Okta’s Cash Margin Is Doing the Talking. Its Operating Margin Is the One I’d Listen To.

The market has decided Okta is an inflection story. Two earnings beats in one quarter-cycle, a 12% net retention figure that held the line, and a freshly shipped product to secure AI agents add up to a stock up roughly 67% in three months, at $123.27, a hair under its 52-week high. The narrative is clean: identity is the control plane for the coming wave of autonomous software, regulation is forcing enterprises to prove who, or what, touched what, and Okta is the independent name they reach for. The post-Q1 30% single-session surge priced that thesis in full. The April AI-agent launch supplied the forward story. Analyst target raises supplied the cover.

And the thing the market is admiring, 30% free cash flow margin on the trailing year, north of 35% in the most recent quarter, is real cash. That part I won’t argue with. I’d point instead to the line directly above it on the income statement. Trailing operating margin is 5.54%. Operating income, GAAP, came in around $166 million on revenue that grew under 12%. A company generating $911 million in free cash flow but only $166 million in operating income is telling you something specific about where the gap went.

The gap has a name: stock-based compensation.

The FCF figure the market is capitalizing at a premium is, in large part, the difference between cash that left the building and equity handed out in its place. That is not a scandal; it is how this sector has always worked. But it changes what the 35% number means. Free cash flow that leans on dilution is FCF margin at maturity that hasn’t actually matured. It flatters the present by deferring the cost to the share count. The market is treating cash conversion as proof of operating leverage. The income statement says the operating leverage isn’t there yet.

Here is where I’d move the lens, because the more interesting number isn’t margin at all. Net retention is 107%. For a platform whose compounding case rests on selling more seats and more modules into the same accounts, the upsell and the cross-sell from workforce identity into customer identity, 107% is the floor doing the work, not the expansion. In this sector, gross retention keeps the base, but net dollar expansion is what turns a 12% grower into a 20% compounder. At 107%, the expansion engine is idling. A few years ago this company ran net retention well into the 120s. The slide from there to 107 is the real story, and it sits quietly underneath a Rule of 40 score of 42 that looks fine precisely because the FCF half is carrying the growth half.

The backdrop, to be fair, is working in Okta’s favor. The EU AI Act’s transparency rules land in August 2026, Colorado’s AI statute takes effect at the end of June, and a cluster of US state rules point enterprises toward the same need: attribution, audit trails, proof of which identity authorized which action inside an AI system. Washington’s 2026 cyber posture leans the same way on the federal and FedRAMP side. This is genuine demand pull, and it is the kind of regulatory tailwind that shows up in contract pipelines before it shows up in net retention. The question is whether it shows up fast enough to re-accelerate a number that has been drifting down, not whether it exists.

The strongest version of the long case is that the AI-agent product changes the unit of identity itself. Every human gets one identity; an enterprise running thousands of autonomous agents needs to authenticate and govern every one of them, and Okta charges for the privilege. That is not seat expansion. It is a new usage tier entirely. If it lands, it pushes net retention back up while the channel partner motion scales it cheaply. The market made the same assumption it made the last time Auth0 was supposed to platformize customer identity into a second growth leg: a credible new module, a believable TAM, and a willingness to pay today for the curve that arrives in FY28.

The trouble is the timing gap between the regulatory pull and the revenue. cRPO grew 12%, the same rate as revenue, which means the contracted-but-unbilled pipeline is not yet running ahead of the top line the way it would if agent identity were already inflecting. The AI-agent revenue is a FY28 and FY29 story by the company’s own framing. The market is paying a forward multiple now for bookings it cannot yet see in deferred revenue. Markets love stacking the regulatory tailwind, the product launch, and the retention recovery into one number, then discounting it back at a rate they picked when sentiment was generous.

That discount rate is the part nobody on the buy side wants to revisit. With the 2-year Treasury sitting above the effective Fed Funds rate, the market is signaling it expects rate volatility ahead, and back-ended earnings stories are exactly what gets repriced when the discount rate moves. A name whose value lives in FY28 cash flows is more sensitive to that rate than a name earning it now. There is also a more prosaic mechanic under the 30% single-session pop: 5.58% of the float was sold short. Some of this rally is shorts buying back borrowed stock, which is repricing of positioning, not of the franchise.

I’m not calling the franchise broken. Two-thirds of the Fortune 100 do not switch identity providers on a whim, and the switching cost is the moat. What I’m questioning is the arithmetic: 5.54% operating margin, 107% net retention, sub-12% growth, capitalized at a near-high price that assumes the agent curve arrives on schedule and the expansion engine restarts. If net retention turns back up toward the mid-110s over the next two or three quarters and operating margin starts closing the gap toward the cash margin, my concern is misplaced and the premium is earned. If net retention stays at or below 107% while the multiple holds, the market is paying for an inflection that the renewal book hasn’t confirmed. The same priced-for-the-inflection setup ran through Snowflake’s growth-multiple repricing, and the question there was identical: when does the curve actually show up in the numbers the market is already paying for?

Priced like the inflection already happened. Operates like a 12% grower with a good cash conversion story and a product that matters in two years.

Those two things are not the same. The market usually figures that out. Eventually is the operative word.

Figures reflect the most recent available data and may differ slightly from live market prices. · © Mathstock