 43% "still trailed Google Cloud's extraordinary 82% segment growth." On policy, Google was initially absent from the open-weight letter but Sundar Pichai later endorsed the effort on the company's behalf — leaving Anthropic as the sole major holdout. A disclosed author holding. (Referenced; not a stance call.) | article ↗ |
| META | Meta Platforms | QT · SA · STK · FA | Neutral | One of the 25 original signatories of the open-weight letter — consistent with Llama being an open-weight family, and Llama is among the model families Microsoft Foundry wraps with a common interface. Notably absent from NVIDIA's Open Secure AI Alliance alongside Anthropic and OpenAI. A disclosed author holding. (Referenced; not a stance call.) | article ↗ |
| PLTR | Palantir Technologies | QT · SA · STK · FA | Neutral | Named among the 25 initial signatories of the open-weight letter (with Microsoft and Meta) that Jensen Huang amplified — placing the defence/government-facing data platform on the side of keeping downloadable, inspectable model weights legal, which is also the precondition for the disconnected and sovereign deployments its customer base requires. A disclosed author holding. (Referenced; not a stance call.) | article ↗ |
| IBM | International Business Machines | QT · SA · STK · FA | Neutral | A founding member of NVIDIA's Open Secure AI Alliance — the ~40-company coalition building tools to defend against AI-powered cyberattacks — alongside Microsoft and SpaceX. Places IBM inside the open-weight camp and the AI-security build-out that Microsoft is simultaneously monetising with its own first-party stack. (Referenced; not a stance call.) | article ↗ |
| SPCX | SpaceX | QT · SA · STK · FA | Neutral | Named as a founding member of NVIDIA's Open Secure AI Alliance, with Microsoft and IBM — an early public alignment of the Musk-adjacent infrastructure complex with the open-weight/AI-cyberdefence camp. (Referenced; not a stance call.) | article ↗ |
| OpenAI | OpenAI (private) | — | Neutral | Both Microsoft's biggest partner and its biggest concentration risk. The closed-model archetype (weights locked behind an API, with the provider controlling guardrails, infrastructure and pricing); initially absent from the open-weight letter, it joined after Sam Altman said he wanted the US to lead in both open and closed models — but stayed out of NVIDIA's Open Secure AI Alliance. Commercially it is the largest disclosed component of Microsoft's $678B commercial RPO, and the article's key diagnostic is that RPO grew 25% excluding OpenAI with all sequential growth coming from non-frontier customers — the concentration isn't worsening at the margin. GPT-5.4 is also the escalation target in Microsoft's new cyber router (the hardest ~10% of tasks); and Microsoft excluded a $480M OpenAI investment gain from adjusted EPS. (Referenced; not a stance call.) | article ↗ |
| Anthropic | Anthropic (private) | — | Neutral | "The clearest holdout." Anthropic did not sign the open-weight letter and is absent from NVIDIA's alliance, arguing it does not want to ban open models but supports stricter chip controls, action against industrial-scale distillation, and mandatory safety testing for all capable systems — the sharpest counter-position in the fight. It is also a Foundry-listed model provider (Claude wrapped by Microsoft's common APIs and governance), and the source of the $3.2B investment gain Microsoft left inside adjusted EPS while excluding a $480M OpenAI gain — the quarter's quality-of-earnings flag. (Referenced; not a stance call.) | article ↗ |
| Mistral | Mistral AI (private) | — | Neutral | The sovereignty play made concrete: Microsoft expanded the partnership by several billion dollars — Medium 3.5 and OCR 4 land in Microsoft Foundry (Medium 3.5 also in Copilot Studio), Azure supports cloud, cloud-connected and fully disconnected deployments, and Microsoft committed multibillion-dollar spending on Mistral's expanding European GPU capacity. The pitch: open weights with an enterprise wrapper — banks, governments, healthcare and defence customers get frontier AI without sending every query to a closed, US-hosted API, while the deployment stays inside Azure. (Referenced; not a stance call.) | article ↗ |
| Moonshot AI | Moonshot AI (private) | — | Neutral | The distillation flashpoint. US accusations that the Chinese lab used a leading American model's outputs to improve Kimi are "the same concern" driving closed-model providers to call unrestricted model distillation intellectual-property theft — while open-weight supporters see it as competition that lowers costs and broadens access. The regulatory outcome matters directly to Microsoft: a narrow crackdown on industrial-scale distillation is manageable, but broader restrictions on open weights or Chinese models "would weaken Foundry's breadth and the sovereignty pitch." (Referenced; not a stance call.) | article ↗ |
"View" here is referenced/neutral — App Economy Insights is financial-analysis journalism; this is an earnings/strategy breakdown, not a set of buy/sell calls (BUY/SELL/HOLD ratings are shared only with App Economy Portfolio members; the author's disclosed holdings among these names are MSFT, NVDA, GOOG, META and PLTR, plus AMD and AMZN which aren't discussed here). Research: QT Qualtrim · SA Seeking Alpha · STK Stock Analysis. The "Source" links open the newsletter (no per-name timestamps — it's a written post). Named only in passing and not given rows: DeepSeek, xAI, Cohere and Hugging Face (model providers listed in Foundry), Maia 200, MDASH, Project Perception, Kimi, Entra, and the individuals quoted (Jensen Huang, Satya Nadella, Sam Altman, Sundar Pichai).
A jargon-free summary of the read behind each name. (Plain-language companion to the table above; renders on each ticker's consolidated page.)
There's a fight in Washington over whether AI companies should be allowed to publish their models' "weights" — the trained internals that let anyone download a model and run it on their own computers, instead of renting it from the owner over the internet. Microsoft has quietly taken both sides, and that's the point of this issue. It keeps a multibillion-dollar partnership with OpenAI, whose models stay locked away, and it is making Azure the best place to run the downloadable rivals. Microsoft doesn't need to guess which type wins; it wants to sell the electricity, plumbing and security underneath all of them.
The quarter was strong. Sales rose 18% to $90 billion and beat expectations, and Azure — the cloud-computing business, the one investors actually watch — grew 43%, better than the 39–40% management had guided. Copilot, the AI assistant bolted onto Office, added 10 million paying seats in three months to pass 30 million, though that's still under 8% of Microsoft's business customers, so there's a lot of runway. And the order book is extraordinary: $678 billion of contracted-but-not-yet-delivered revenue, more than twice what the company earns in a year.
Two cautions. First, the order book leans heavily on one customer, OpenAI — the useful check the article performs is that it still grew 25% with OpenAI excluded, and all the growth this quarter came from ordinary businesses, so the concentration isn't getting worse. Second, the cost: Microsoft spent $115.9 billion on data centres this year, up 80%, which is why free cash flow — the money left after building things — dropped 23% even though the business generated 30% more cash. Profits are also being flattered a little by investment gains: Microsoft removed a $480 million gain on its OpenAI stake from its adjusted profit, but kept a $3.2 billion gain on its Anthropic stake.
The one item worth writing in a diary: from next fiscal year, Microsoft will assume its data centres last 25 years instead of 15 (which spreads the cost over more years and lifts reported profit), and it will sign more leases in a form that legally sits outside the capital-spending line. The money still goes out the door; it just becomes harder to see. As the article puts it, "the spending remains, but the accounting optics improve." The author owns the stock; this is analysis, not a recommendation.
Nvidia makes the chips nearly all AI runs on, so anything that widens the number of companies training and running their own models is good for it. That's the subtext here: CEO Jensen Huang used his first-ever post on X to back an industry letter — signed by Microsoft, Meta, Palantir and 22 others — asking Washington not to restrict "open-weight" AI models, the kind any business can download and run on hardware it owns.
Nvidia followed up by launching the Open Secure AI Alliance, a roughly 40-company group building defences against AI-powered cyberattacks, with Microsoft, SpaceX and IBM as founding members. Notably absent: Anthropic, OpenAI and Meta. The line-up is a decent map of who benefits from open models spreading and who prefers them contained. Referenced in passing, not a stance call.
Mistral is a French AI lab whose selling point is that it publishes its models openly, so customers can run them on their own machines rather than piping every question to an American company's servers. Microsoft just expanded its partnership with it by several billion dollars: Mistral's newest models go into Microsoft's model marketplace and its Copilot-building tools, and Microsoft committed to buying capacity on Mistral's growing European GPU fleet.
Why does Microsoft pay for a rival's models? Because of who's buying. Banks, hospitals, governments and defence agencies want modern AI but can't send sensitive data over the public internet. Azure can now run these models in a normal cloud, a partly connected setup, or an entirely disconnected one — a private, air-gapped installation. That's the "sovereignty" pitch: you keep control of your data and your jurisdiction, and Microsoft still collects the bill. Referenced, not a stance call.
Anthropic (the maker of Claude) is the odd one out in this fight. It refused to sign the industry letter defending downloadable AI models and stayed out of Nvidia's new security alliance. Its position isn't that open models should be banned — it wants tighter controls on advanced chips, enforcement against "industrial-scale distillation" (rivals cheaply cloning a model by training on its answers), and compulsory safety testing for any sufficiently capable system.
It shows up in Microsoft's numbers too. Microsoft owns a stake in Anthropic, and a $3.2 billion paper gain on it was left inside Microsoft's adjusted earnings while a smaller $480 million gain on its OpenAI stake was stripped out — meaning some of Microsoft's headline profit came from a share-price mark, not from selling software. Anthropic's models are also offered on Microsoft's own marketplace, which is the whole "we host everyone" strategy in miniature. Referenced, not a stance call.
OpenAI is simultaneously Microsoft's most important partner and its biggest single risk. It is the archetype of a "closed" AI company: the model lives behind an internet connection it controls, and it sets the guardrails, the hardware and the price. It initially skipped the industry letter defending open models, then signed on once Sam Altman said America should lead in both approaches.
Financially, OpenAI is the largest identified chunk of Microsoft's $678 billion of contracted future revenue — wonderful visibility, uncomfortable concentration. The article's useful check: strip OpenAI out and the rest of that backlog still grew 25%, and every dollar of this quarter's increase came from ordinary corporate customers rather than AI labs. So the dependence is real but no longer deepening. OpenAI's GPT-5.4 also plays a supporting role in Microsoft's new security product, handling only the hardest 10% of tasks that Microsoft's own cheaper model can't. Referenced, not a stance call.
Key points & figures extracted from the paid App Economy Insights newsletter (in transcript.txt) for personal study. Not investment advice. © App Economy Insights for source material.