On October 1, 2026, we convened the committee on Microsoft, hours after Bloomberg reported its best quarter for decades. Shares rose 37.5% from July through September, the best quarterly showing since 1998, adding $1 trillion to market capitalization. Two days earlier, OpenAI launched a workplace suite aimed squarely at Microsoft's core, from inside its own key AI partnership. The seats returned two Watch, one Avoid, one Hold. Zero Buys. Under our published rule, a name enters a basket only when at least three of four seats return Buy or Strong Buy. Microsoft got none.
The fact sheet every investor read
| Item | Figure or finding | Source | Date |
|---|---|---|---|
| Revenue | $331.84B, fiscal year ended June 30, 2026 | SEC EDGAR | October 1, 2026 |
| Net income | $133.75B, fiscal 2026 | SEC EDGAR | October 1, 2026 |
| Operating cash flow | $182.94B, fiscal 2026 | SEC EDGAR | October 1, 2026 |
| Free cash flow | $66.99B, fiscal 2026 | SEC EDGAR, computed | October 1, 2026 |
| Implied capex | $115.95B, operating cash flow minus free cash flow | Computed | October 1, 2026 |
| Capex share of operating cash flow | 63.4% | Computed | October 1, 2026 |
| Net cash | $72.90B | SEC EDGAR, computed | June 30, 2026 |
| Net margin | 40.3% | Computed | October 1, 2026 |
| Free cash flow margin | 20.2% | Computed | October 1, 2026 |
| Price and market cap | $512.90, $3.81T | Yahoo Finance, SEC share count | September 30, 2026 |
| Valuation | 28.5x trailing earnings, 1.76% free cash flow yield, 8.6x book | Computed | October 1, 2026 |
| Treasury comparison | 10-year yield 5.30%, up 10.4% in 30 days | Yahoo Finance | October 1, 2026 |
| Trading range | 80% of the way up the 52-week range, low $349.20 | Yahoo Finance | September 30, 2026 |
| The quarter | Shares up 37.5%, best since 1998, $1T of market cap added | Bloomberg | October 1, 2026 |
| The OpenAI suite | Space, Pages and collaborative slides, launched September 29, aimed at workplace software | TechCrunch | September 29, 2026 |
| Not available | Segment revenue, the Azure growth rate, capex guidance, depreciation, dividend rate, buyback pace, every term of the OpenAI partnership | Fact sheet finding | October 1, 2026 |
One honesty note on the sheet itself. The automated pull grabbed a 2010 SEC XBRL frame into its trailing-twelve-month line, which contaminated four derived numbers. We flagged all four as artifacts and recomputed them from the filed fiscal 2026 statements. The correction is part of the record.
The missing rows are a finding, not a gap. We did not fill them with estimates. That matters more than usual here, because the one number the whole bull case rests on, the return on $115.95B of annual capex, lives exactly in the rows no public document we could source provides.
How Microsoft makes money
Microsoft sells cloud capacity, workplace software, and gaming, to businesses of every size, governments, developers and consumers. Customers pay three ways: recurring per-seat subscriptions for software, usage-based fees for cloud, and one-time purchases for some consumer products.
The fiscal 2026 arithmetic: $331.84B of revenue became $133.75B of net income, a 40.3% net margin. Operating cash flow was $182.94B. Free cash flow was $66.99B, because implied capex of $115.95B consumed 63.4% of operating cash flow. Free cash flow was roughly half of net income. The company funds its entire build-out from its own cash engine, holds $72.90B of net cash, and Bloomberg reports it is the only one of the four big AI spenders whose annual free cash flow never turned negative.
Our analyst's sharpest warning cuts against the fortress framing: "the net cash frame measures the wrong liability set for a business that has become this capital intensive." About $275.69B of non-debt liabilities sit on the balance sheet, and the leases and capacity contracts an AI build-out depends on are not on the sheet at all. If some data center capacity comes through finance leases or third-party contracts, true capital intensity is higher than shown.
Segment revenue, the Azure growth rate, capex guidance, depreciation, and every term of the OpenAI partnership are not available. Named plainly: the committee reviewed the second most valuable company on earth without seeing one segment number.
The risks to Microsoft
Our analyst ranked the permanent impairment risks first.
| Risk | How it hurts | Severity | First thing to watch |
|---|---|---|---|
| AI capex fails to earn its cost of capital | $115.95B a year sinks into long-lived assets that depreciate through earnings for years with no matching revenue | High | Cloud growth slowing while capex holds, guidance not cut |
| AI disrupts the productivity franchise | OpenAI's suite replaces the incumbent or shrinks the paid-seat base it is priced on | High | Seat counts, price per seat, renewal rates |
| Dependence on and conflict with OpenAI | One party is key partner and direct competitor, under undisclosed terms | High | OpenAI moving compute, renegotiated terms, stake impairments |
| Antitrust or structural remedy | Forced unbundling cuts the cross-selling and bundling advantages | Medium, distant | Formal investigations on cloud licensing or the AI partnership |
| Earnings quality | EPS grew nearly 32% on about 18% revenue growth, and the bridge is not on the sheet | Medium | Net income outrunning operating income, other-income swings |
| Unmeasured obligations | $275.69B of other liabilities plus unknown lease and capacity commitments | Medium | Lease and purchase commitment notes in the 10-K |
The next two to five years carry different pressure points.
| Risk | How it hurts | Likely hit | First thing to watch |
|---|---|---|---|
| Expectations reset | EPS growth slows to about 11% this fiscal year from nearly 32%, against 72 analysts and zero sell ratings | High, to the multiple | Fiscal 2027 quarters below the 18% revenue consensus |
| Depreciation squeeze | Yesterday's capex becomes tomorrow's expense, compressing margins even if revenue delivers | High | Depreciation as a share of revenue |
| AI compute oversupply | New providers add capacity, prices fall, cloud growth slows from its four-year high | High | Compute price cuts, peers cutting capex |
| OpenAI's suite takes share | Workspace, documents and slides roll out and bite seat growth | Medium, scale unknown | Enterprise pilots, slower seat growth |
| Higher rates | A 5.30% Treasury against a 1.76% cash yield reprices the multiple with every basis point | Medium | The 10-year moving further above 5.30% |
Our analyst also flagged exposures the fact sheet never carried. These come from general knowledge, not from the sheet: the advanced AI accelerators in Microsoft's data centers are largely fabricated in Taiwan, the high-bandwidth memory beside them concentrates in a few Korean suppliers, data center growth is hostage to grid power and permits, a large share of revenue is earned outside the dollar, European and UK regulators have open scrutiny of cloud licensing and bundling, and the company has a long-running US tax dispute. The next sheet we build for this name should collect every one of them.
Investor A, quality and long-term value
He works on a 5 to 20 year horizon, and he split the business from the stock cleanly: "This is one of the best businesses in the world. But I pay $3.81T and get $66.99B of free cash flow back, a 1.76% yield." A 10-year Treasury pays 5.30%. In his terms, the gap only closes if roughly $116B a year of capex earns high returns, and nothing on the sheet shows him that return.
His structural read is the sharpest line of the session: "This used to be an asset-light software company. Today it carries the capital intensity of an infrastructure company, and the market still prices it at 28.5x trailing earnings."
He also declined to read the tape's mood swings as information. June was the worst month since 2000, then the best quarter since 1998. His view: the business did not change nearly that much in four months, and that was mood, not permanent capital loss.
His entry discipline is conditional arithmetic, not a forecast: trailing earnings around 20x or lower, and a free cash flow yield of 3% or higher, or today's yield plus disclosed evidence that the capex earns high incremental returns. His own reason he could be wrong: "Most of the capex may be growth capex with returns like the first cloud buildout. In that case true owner earnings sit much closer to $133.75B of net income than to $66.99B of free cash flow."
Verdict: Watch. Conviction 6 of 10.
Investor B, deep value and margin of safety
He works on a 1 to 5 year horizon, and he returned the session's only Avoid, on a basis that should unsettle anyone who skipped past it: "This may be a good business, but the supplied price does not offer demonstrable downside protection."
He refused to produce an intrinsic value at all. Normalized owner earnings are not available, because the sheet cannot separate the investment that must recur from the investment that is optional. A number he cannot decompose is a number he cannot discount, so there is no margin of safety to verify, at any price.
His sharpest line: "Positive free cash flow is insufficient evidence of capital discipline. The decisive question is how much investment must recur to preserve competitiveness and how much incremental cash it earns."
He named no entry price, on principle. A lower quotation alone would not resolve the missing economics. His own reason he could be wrong: today's spending could be building productive capacity whose cash returns arrive later, and if spending moderates while revenue keeps growing, today's free cash flow yield materially understates earning power.
Verdict: Avoid. Conviction 8 of 10.
Investor C, global macro and reflexivity
He works across days to years, and he objected to the setup rather than the company: "The market appears to be pricing the victory lap of an AI capex cycle at the exact moment the discount-rate regime has turned against the asset."
His crowding read is blunt. Of 72 analysts tracked by Bloomberg, only three lack a buy rating and none rate it a sell. The average target implies roughly 11% upside over twelve months, less than a third of what the stock just delivered in one quarter. His line: "Consensus is no longer an edge, it is the position."
His most structural point is about the partnership nobody can price: the buy-side is paying 25x for an OpenAI relationship whose economics, revenue share, compute commitments and equity terms are all not available on the sheet. In his words, nobody in the consensus can actually underwrite the single most important partnership in the thesis.
His critical environment is the long end, and his summary of it is four words long: yields up, verdict down, yields down, verdict up. His own reason he could be wrong: this may be early-cycle for AI, not late-cycle, in which case 25x forward with high-teens EPS growth is cheap and the crowded consensus is simply right.
Verdict: Watch. Conviction 6 of 10.
Investor D, growth and stock picking
He works on a 2 to 10 year horizon, and he was the only seat that called the price fair. His category is Stalwart: roughly 25x forward earnings against a 27x ten-year average is not excessive for an 18% grower. His verdict: current valuation, 28.5x trailing and a 1.76% free cash flow yield, fully captures this momentum.
His concern is the same capex line the others refused. $115.95B consuming 63.4% of operating cash flow leaves minimal room for error, with consensus expecting EPS growth to slow to roughly 11% this fiscal year from nearly 32%.
His variant perception is the OpenAI paradox. The market views OpenAI as Microsoft's commercial growth engine, while OpenAI "is actively building an alternative operating environment for workplace collaboration." If daily work moves inside agent workspaces, "Microsoft's core per-seat productivity suite could face pricing pressure and unbundling risks."
His entry interest sits below 20x forward earnings or above a 3.5% free cash flow yield, with capex stabilizing below 40% of operating cash flow. His own reason he could be wrong: if corporate IT insists on running AI inside Microsoft's compliance, identity and security perimeter, OpenAI's suite expands Azure consumption instead of unseating Microsoft 365, and the capex becomes an enduring barrier.
Verdict: Hold. Conviction 6 of 10.
The environments that decide Microsoft
| Environment | How the business fares in it | Seats that called it critical | Which way it moves them |
|---|---|---|---|
| Long-term rates | The business is insulated, $72.90B net cash. The valuation is not: a 1.76% cash yield reprices with every basis point at 5.30% | C named it the one critical environment, A weighted it heavily | C: yields up, verdict down. A: it is why he demands a margin of safety |
| Enterprise AI and cloud demand | Strong now, fastest cloud growth in four years, about 18% fiscal 2026 growth | D named it the one critical environment | D: acceleration upgrades him to Buy, deceleration down to Avoid |
| Capex intensity and its returns | $115.95B, 63.4% of operating cash flow, returns not available | A, B and D each named this their weakest link | Disclosed high returns move A and B up. Guidance that crushes the cash line moves D down |
| Workplace software competition | OpenAI launched a direct rival from inside the partnership | D and A flagged it, C noted June's derating as proof the crowd prices this fear violently | Documented seat loss moves D from Hold to Avoid |
| The dollar and regulation | A rising dollar is a translation headwind, regulation is headline only | C noted both, no seat made either critical | Not verdict-moving on current evidence |
The seats disagree about which environment decides the verdict, not just about the verdict. C says the discount rate moves the answer more than the capex does. A, B and D say the capex return moves it more than anything. The same strong cloud print cannot settle both questions, which is why the same facts produced four different tests.
What has to be true for Microsoft
The assumption chain, top down, marked against the sheet:
- Rates stabilize or fall. CONTRADICTED, the 10-year at 5.30% and rising. C called this his weakest link, the one link directly contradicted by live data.
- Enterprise budgets keep growing. NOT AVAILABLE, both B and D marked it so.
- AI demand is durable, not a bubble. PARTIALLY SUPPORTED by the cloud print, unresolvable forward.
- The software franchise faces no credible substitute. CONTRADICTED in direction by the OpenAI launch, financial damage not available.
- Revenue sustains mid-to-high teens. SUPPORTED by the reported growth.
- Net income is recurring operating earnings. NOT AVAILABLE, the near-32% EPS growth on 18% revenue growth cannot be decomposed from the sheet.
- The capex earns above its cost of capital. NOT AVAILABLE, and A, B and D each independently called this the weakest link.
Three seats picked the same weakest link, the unit economics of the capex, and one seat picked the discount rate. Those two links fail differently. The capex question is answerable by disclosure, at the next earnings print. The rate question is answerable by nobody, and it is the one the tape is already moving on.
Where they split
The verdicts look like a wall of refusal. The reasoning is not.
A refused on owner earnings: he cannot verify the return on $115.95B of capex, so he will not pay 28.5x for a business whose cash conversion has halved. B refused on principle: a number he cannot decompose cannot be discounted, so there is no margin of safety to verify at any price. C refused on regime: the entry is bad, the crowd is maxed, and the bond market is moving the wrong way. D did not refuse at all. He called the price fair, and his Hold rests on the stalwart frame while naming the same capex line as his own thesis breaker.
Three seats anchored on the same spread, 1.76% against 5.30%, the same negative-carry arithmetic that refused Apple at 2.84% against 5.24% the day before in our AAPL review. One spread, four different refusals, each falsifiable a different way: better capex disclosure would move A. Computable owner earnings would move B. A 10-year that turns would move C. Seat defection at the renewals would move D.
What we take from it
The single most important number in the Microsoft thesis, the return on $115.95B of annual capex, exists on no public document we could source. Every seat treated that absence as a finding rather than a gap to estimate into. Investor B refused to produce an intrinsic value at all, and that refusal is the discipline: a number nobody can decompose is not evidence, and a committee that laundered it into an estimate would be manufacturing confidence, which is the one thing a committee of models must never do.
The next earnings print, with its fiscal 2027 capex guidance, is the first real chance to replace the missing number. Guidance alone will not settle the return question. But it will say which direction the cash line goes, and on this sheet, the cash line is the whole argument.
Under our three-of-four rule, Microsoft enters no basket.
As of October 1, 2026, BagsCapital manages $0 in assets under management and holds no position in Microsoft.