Three American technology companies reported quarterly results on Wednesday evening, and the market responded in a way the usual framework cannot explain. Marvell Technology posted record revenue of $2.74 billion, up 37 percent year over year, beat the consensus on adjusted earnings per share at $0.94 against $0.93, and raised its outlook. The stock fell 7.6 percent. Okta reported $805 million of revenue, up 11 percent, and rose from a close of $134.42 to $172.91 — nearly 29 percent in a single session.
Eleven percent growth was rewarded with a 29 percent gain. Thirty-seven percent growth was punished with a 7.6 percent loss. That is not a nuance of expectations; it is a change of sign. And it does not go away by noting that Marvell had already run up 179 percent year to date while Okta had been a laggard for the better part of two years. Positioning explains the magnitude. It does not explain the direction. The direction says something else: in one night, the market moved the artificial intelligence trade down a layer — away from who builds the compute, toward who controls what that compute is permitted to do.
What CrowdStrike actually reported
CrowdStrike reported second quarter fiscal 2027 results for the period ended July 31, 2026. Revenue was $1.47 billion, up 26 percent, with subscription revenue of $1.40 billion, up 27 percent. The number that matters more in a subscription business is annual recurring revenue, which reached $5.84 billion, up 25 percent.
The figure that produced a 20 percent move — the best single day in the company’s history — was neither of those. Net new annual recurring revenue was $332.8 million, up 51 percent from the year-ago quarter. That is an acceleration, and a sharp one: the installed base is compounding at 25 percent while the increment is growing at 51. Management beat its own guidance by more than $45 million and raised the full-year net new ARR growth outlook by 630 basis points, to 34 percent at the midpoint — a cumulative 1,150 basis points above where the year started. Full-year revenue guidance moved to $5.991 billion to $6.011 billion, with ending ARR guided to $6.603 billion to $6.612 billion.
The engine is Falcon Flex, a contracting model under which customers commit a total budget and draw modules against it. Ending ARR from accounts that have adopted Flex passed $2.29 billion, up 101 percent. Module adoption keeps deepening: 51 percent of customers run six or more modules, 35 percent run seven or more, 26 percent run eight or more. Operating cash flow of $530.3 million and free cash flow of $377.4 million were both second-quarter records, and the balance sheet carries $5.01 billion in cash. George Kurtz called it “the best quarter in CrowdStrike’s history.”
One line in the same release stayed out of every headline. GAAP net income was $5.3 million — one cent per share. At the operating line, GAAP showed a loss of $33.2 million. On a non-GAAP basis the company reported $322.9 million of net income, or $0.31 per share, and $371.6 million of operating income. The distance between one cent and thirty-one cents is, in substance, stock-based compensation. The best quarter in company history was, under the rules by which the company actually keeps its books, a break-even quarter. The free cash flow is real and substantial. The earnings that belong to shareholders are not yet.
What Okta actually reported
Okta is the slower business, which is precisely what makes it the more interesting one. Revenue of $805 million beat the $793 million consensus; adjusted earnings of $1.05 per share beat the $0.96 to $0.97 range and compared with $0.91 a year earlier. Subscription revenue grew 12 percent and accounts for 99 percent of the total. Remaining performance obligations — the subscription backlog — rose 17 percent to $4.858 billion against an average contract term of roughly 2.5 years, and current RPO, the portion expected to convert to revenue within twelve months, rose 14 percent to $2.585 billion.
Profitability is steady and, in the best sense, dull: a 28 percent non-GAAP operating margin, unchanged year over year, and free cash flow of $227 million at a 28 percent margin, up from $162 million and 22 percent. The balance sheet holds $2.299 billion in cash and short-term investments. The 2026 convertible notes matured in June and Okta retired the remaining $350 million principal in cash, leaving no convertible debt outstanding. During the quarter the company repurchased 1,542,442 shares at an average cost of $81.06, spending $125 million of the $1 billion authorization approved in January, with $555 million remaining. Set an $81.06 average cost against Thursday’s $172.91 close and you have the one capital allocation decision of the quarter that has already more than doubled.
None of that justifies 29 percent on its own. Eleven percent revenue growth, guided to ten percent for the third quarter, with current backlog growth guided down from 14 percent to a range of 11 to 12 percent, is the arithmetic of a mature software company, not a growth stock.
Why the market paid more for the slower company
The answer is not in the income statement; it is in the product list. Okta sold Okta for AI Agents and Auth0 for AI Agents during the quarter — to an appliance manufacturer that needs to discover shadow AI and govern hundreds of autonomous agents running inside Gemini Enterprise, to a commercial insurer that wants short-lived permissions, human-in-the-loop review and clean attribution separating human from agent actions, to a North American bank, to a Department of Defense organization. Cross App Access, the standard by which agents connect to enterprise applications, added more than 25 integrations. And Okta now appears as a named identity layer in the enterprise agent programs of nearly everyone shipping agents: Anthropic, for governing Claude Enterprise and the connectors behind it; Google Cloud, for the Gemini Enterprise agent platform and Chrome Enterprise; Cisco; Databricks; Snowflake; and an expanded multi-year strategic collaboration agreement with Amazon Web Services.
In the same quarter CrowdStrike introduced Continuous Identity for AI Agents, extending risk-aware authorization across human, non-human and AI agent identities. Two competitors, two product launches, one thesis.
That thesis attacks the argument that has weighed on enterprise software multiples for three years. If agents do the work of employees, seat counts fall, and any business that bills per seat shrinks with the headcount that uses it. But identity is not billed per human. It is billed per identity — and an agent is an identity that has to be authenticated, authorized, logged and revoked when it is decommissioned. Not one per employee. Hundreds per department. On Wednesday night the market did not buy this quarter’s growth. It repriced the unit being counted.
The bill Palo Alto already paid in February
The thesis is not new; it has simply never had a quarterly print behind it. Palo Alto Networks closed its acquisition of CyberArk on February 11, 2026 for roughly $25 billion in cash and stock — $45 plus 2.2005 Palo Alto shares for each CyberArk share. The stated rationale was exactly what surfaced this week in two earnings reports: securing every identity in the enterprise, human, machine and agentic. The deal materials carry the ratio that drives all of it. Machine identities outnumber human ones by more than eighty to one. Close to 90 percent of enterprises report identity-related breaches, and roughly three-quarters still run outdated privilege models.
Use February’s $25 billion as the price anchor and Thursday makes more sense. The move was sector-wide rather than company-specific: Palo Alto Networks, SailPoint, Zscaler and Rubrik each gained at least 10 percent. What got repriced was not CrowdStrike and not Okta. It was a layer.
The counter-argument is in the same filings
There are sound reasons to decline this re-rating, and all of them come from the companies’ own disclosures. The cleanest test of the claim that existing customers are spending more because of agents is the dollar-based net retention rate, which measures what the same customers pay twelve months later. At Okta it is 107 percent. That is expansion, but thin expansion, and nowhere near the levels that prevailed the last time the stock carried this multiple. Customers with more than $100,000 in annual contract value grew 6 percent to 5,255. Current backlog growth is guided lower, not higher. The agent business appears in the filings as individually narrated wins — anecdotes, however impressive — and not yet as a line in the income statement. Press accounts put calculated billings down about 5 percent in the quarter; with a 2.5-year average contract term that metric is notoriously noisy, but it is not showing acceleration either.
At CrowdStrike the risk sits elsewhere. Falcon Flex front-loads commitment: customers pledge a budget and draw modules later. Commercially that is smart and it locks customers in early, but it pushes the question of when a commitment becomes consumption into the future, and ARR composed increasingly of committed rather than consumed budget is a different quality of ARR than ARR composed of live subscriptions. Add the GAAP operating loss in the best quarter the company has ever had, and the honest summary is that growth currently costs roughly what it brings in.
Then the most obvious risk of all: Microsoft bundles identity management into enterprise agreements many customers already pay for. Every thesis about pricing power at specialist identity vendors has to survive that bundle. Entra is not better than Okta at the high end; it does not have to be. It has to be adequate and already purchased.
Sizing it for a US portfolio
Investors reaching for this theme should be clear about the arithmetic of the opportunity. CrowdStrike guides to about $6.0 billion of revenue this fiscal year and Okta to roughly $3.2 billion. Together that is around $9 billion of annual revenue set against AI infrastructure spending running into the hundreds of billions. Security is a small levy on a very large base — which is the bull case and the constraint at once. A small levy on a growing base compounds well; it also means the theme cannot absorb capital the way the compute layer did, and single-name concentration is the practical risk in a group whose members just moved 10 to 29 percent in a day.
Two practical notes. First, holding period: positions taken into a 29 percent single-day move are, by definition, short-term positions, and gains realized inside twelve months are taxed as ordinary income rather than at the long-term capital gains rate — a spread that at higher brackets exceeds the entire dividend yield of most large-cap equities, and neither of these companies pays a dividend. Second, overlap: broad technology index funds and most large-cap growth funds already hold Microsoft, Palo Alto Networks and CrowdStrike at meaningful weights. Adding a cybersecurity thematic fund on top is often less diversification than it looks.
Four numbers for the next report
This week’s thesis is testable in four places. First, Okta’s net retention rate: if it rises from 107 percent, existing customers really are paying more for agent identities. If it holds flat, the move was multiple expansion. Second, current RPO against the guided 11 to 12 percent — anything above is evidence, anything below is the opposite. Third, at CrowdStrike, the relationship between committed and consumed Falcon Flex budget, and whether the GAAP operating loss closes while net new ARR compounds at 51 percent. Fourth, across the group, whether specialists hold price as Microsoft tightens the bundle.
Conclusion
Kevin Warsh delivers his first Jackson Hole keynote as Federal Reserve chair on Friday, and the 30-year Treasury yield sits at multi-decade highs. That is not a backdrop in which high multiples get discounted generously. Which makes the week’s scoreboard all the more striking: 29 percent awarded for 11 percent growth, 7.6 percent taken away for 37 percent growth.
The honest reading is not that security software has become a better business than semiconductors. It is that the first phase of the AI trade — who builds the data centers — is now priced so completely that a record quarter growing 37 percent can disappoint, while the second phase — who governs what acts autonomously inside those data centers — was not priced at all. The evidence for that second phase, however, does not yet exist in the financials. It exists in product launches, partner lists and narrated customer wins, not in net retention. Until one becomes the other, Thursday’s re-rating is exactly what it appears to be: a bet that the market has started counting the right unit before the unit starts paying.
Try TradingView Free for 30 Days
Plus get a $15 discount on your first subscription through this link.
