Thesis: SentinelOne is winning the product argument and losing the scale argument, and the place those two collide is gross margin. The platform is genuinely AI-native rather than retrofitted, non-endpoint solutions now approach half of recurring revenue, and net new business is running at record levels. Yet revenue growth has settled at twenty-one per cent, large-customer additions are decelerating, and gross margin is falling while every larger competitor holds theirs. The question is not whether the technology is good. It is whether a company at roughly a fifth of CrowdStrike’s scale can fund a frontier-model arms race without surrendering the unit economics that justify independence.
The business
SentinelOne sells autonomous endpoint protection and, increasingly, everything adjacent to it. The Singularity platform detects and neutralises threats at the point of execution on the host machine, using behavioural models rather than signatures, and it does so without a human analyst in the loop.
The strategic bet is that the endpoint remains the ultimate control plane. Chief executive Tomer Weingarten argues that verifying identity at the door is useful but insufficient, because attackers routinely operate post-authentication inside environments where trust has already been granted. The recent supply-chain compromises he cites were exploit-free, carried no signature and triggered no perimeter control; the only effective defence was behavioural interception at the host.
Around that core the company has assembled four product lines. Endpoint remains the base. Prompt Security, acquired in August 2025, secures enterprise use of generative AI and now extends to agent security and AI red teaming. Observo AI, bought for two hundred and twenty-five million dollars a month later, supplies the telemetry pipeline underneath an AI-native security information and event management product positioned against Splunk. And Purple AI, the agentic security operations layer, opened to all customers in June 2026 and by August was running more than eight thousand five hundred autonomous investigations per day.
Distribution has shifted decisively toward partners: an expanded global alliance with LevelBlue, the largest managed security provider in the world; Google Cloud partner of the year for security in April 2026; and consumption-based purchasing directly from the AWS console, which removes enterprise procurement from the sales cycle entirely.
Where the numbers actually sit
The second quarter of fiscal 2027, reported on 27 August 2026, contained both halves of the argument in a single release.
Revenue reached two hundred and ninety-two million dollars, up twenty-one per cent. Annual recurring revenue passed one point two billion, up twenty-two per cent. Net new recurring revenue of fifty-six million was a record for a second quarter, and remaining performance obligations accelerated sharply to forty-five per cent growth. Non-GAAP operating margin reached ten per cent, an improvement of roughly eight hundred basis points year on year, which is a genuinely impressive demonstration of leverage.
Three numbers cut the other way. Revenue growth has now printed twenty-one per cent for two consecutive quarters, and full-year guidance implies roughly twenty at the midpoint against twenty-two last year. Customers spending above one hundred thousand dollars grew thirteen per cent, decelerating from seventeen in the prior quarter, with only thirteen net additions sequentially. And gross margin fell year on year on both a GAAP and non-GAAP basis, from seventy-five to seventy-two per cent and from seventy-nine to seventy-seven respectively.
The market read the combination as mixed. Despite the revenue beat, non-GAAP earnings guidance was trimmed and the shares fell around eight per cent.
Bull case
The platform argument is not marketing. Singularity was built on machine learning from inception rather than acquiring it, which matters in a threat environment where adversaries increasingly deploy frontier models. Weingarten’s framing is that enterprises cannot defend against AI-driven attacks by consolidating onto legacy platforms with tools bolted together.
The product mix is validating that claim. Non-endpoint solutions approached half of total recurring revenue for the first time in the first quarter of fiscal 2027, with AI security revenue nearly doubling sequentially and data solutions posting a fourth consecutive quarter of accelerating growth. Purple AI attach rates have exceeded half of new licences, and in early full deployments management reports the agentic security operations revenue can exceed a customer’s core endpoint spend. industry research has named the company a leader in endpoint protection for a sixth consecutive year.
Two structural advantages are underappreciated. The first is operating system coverage: much of the world’s artificial intelligence is built and run on Linux and macOS, where many competitors remain effectively Windows-centric. The second is sovereignty. SentinelOne can secure fully air-gapped and highly restricted on-premises environments that pure cloud architectures cannot reach, an increasingly relevant capability for government and defence buyers.
Nearly half the installed endpoint base still runs legacy antivirus. That is the market share opportunity, and it does not require taking a single customer from CrowdStrike.
Finally, the frontier-model relationships are real rather than press-release deep. Wayfinder, launched in April 2026, is built on Anthropic’s Claude Security offering and Google threat intelligence, and the company is a named early-access partner in OpenAI’s trusted access programme for cyber. Management has also stated on the record that it is a participating vendor in Project Glasswing.
Bear case
Microsoft is the structural threat, and it is the only participant with a credible published share number. industry research put Microsoft at 28.6 per cent of the modern endpoint security market in 2024, growing faster than the market. Defender arrives bundled inside enterprise agreements customers have already signed. Any reader encountering a precise market share figure for SentinelOne, CrowdStrike or Palo Alto Cortex should treat it with suspicion; no credible public source publishes one.
Large-customer momentum is the near-term worry. Thirteen net additions above the hundred-thousand-dollar threshold in a quarter, with cohort growth decelerating from seventeen to thirteen per cent, suggests that record net new recurring revenue is being carried by expansion within existing accounts rather than by winning new enterprise logos. That is a materially less durable growth engine.
Gross margin is the most under-discussed number in the story, and it is going the wrong way in both quarters of the current fiscal year. Frontier-model inference, petabyte-scale data ingestion and a channel mix weighted toward managed service providers all consume margin. Competitors operate in the high seventies to low eighties and are stable. A structurally lower gross margin caps the operating margin ceiling permanently, which matters for a company whose entire equity story now rests on reaching the Rule of 40.
Execution history invites caution. The 8-K filed on 28 May 2026 discloses a reduction of approximately eight per cent of full-time employees and a charge of about twenty-five million dollars. Management framed it as an AI-enabled productivity gain rather than a demand response, and the annualised saving of roughly forty-five million was asserted on the earnings call rather than in the filing. It followed a period of substantial turnover at the top of the organisation, including two chief financial officer transitions in twelve months before Sonalee Parekh took the role in March 2026.
The deciding question
Can SentinelOne monetise agentic security fast enough to offset the margin cost of delivering it?
The company is moving from seat-based licensing toward usage-based metering, charging for consumption of the security data lake and Purple AI through a credits system, alongside prepaid Flex commitments that crossed two hundred million dollars of contract value within a year of launch. That is the correct architecture for an agentic product, and notably it is the transition Okta has so far declined to make.
But agentic security is expensive to run. Every autonomous investigation consumes frontier-model inference that SentinelOne buys from Anthropic, OpenAI and Google, supplemented by its own models. Eight thousand five hundred investigations a day is an impressive adoption statistic and a real cost line. The margin compression already visible in the reported numbers is the early evidence of that trade.
So the platform expansion that makes SentinelOne strategically interesting is simultaneously diluting the unit economics that would make it independently viable. Either usage-based pricing captures the value and gross margin stabilises, or the company grows into a structurally lower-margin business than the competitor it needs to catch. Both outcomes are live, and the reported gross margin line resolves the question before any narrative does.
What to monitor
Gross margin. Two consecutive quarters of year-on-year decline. Stabilisation would confirm that usage-based pricing is recovering the cost of agentic delivery; a third decline would establish it as structural rather than transitional.
The hundred-thousand-dollar cohort. Thirteen net additions and decelerating growth is the clearest evidence available on whether the upmarket motion is working. Record net new recurring revenue means considerably less if it is expansion within a static enterprise base.
Purple AI and data solutions as a disclosed proportion of recurring revenue. Non-endpoint approaching half is the strongest single datapoint the company has produced. Crossing it decisively, with attach rates holding, would confirm the platform thesis.
Operating leverage against the restructuring. Roughly forty-five million dollars of annualised savings against a ten per cent operating margin should compound quickly. If it does not appear in the second half, the reduction was defensive rather than strategic.
Consolidation. Palo Alto Networks was reported in talks to acquire SentinelOne in July 2025 at a price near ten billion dollars, which the company declined to comment on. Nothing comparable has surfaced in 2026, and the identity-adjacent acquirers have since spent their capital elsewhere. A subscale leader with best-in-class technology in a consolidating market remains a logical target, but that is an argument about ownership, not about operations.
Bottom line
SentinelOne is the strongest independent product in endpoint security and the weakest balance sheet position among the platforms it competes with. Those two facts have coexisted for years and the current fiscal year is the first in which they are being genuinely tested against each other.
The bull case no longer depends on winning the endpoint war. It depends on non-endpoint solutions, already close to half of recurring revenue, compounding faster than the core decelerates, with usage-based pricing capturing enough of the agentic opportunity to hold margin. The bear case is arithmetic: a smaller company funding the same frontier-model costs as larger rivals, with falling gross margin, decelerating enterprise logo growth and no independent efficacy benchmark to point at.
Management has been notably honest about the operating discipline required, cutting eight per cent of staff while explicitly targeting the Rule of 40. The technology has never been the problem. The question this year, and it is a fair one, is whether independence at this scale is a strategy or a phase.