SentinelOne, and the Autonomous SOC

We went long SentinelOne common stock on 23 October 2026 for a three-to-five-year hold. This is the highest-risk of the three names we initiated this week, and we are putting that at the top rather than the bottom. The case is a valuation gap, not a safety story: SentinelOne grows annual recurring revenue at roughly 22% with 72% gross margins and a net-cash balance sheet, and it trades at about 7.6 times sales while the category leader, CrowdStrike, trades near 45 times. We do not think a company growing ARR in the low twenties should trade at a sixth of its closest comparable's multiple, and over three to five years we are underwriting partial convergence of that gap, not a miracle.

The bottleneck framing is what makes this more than a cheap-stock bet. Security teams are drowning: the volume of alerts has outrun the number of analysts who can triage them, and the constraint in a modern security operations centre is no longer detection, it is autonomous response at machine speed across the whole estate, from endpoint to cloud to identity to the SIEM where the logs live. SentinelOne's bet is an AI-native platform, Singularity, with an agentic analyst, Purple AI, designed to do the triage a human cannot keep up with, plus an AI-SIEM aimed squarely at the data-and-search layer Splunk has owned. Own the autonomous layer of the SOC and you own the constraint. That is the thesis, and it is why the discount to CrowdStrike looks like an opportunity rather than a verdict.

We are not pretending this is derisked. It loses money on a GAAP basis, its enterprise growth has slowed, and it cut its profit guidance in August. We size it accordingly and we spend a full section on exactly how it could go wrong.

SentinelOne sells a cybersecurity platform built around autonomous endpoint and extended detection and response. In its most recent quarter, reported in late August, revenue was $292 million, up 20.6%, and annual recurring revenue reached $1.22 billion, up 21.7%. Gross margin is about 72%, which is the signature of a real software business, and the company turned a positive adjusted operating margin of about 10.5%, a meaningful improvement from losing money at the operating line a year earlier. It holds about $655 million of cash and investments against no debt.

It is the clear number-two independent platform in endpoint security behind CrowdStrike, and it competes with Microsoft's bundled Defender from below on price and from a different angle on architecture. Two things distinguish its approach. First, it leans on an AI-native design and a channel-heavy go-to-market, with managed-service-provider partners and hardware alliances such as Lenovo shipping it on devices. Second, it is pushing beyond endpoint into the AI-SIEM, the security data lake and search layer, which is a much larger market than endpoint alone and the one where the agentic-analyst idea has the most leverage.

The product that carries the story is Purple AI, an agentic security analyst that investigates and responds to threats in natural language. If it works at scale, it is the thing that lets a security team cover more with the same headcount, which is the exact constraint we described. The quarter showed the platform is real and growing; the question the next three years answer is whether it re-rates.

Apply the house question. The structural constraint in security is not that threats cannot be detected; modern tools detect far more than anyone can act on. The constraint is human. A security operations centre generates more alerts than its analysts can investigate, the talent shortage is chronic, and attackers now move at machine speed using the same AI the defenders have. The only way the math works is to automate the analyst, and that is a software problem that favours whoever has the most AI-native architecture and the broadest view of the estate.

This is why the category is converging on three layers that have to work together: the endpoint and workload sensors that see what is happening, the identity and cloud telemetry that shows where an attacker moves, and the SIEM, the data lake where all of it is stored, searched and correlated. The winners own the autonomous layer that sits on top of all three and acts without waiting for a person. SentinelOne is building toward exactly that stack, and its AI-SIEM push is the attempt to own the data layer rather than rent it from Splunk or Microsoft.

We are honest that SentinelOne does not own this bottleneck the way RTX owns interceptor production; it is competing for it against a larger, profitable, faster-consolidating rival. But the prize is the autonomous SOC, the direction of travel is clear, and SentinelOne is a credible number two trading as though it were an also-ran. The thesis is that a credible number two in a market this large and this structurally short of human capacity is worth far more than 7.6 times sales.

The entire quantitative case is one comparison. SentinelOne grows ARR about 22% and trades near 7.6 times sales. CrowdStrike grows at a broadly similar rate and trades near 45 times sales. The market is paying roughly six times as much per dollar of revenue for the leader as for the number two, despite comparable top-line growth.

Some of that gap is deserved and we will say so. CrowdStrike is larger, GAAP-profitable, generates strong free cash flow, and has arguably the better brand and the faster enterprise momentum. A premium is warranted. A six-fold premium is not obviously warranted, and that is the inefficiency we are buying. Over three to five years, two things can close it: SentinelOne's own fundamentals improving (sustained 20%-plus ARR growth, the operating margin turning durably positive, free cash flow going consistently positive), and the AI-SIEM and Purple AI giving it a story the market is willing to pay up for. We do not need the gap to close fully. We need it to close partway, because from 7.6 times sales even a move to, say, twelve to fifteen times on a larger revenue base is a large return, and the downside is cushioned by a net-cash balance sheet and a business that is now around breakeven on an adjusted basis.

There is also an unpriced optionality we will name and not lean on: SentinelOne has been a perennial takeover candidate, and a profitable platform growing ARR at 22% with a sub-eight-times-sales multiple is the kind of asset a larger security or infrastructure company could rationally buy. We do not underwrite a buyout. We note that the same cheapness that makes it attractive to us makes it attractive to an acquirer.

We are long SentinelOne common stock, opened 23 October 2026, for a three-to-five-year horizon. Shares, not options. The disclosure block at the top of this note carries the terms.

Sizing matters more here than on the other two names we initiated this week. This is a Constructive stance, not a Speculative one, because SentinelOne is not a going-concern story: it has real recurring revenue, net cash and improving margins, so the risk is underperformance and dilution, not zero. But it is the riskiest of our three new longs, and under our own framework Constructive means the position is taken with conviction and sized to be added to on weakness, not maximised on day one. We opened it after a year in which the stock ranged from about $12 to $26 and currently sits around $24, so we are buying in the upper half of its own range on the view that the fundamentals and the valuation gap justify it, not because it is at a low.

This section is longer than usual because this name earns it.

Stock-based compensation and GAAP losses mean dilution. SentinelOne lost roughly $450 million on a GAAP basis over the trailing twelve months, and a large part of that is stock-based compensation. That compensation is a real cost paid in shares, so the share count grows, and a holder's slice is diluted over time even as the business grows. The adjusted operating margin turned positive, but the gap between adjusted and GAAP profitability is the single most important thing to watch, because it is the difference between a compounder and a treadmill. This is the primary risk and we are not minimising it.

Enterprise momentum has slowed. Customers paying over $100,000 a year rose by only 13 in the quarter to 1,715, which is a soft number for a company that needs to keep landing large accounts to justify the thesis. If the enterprise motion stalls, the ARR growth rate falls and the valuation-gap case weakens.

It cut profit guidance. In August the company trimmed full-year adjusted EPS guidance by roughly 11% even as it nudged the revenue outlook up, and free cash flow was negative that quarter. The market read a company spending to grow with less to show on the bottom line, and that read is not unreasonable.

The competition is ferocious and better-capitalised. CrowdStrike is larger, profitable and consolidating the category, and Microsoft gives Defender away inside bundles enterprises already pay for. SentinelOne has to win on architecture and price against two opponents who can outspend it, and the AI-SIEM push puts it into Splunk's and Microsoft's territory as well.

Finally, 7.6 times sales is cheap relative to CrowdStrike but it is not cheap in absolute terms, and the forward earnings multiple is high. If software multiples compress or growth disappoints, the stock has room to fall well before the three-to-five-year thesis is disproven. We expect volatility and we are sized for it.

The open questions we consider the next highest-value work on this name.

1. 01

   The GAAP-to-adjusted gap, and share count. Track stock-based compensation as a share of revenue and the growth in diluted shares. This is the number that decides whether the thesis compounds or dilutes away.

2. 02

   Net revenue retention and $100k-plus customers. These show whether the enterprise motion is reaccelerating or stalling. The slowdown to +13 large customers is the figure to watch next quarter.

3. 03

   Purple AI and AI-SIEM traction, in dollars. The re-rating story depends on the AI products becoming a disclosed, growing revenue stream rather than a demo. Press for a number.

4. 04

   The path to durable free cash flow. Adjusted operating margin is positive; free cash flow was negative last quarter. Model when FCF turns consistently positive, because that is when the valuation gap has the best chance of closing.

5. 05

   Benchmark the gap to CrowdStrike continuously. Track the price-to-sales ratios of both side by side. The thesis is convergence; the data either shows it happening or it does not.

• SentinelOne Q2 CY2026 results, reported 27 August 2026
• ARR, revenue growth and billings
• Adjusted operating margin and free cash flow
• Customers over $100k ARR
• Full-year guidance revision
• Purple AI and the AI-SIEM
• CrowdStrike valuation comparison
• S price, market capitalization and cash
• S short interest
• Lenovo and channel partnerships