Microsoft is down roughly 24% over the past year, sitting closer to its 52-week low ($349) than its high ($555), and trading around a 21x forward multiple that's cheap by its own recent history.

The easy story writes itself: dominant franchise, AI tailwind, temporarily out of favor, load up.

Except, I don't think it's that simple.

Going into this, I had it in mind to never bet against Microsoft. But based on my research, I just don't think it's the opportunity others are touting.

Let me lay it out for you.

Microsoft has committed to roughly $190 billion in capex for fiscal 2026 — a 61% increase from 2025 — and management has said it will increase total AI capacity by over 80% this year and roughly double its datacenter footprint over the next two years.

That is not the language of a company hedging its bets. It's a full commitment to a single wager: that this spending converts into durable free cash flow within a reasonable window.

The bet rests on three layers working together: Copilot at the application layer, Azure at the infrastructure layer, and Microsoft's enterprise AI tooling on top of both. Combined, the pitch is that Microsoft doesn't need to win the model race outright — it just needs to own enough of the surface that it benefits regardless of which model wins underneath.

That's a clean story, but there's more to it.

Because the crux of the current valuation is all about timing. Specifically, when does capex intensity decline and start converting to cash flow.

Everyone looking at this stock agrees on the setup. Where they disagree is on the clock.

With that in mind, here's how different perspectives are viewing that.

Analyst targets cluster well above today's price — consensus sits around $556.75, with dispersion as high as $870.

That optimism comes with a specific, near-term clock attached. The bull case is explicitly pinned to Azure holding 30%+ growth into FY27, with new AI capacity coming online in the back half of 2026.

Analysts aren't just optimistic — they're betting conversion shows up within roughly the next 12 to 18 months.

The market isn't pricing "never." It's pricing "not on the Street's schedule."

The current price sits below every bear case I found in sell-side research. This indicates capex conversion is expected to take longer than what analysts are hoping for, though a specific timeline isn't known.

The clearest signal on timing isn't what management says — it's what they've actually done with two decisions this year.

In April 2026, Microsoft and OpenAI formally ended Azure's exclusive distribution rights to OpenAI's models — a structural moat that had held since 2019. Microsoft kept a non-exclusive IP license through 2032, a first-to-ship provision, and its ~27% OpenAI equity stake, but gave up the "if you want GPT at scale, you run it on Azure" lock. At the same time, Microsoft has pushed its own in-house MAI models into production, routing real Excel and Outlook volume away from OpenAI and Anthropic, with Mustafa Suleyman stating the explicit goal of eliminating what Microsoft pays Anthropic.

Neither move is a bet that capex converts to cash quickly. Both read as hedges against being fully dependent on someone else's model economics while the infrastructure bet plays out on its own timeline, however long that takes.

Notably, Microsoft hasn't put a number on that timeline — no explicit capex-deceleration guidance yet. That vagueness is itself the signal: of the three views here, Microsoft's own actions look the least anchored to a near-term clock. Not betting on fast conversion like the Street, but building for a runway long enough that it hasn't needed to specify an end date.

We modeled these as a discounted cash flow, tying revenue growth to the same capex-conversion story as each scenario — because a bear case that keeps spending high without also assuming demand disappoints isn't really modeling anything. Built off trailing financials (TTM revenue ~$325B, TTM capex ~$97B, or roughly 30% of revenue), discounted at a 9% WACC with 4% terminal growth:

Scenario 1 (bear) — growth disappoints, fading from ~12% toward Microsoft's slower pre-AI trend of ~7% as the AI spend fails to convert to demand. Capex still peaks high (50-55% of revenue) but grinds down to 24% by year six — because construction capex is inherently finite (a datacenter eventually gets built whether or not it pays off), separate from a maintenance-capex floor of roughly 9% of revenue that persists either way. ~$261/share. Roughly 32% downside from today — real, but not a collapse.

Scenario 2 (bull) — capex moderates fast (down to 17% of revenue by year six, in line with the Street's 12-to-18-month clock) and the spend actually converts: Azure/Copilot growth accelerates to 16–19% as the infrastructure captures real demand. ~$516/share. Roughly 35% upside from today — still below the Street's ~$557 consensus, meaning even our own optimistic case doesn't get us to where analysts sit.

Scenario 3 (base) — capex grinds down gradually (50% this year to 20% by year six) while growth holds at a moderate, AI-boosted-but-not-euphoric pace (14% fading to 9%) — partial conversion, not full payoff. ~$365/share.

That last number is worth sitting with: the base case now lands slightly below where the stock trades, not above it — a different read than "cheapest multiple in years, must be a bargain," but also not "richly priced."

And notice what the bear and bull numbers do to the shape of the range: −32% against +35% is roughly symmetric, not the lopsided setup you might expect from a stock that's already down 24% over the past year.

Q4 earnings land July 29, two days from today, with capex guidance the single most-watched line item. That print won't resolve the timing question above, but it's the next real data point on it — and given Microsoft's own MAI push is still recent and unproven at scale, it's also worth watching whether the margin-expansion story it implies actually shows up in the numbers.

We're not calling this cheap, and we're not calling it overvalued. We ran the base case across a reasonable range of discount rates (8%–10% WACC, 3.5%–4.5% terminal growth) to see how much of this conclusion depends on the assumptions rather than the business, and the honest answer is: quite a bit.

But what's more interesting is what happens when you run the same sensitivity range on bear and bull. Bear ranges roughly $195–$383/share across the grid; bull ranges roughly $391–$747. At the generous end of the assumptions, the two cases nearly meet in the middle of where the stock trades today.

It's closer to genuinely balanced: real risk if the AI spend doesn't convert, real reward if it does, in roughly the same order of magnitude either way.

That's a less dramatic story than "cheapest multiple in years, asymmetric bargain".

If there's an edge here, it isn't in the direction of the bet. It's in being honest that this is closer to a coin flip on timing than a mispriced asymmetry — and being clear-eyed about which side of that flip you're on before putting money behind it.