JFrog is the software supply chain system of record, and it is repositioning the role of its registry around the flood of machine-generated code. AI agents now produce artifacts faster than human teams can vet them, and the company is turning that volume into a security and governance toll rather than letting it erode the moat.
Trailing net dollar retention reached 121 percent, up from 118 percent a year earlier. The cohort of customers above annual recurring revenue of a million climbed to 97 from 61. That expansion engine is the load-bearing number, and it is accelerating rather than cooling.
Q2 revenue of 163.8 million grew 29 percent. Non-GAAP operating margin reached 19.9 percent, while GAAP earnings stayed negative. Can retention and margin expansion outpace the multiple the market has already paid for this growth?
JFrog is an Israeli company listed on Nasdaq that sells subscriptions to a unified platform for the software supply chain. It spans DevOps, DevSecOps, DevGovOps, and AI and MLOps workflows. The platform is organized around Artifactory, the artifact registry at its center, plus Xray scanning, Curation, AppTrust, an AI Catalog, and Runtime Security. It operates a single reportable segment and deploys as SaaS or self-managed. Roughly 6,600 organizations use it, including a majority of the Fortune 100. The business model is self-reinforcing, because free trials and an open source registry pull developers in, the paid tiers monetize teams and enterprises, and the Enterprise Plus bundle now captures the whole supply chain instead of a single point.
The competitive set is broad and fragmented. GitHub, GitLab, and Sonatype attack the developer-facing edge, while the hyperscalers offer overlapping registry and pipeline functions, and Snyk, Aqua, Checkmarx, and Synopsys contest the security layers. JFrog's standing rests on spanning all four workflows on a single system of record rather than selling point solutions. That breadth is also the reason the platform is hard to rip and replace, because every workflow that gets added becomes another dependency the customer would have to rebuild if it left.
The strategic shift is a move from an artifact registry into the control plane for AI-generated software. When coding agents emit code at machine speed, the registry becomes the chokepoint where trust and governance get enforced before production. Cloud revenue reached 53 percent of the total, up from 45 percent a year earlier. That mix shift compounds the toll, because SaaS generates committed consumption that licenses did not.
The Israel domicile is a structural variable that belongs in context, not in the margin line. The company maintains redundant infrastructure and code outside the country, and the February 2026 escalation with Iran and the April 2026 ceasefire sit over a workforce present in the country. The consequence for shareholders is a geopolitical premium and discount that alternate on headlines, independent of the operating trajectory. The practical read is that the multiple carries a component that has nothing to do with the P and L, and that component can compress on a single headline even when every operating metric is improving.
The product moat is positional rather than architectural. No single component is hard to copy, but the accumulation of every artifact, dependency, scan result, and policy across a customer's software life is expensive to replicate and awkward to migrate. That is where switching costs actually live. The goodwill on the balance sheet reflects a decade of bolt-on acquisitions that filled the platform from a registry into a suite, and each acquisition added a layer that deepens the data gravity holding customers in place.
Three events this quarter moved the moat in the AI direction. The first is the JFrog plugin built with Anthropic that brings supply chain governance and security to Claude Code. The mechanism is a distribution wedge. By embedding inside the tool where AI code is written, JFrog becomes the enforcement layer at the moment of generation rather than a downstream gate. For shareholders that is a positioning gain before it is a revenue line, because it preempts competitors from owning the AI entry point.
The second event is the security solution delivered to more than a million AI developers through the Cursor coding agent plugin. The same wedge measured in reach shifts the top of the funnel from procurement-led to developer-led. It matters to the thesis because developer-led penetration historically precedes the large account cohort that drives the trailing retention, and the bundle sales team then closes on the footprint the plugin created.
The third event is being positioned as a Leader in the first Gartner Magic Quadrant for software supply chain security. The consequence is category creation. JFrog is helping define a new quadrant rather than competing in an established one, and that confers pricing power and a defensible narrative in RFPs. The limit is that a Leader box is a one-time stamp, and the real test is whether the category converts into security attach that lifts net dollar retention further.
Q2 revenue of 163.8 million grew 29 percent year over year. Roughly 30.3 million of the increase came from existing customers, with the remainder from new logos. That split is the cleanest signal in the quarter. Growth is coming from expansion inside the base rather than from hunting fresh accounts, which is the more durable form because it does not require paying full sales cost on every incremental unit.
Gross profit was 127.6 million at a 77.9 percent GAAP margin. Non-GAAP gross margin reached 83.2 percent. The improvement over the prior year came from operating leverage as revenue outgrew the hosting and support cost base. Third-party hosting rose on higher SaaS consumption, but it was absorbed by the larger top line, which is the signature of a software business scaling past its cost inflection.
Operating expenses were 140.9 million, down from the prior year pace as the mix shifted. They represented 86 percent of revenue, compared with 96 percent a year earlier. Research and development, sales and marketing, and general and administrative each grew slower than revenue. The GAAP operating loss was 13.2 million, while non-GAAP operating income was 32.6 million. The gap is dominated by share-based compensation, which is a real economic cost but a non-cash one.
Cash generation is where the model is strongest, and it is the part of the print that deserves the most weight. Operating cash flow was 57.1 million, and free cash flow was 53.7 million. Cash, equivalents, and investments reached 824.5 million at quarter end. Deferred revenue rose to 351.5 million, a backlog proxy growing faster than recognized revenue and supporting forward quarters. The GAAP net loss narrowed to 4.2 million, and the accumulated deficit sat at 443.9 million.
Management guides third quarter revenue to a range of 164 million to 166 million. Full year revenue guidance is between 648 million and 652 million. Non-GAAP operating income is guided to 116 million to 120 million for the year. The guide implies a full-year growth rate in the mid-to-high twenties, a deceleration from the 29 percent printed in the second quarter. The forward question is whether AI-related security demand holds growth near the high end of that range or normalizes to the low end as the base grows.
In February the board approved a share repurchase program authorizing up to 300 million of ordinary shares. It became effective in March with no expiration, and 298 million remained available at quarter end, with only 2.0 million retired in the first half. Against a large net cash position, an active repurchase program is the clearest signal that management views the shares as cheap relative to the cash generation. A program that sits nearly untouched for two quarters is an authorization, not yet a commitment. The next several quarters resolve whether it becomes a genuine offset to the dilution from the annualized share-based compensation load. The strategic point is that the buyback is the clearest available signal of management's view of the share price, and a large program paired with a near full cash balance is a management team telling the market it does not consider the equity expensive. That is a more credible form of cheapness than any commentary, because it is money rather than words.
Execution risk also concentrates in the new office lease that commenced in July, a ten-year commitment that locks in overhead ahead of the revenue it is meant to support. The second risk is the AI buildout, where the company competes for the same developer mindshare as the very agent platforms it now integrates with. The third is integration discipline. A platform assembled from dozens of acquisitions carries a standing risk that the next bolt-on disrupts the product more than it extends it, and the goodwill balance makes that a balance-sheet item, not just an operational one.
The dominant downside risk is retention decay. The entire valuation rests on a trailing net dollar retention of 121 percent, and the mechanism of that number is expansion within the installed base. If a customer consolidates on a platform provider that bundles a cheaper registry and scanner, retention can step down faster than the bundle mix can offset it. A slip from 121 percent toward the high 90s would not cut much near-term revenue, because deferred revenue cushions the print. But it would reframe the multiple from a compounding growth story to a mature plateau, and that reframe is where most of the downside lives.
The second risk is competitive encirclement from both directions. From above, the hyperscalers bundle registry and pipeline functions into accounts they already own, which pressures the SaaS portion that is now 53 percent of revenue. From beside, specialized security vendors compete on the scanning and governance layers where JFrog is trying to build the new category. The consequence is a two-front margin pressure that is slow to show up in the income statement because the bundle is still gaining, but that can surface in a pricing cycle if competition hardens at renewal.
The third risk is structural and geographic. An Israel-domiciled company with a workforce present in a conflict region carries a binary event risk that no business continuity plan fully removes. A renewed escalation would move the stock on headlines before any operating metric does. There is also a lighter but real currency item, with foreign exchange moving the interest line quarter to quarter.
The honest downside scenario is a combination, not a single event. Retention drifts toward the high 90s, competition erodes pricing at renewal, and a geopolitical headline suppresses the multiple, and the three together take the stock to a multiple the cash flow alone cannot support. The offset is that the large net cash position and the free cash flow give the equity a floor that a pure growth name without balance-sheet strength would not have. In other words, the downside is a multiple compression risk more than a cash flow risk, and that distinction matters because a multiple can recover when the multiple is the problem, whereas a broken cash flow cannot.
The framework starts from non-GAAP operating income, the number that matches how the business generates cash once non-cash share-based compensation is set aside. The second quarter non-GAAP operating margin of 19.9 percent, annualized, points to a run rate in the mid 200 million range against full year guidance. The multiple that matters is enterprise value to forward non-GAAP operating income, and the bear, base, and bull cases turn on where that margin lands by year end.
In the bear case, growth decelerates to the low end of the guide, retention slips toward the high 90s, and the market prices the equity as a mature platform that has already captured its AI narrative. Against that, a multiple in the high teens to low twenties is the starting point. The non-GAAP operating income base for the year sits in the range of 116 million to 120 million. Applied to that base, the multiple implies a value in the region of 2 billion to 2.6 billion. After subtracting the net cash cushion, that equity value lands in the range of 16 to 21 per share.
In the base case, the margin holds near 20 percent, retention stays in the low 120s, and the multiple settles in the mid twenties. That supports an equity value near 3 billion, or about 23 per share, with the net cash providing a cushion under the print. The bull case is the one the market is already pricing. If the AI supply chain security category matures on schedule, retention holds above 120 percent, and the repurchase program retires a meaningful fraction of the float, the multiple can stretch into the low thirties. That supports an equity value toward 3.6 billion or more, or the low to mid 20s per share.
The same input set carries from the financials into the conclusion. The net cash figure, the margin trajectory, and the retention number are the three variables that move all three scenarios, and the quarter answered them in a direction that favors the base and bull cases while leaving the bear case alive on retention alone. It is worth stating the asymmetry plainly. The bear case requires a deterioration in retention that the most recent print is not showing, while the base and bull cases require only that the margin and retention trends continue at their present pace. That asymmetry is what makes the downside scenario a real but conditional risk rather than the default outcome.
The quarter revealed that JFrog's repositioning from registry to AI control plane is generating real traction, not just narrative. Retention rose to 121 percent, and non-GAAP operating margin reached 19.9 percent. The cohort of customers above a million of annual recurring revenue jumped to 97 as well. Those three together say the security attach and the cloud shift are doing exactly what the strategy promised, even while GAAP earnings remain negative.
The central initiatives are now load-bearing rather than aspirational. The Anthropic and Cursor integrations put supply chain governance at the point where AI code is generated, the Gartner Leader placement creates a new category the company helps define, and the repurchase authorization, backed by a large cash position, gives management a tool to retire dilution from the annualized share-based compensation load. The combination is a company building the AI toll, validating it with third parties, and holding a war chest to protect the per-share economics.
What the next several quarters resolve is specific to this print. The variables to watch are the pace of the repurchase program, with most of the authorization still available against only 2.0 million retired in the first half. Also the retention number and whether it holds above 120, the third quarter revenue against the guide, the Enterprise Plus bundle at 59 percent of revenue, and the AI plugin adoption as it converts developer reach into enterprise security attach. If the buyback accelerates and retention holds, the base case is the right one, and the stock is a compounder with a cash floor. If the program stays dormant and retention slips, the multiple has already paid for the growth and the bear case takes over.