TEAM

Avoid at $171 a share against $223.40. The price is 31% above our value, past the line (0.85 times the price) below which we avoid. The Street's revenue path is ours; the price needs a higher margin or a richer exit multiple than we can support. The call rests on one assumption the company has not confirmed: that after FY2027 it keeps paying cash for about half of each cut in stock pay, as it does in FY2027. If later cuts cost no cash, the value is $205 and the call is Hold; the line is a 24% replacement share. The blind draft said Hold at $201; two outside reviewers showed that its stock-pay cut was not paid for in the margin and that Atlassian's own history caps its long-run return (section 9). The shares have roughly tripled in six months (from about $69 on 2026-04-27, research-check), so the market has already moved from pricing the bear toward our central case and a little beyond it.

The answer on AI. Frontier labs, Anthropic included, are a real threat to Atlassian's seat-based pricing but, on the evidence through October 2026, not to its products: agents that write code and resolve tickets need somewhere to receive work, keep context and record what they did, and today they plug into Jira and Confluence (Atlassian's interface for outside agents passed one million monthly users in the June quarter) rather than replace them. The threat lands if agents do the work without anyone buying a seat and Atlassian cannot charge for agent activity; that world is our bear case, worth $63 a share, and the reason our range is so wide.

Terms used throughout: the central case is our most likely outcome; the bear and bull cases are coherent worlds below and above it. The long fade values the business after 2031 by letting growth and the return on new investment drift to mature levels over 15 years. The outside view values 2031 operating profit at the multiple similar companies have fetched. Owner free cash flow is cash after stock pay, acquisitions and working capital. ARR is annual recurring revenue.

1. What Atlassian sells, who pays and what limits it

Atlassian sells software that teams use to plan and track work (Jira, more than $2.5B of ARR), write and share knowledge (Confluence, more than $1.5B), run help desks and IT service (Jira Service Management, inside the Service Collection, more than $1B of ARR growing about 30%), record video messages (Loom), and an AI layer across all of them (Rovo). The ARR figures are management's, from its May 2026 Investor Day. About two-thirds of Jira and Confluence users are not engineers (CFO, Q4 FY26 call).

Customers pay mostly per user (seat) per month, more for higher editions. Since 2025 Atlassian has also sold Collections: bundles such as the Teamwork Collection (Jira, Confluence, Loom and Rovo) with about ten times the AI credits of the standalone apps. Rovo comes with paid editions up to a credit allowance (10-K FY26); credits beyond it are one of the usage meters Atlassian lists as available from 2026-09-01, with billing for most meters starting 2026-12-03 (atlassian.com/licensing/usage-based-pricing, read 2026-10-26; the page gives no prices). So AI earns money through edition upgrades, Collections and, from December, metered overage.

FY2026 (the year to June 30) revenue was $6,572M, up 26%: cloud $4,411M (+28%), Data Center $1,831M (+25%), marketplace and other $331M (10-K). Data Center is the version customers run on their own servers. In September 2025 Atlassian announced its end of life: no new customers after March 2026, no new licences or expansions after March 2028, end of support in March 2029, with about three years of extended maintenance for some (8-K 2025-09-08). That pulled licence revenue forward into FY2026, so FY2027 revenue grows about 13% on the company's guide while cloud grows 25.5%.

Profit is the weak spot. GAAP operating income was $10M in FY2026 against a non-GAAP $1,996M; the gap is mostly stock pay ($1,605M, 24% of revenue) and $285M of restructuring after a 10% cut to the workforce in March 2026 (8-K 2026-03-11). We count stock pay as a cost. The FY2027 guide is a 4.5% GAAP operating margin with stock pay at 19% of revenue.

What limits it: how many people use work-tracking and knowledge tools, what each will pay, and how much of a customer's work runs through Atlassian rather than Microsoft, Google, ServiceNow, GitHub or GitLab.

2. Recent results beat the guide every quarter

Over the eight quarters to June 2026, revenue beat the midpoint of the company's guide by a median 3.05% (range 0.6% to 6.6%), cloud growth by 3.8 points and the non-GAAP operating margin by 3.2 points (output.txt, from nine earnings releases). The last two beats were the largest, mostly Data Center licence revenue pulled forward by the end-of-life announcement (about $50M more than expected in Q3, CFO).

The underlying signals were strong (management claims unless marked): cloud net revenue retention "north of 120%", up for three or four quarters; remaining performance obligations (contracted revenue not yet recognised) up 44% to $4.8B (reported); record $1M, $3M and $5M deals; cloud growth up from 26% to 31% over two quarters (reported); and "no signal of seat compression" (CEO, Q3 FY26 call). The weak points: subscription ARR growth guided down from 23% to 18% for FY2027, which the CFO puts down to prudence and lapping the DX acquisition (about 1 point), and stock pay still about 19% of revenue.

3. Is AI a threat to Atlassian? Product by product

We read Anthropic's registration statement (S-1, filed 2026-10-26) for this question, alongside Atlassian's filings and calls.

What the S-1 says, and does not. It names neither Atlassian nor any of its products. It calls Anthropic primarily a platform others build on, and says that as it builds its own applications it "may from time to time compete with enterprise software and application companies in select categories" (fact, competition section). Its risk factors call the conflict with customers "particularly acute with software-as-a-service and other software companies" (fact, p. 56). Its applications are Claude Code (a coding agent), Claude Cowork (a knowledge-work agent, being merged into Claude.ai from September 2026), Managed Agents and vertical solutions (fact). It says Claude wrote over 90% of the contributions to Anthropic's own code in September 2026 (company claim). It describes no issue tracker, wiki or service desk. Read plainly, Anthropic builds the agents that do the work, not the system that assigns and records it, and says it may move into categories it does not name. We read no more into it than that.

Atlassian and the labs today. Atlassian lists Anthropic, OpenAI and Cursor among the platforms it integrates with (10-K FY26), lets customers assign a Jira issue to Claude or Cursor and get a draft code change back (Q4 FY26 letter), and runs its own AI on several labs' models through a gateway (10-K). Its MCP server and command-line tool, the interfaces other agents use to read and write Atlassian data, passed one million monthly users in the June quarter with calls up more than 400% (letter). Management says 98% of MCP users are also active in Jira's own screens in the same month and that MCP adopters grow ARR twice as fast (letter, management claims). For now the labs are channels into Atlassian more than rivals.

Seats against agents doing the work. This is the real risk. Atlassian sells seats; if a team of ten engineers becomes six engineers and a set of agents, Atlassian loses four seats unless it charges for the agents. Its answers so far are bundles with credits, usage meters whose billing starts in December 2026, and a commitment model ("Flex") in trials (Investor Day). None has revenue reported yet. The CEO argues the opposite effect: cheaper software means more software and more builders, including non-engineers (Q4 FY26 call). Seat counts are not disclosed, so neither story can be checked; what is visible is cloud growth rising to 31% in the June quarter with seats named as a driver.

Do AI coding agents shrink or grow developer seats? Not yet visibly in Atlassian's numbers. Engineers are about a third of Jira and Confluence users, so even a large fall in developer hiring touches about a third of the seat base directly. We have no dated, sourced measure of developer employment and use none.

Will labs or their partners build issue tracking, knowledge or service management? Microsoft (with GitHub), Google and ServiceNow already sell products in these markets, and GitLab and Linear compete in tracking (our knowledge of the market, not sourced in this report). A lab entering would add a rival in a market Atlassian already shares. The larger risk is not a lab-built Jira but work moving to agent screens that treat Jira as a database behind an interface, which caps what Atlassian can charge.

Data Center end of life is a separate, dated risk. Customers must move or leave by March 2029. Management says retention is "outperforming" and 93% of migrants land in Premium or Enterprise (Q3 FY26 call; Investor Day). A forced move is also when a customer can switch vendor, so the bear loses 25% of Data Center ARR with no price uplift.

How the threat enters the forecast

Each lever is a driver with a range set from history and dated evidence (drivers.json), not a cut to the result:

Net revenue retention is not a separate input: the seat and price paths imply cloud base growth of 19% falling to 12%, against management's "120%+" retention today (its definition and base are not disclosed with a number).

From the central case to the bear, one lever at a time (output.txt): seats −$48, price per seat −$17, Data Center loss and uplift −$7, marketplace −$2, non-GAAP margin −$17, stock pay $0, acquisitions +$2, then the bear's lower exit multiple −$18, ending at $63. Seats are most of the threat's cost.

4. The forecast and the value

Central case. Revenue grows 15% in FY2027 (the guide is 13%; we add the median beat to cloud and take Data Center at the guide), then 13-15% a year to $12.7B in FY2031 as cloud absorbs the migrating Data Center base and organic growth slows. Stock pay is $1,411M in FY2027 (the guide's 19% of guided revenue, 18.7% of ours) and falls to 12.5% of revenue by FY2031. Atlassian is cutting it partly by paying cash instead: in FY2027, 3.0 of the 5.4-point cut (55%) moves into cash pay (Q4 FY26 letter), and we carry the same share into each later year's cut. The non-GAAP margin before that cash rises from 28.2% (the guide plus the median beat) by 1.5 points a year; after it, the reported non-GAAP margin reaches 30.8% and the GAAP operating margin goes from 8.0% to 17.3%. Acquisitions take 4% of revenue a year (10.6% over FY17-26). Owner free cash flow, all paid out, rises from $422M in FY2027 to $1.5B in FY2031. Shares stay at 253.1M; stock pay is charged as a cost, not as dilution.

Where we differ from the company: 1.8% above its FY2027 revenue guide (its median beat), and our FY2031 margin needs operating leverage it has promised ("accelerate and grow our GAAP operating profit", Investor Day) but not quantified past FY2027.

Cost of equity 9.64%: risk-free 5.29% plus beta 1.04 times Damodaran's 4.20% premium (2026-10-01). Beta is the median Blume-adjusted beta of twelve software peers; Damodaran's software beta (1.28) gives $163. Atlassian's own beta was not read (its data source carries the price).

The exit, and the range of defensible choices. The outside view uses the method's default, the median EV/EBIT over fiscal 2016-2025 of the peers with five profitable years: Adobe, Intuit, Autodesk, Dropbox and Zoom, 41.2x. The rule leaves out Atlassian's own history, because it has fewer than seven profitable GAAP years, so the multiples it earned while growing 20-40% a year never set its exit. The peers' multiples were earned at a median 13.6% revenue growth, close to our 11-13% in FY2031-32, so we make no growth adjustment. Other defensible choices, with the operating case fixed (exit_choices.txt):

The exit choice alone spans $123-$235 around $171. In the central case the outside view is 192% above the long fade. The multiple drives the gap: 41.2x operating profit is what a decade of low-rate software markets paid, while the fade prices Atlassian at about 17 times year-six after-tax profit from its own growth and a long-run return at the cost of capital. We keep the method's mean of the two; the 1996-2009 era and the 10-15% growth band both land at $157-159, 7-8% lower, and the 2010-2021 era's high ratio we treat as that era's level, not as evidence about Atlassian.

Long-run return on new capital follows the method's rule: each case keeps the capital per revenue dollar it reaches in FY2031, capped at the company's median five-year return on the same basis and floored at the cost of capital. Atlassian's six five-year windows ending FY2021-26 on US GAAP operating profit (restated from FY2021 in the FY2023 10-K; earlier years were IFRS) run from −43% to +11%, median −3.6% (hist_returns.txt), so the floor binds: 9.64% in every case. The blind draft said the history could not be measured and applied no cap; one reviewer showed it can (section 9).

The weighted value ($194) is above the central ($171) because the bull is further from the central than the bear: more upside in margin and exit multiple than the bear has downside once its revenue stalls. It is still below the price.

5. The market pays for a richer margin or exit than ours

This section was written after the price, targets and estimates were read (unblinded), and revised after the reviews in section 9; the model changes there rest on company documents, not on the price.

At $223.40 our central case earns 3.3% a year to the exit, below the 9.64% cost of equity. Holding everything else at our central, the price needs one of: paid-seat growth 6.2 points a year above our path, a non-GAAP margin 5.2 points higher, or an exit at 1.46 times the peers' multiple (about 60x operating profit). Price per seat at the top of its range reaches $222; no cash replacing stock pay, $205; stock pay 6 points lower, $198; Data Center, migration uplift and acquisition spend cannot reach it either (odds.json). 30% of our simulated values sit above the price.

The Street's revenue path is ours: $7.5B in FY2027 and $12.4B in FY2031 against our $7.56B and $12.67B, never more than 3% apart in any year (market.txt). So the gap is not growth. Its average 12-month target, $209.67, discounted a year at our cost of equity is $191, and our value is 0.89 times that; the eight targets set in the last month average $244.88, up from $178.68 over the year, as the shares rose. The market is paying above our whole defensible exit range: about 60x sits beside the 2010-2021 era's 64.0x ($235) in section 4. We keep the default because the peers earned 41.2x at growth close to ours; what would move us is evidence on seats and on how Atlassian cuts stock pay, not the multiple. The 2026-11-05 print settles part of it: cloud growth at or above our 32.2% with seats named as a driver supports the bull's path; below 30% the bear's.

Alpha's view, 2026-10-26. Alpha, the quantitative model, scores Atlassian 91 of 100 (tier Strong), with momentum A, expectations B, growth B, quality C and valuation D; it gives no rank today and flags its own data as limited, because GAAP profit is near zero and goodwill jumped with recent acquisitions (alpha ticker analyze TEAM). Alpha's strength is the six-month price move and earnings surprises; its D on valuation agrees with ours that the price is ahead of value. We differ on what the beats mean: Alpha reads them as momentum, we read the last two as Data Center licences pulled forward by the end-of-life announcement.

The chart, as timing only. A technical quick read on 2026-10-26 grades the chart B: an advancing trend above a rising 30-week average ($131), above the 50-day ($192) with the last swing low ($168, 2026-09-29) holding, 1.7 average daily ranges above the 21-day average; results are 8 sessions away (2026-11-05 after the close, confirmed by market data), so any fresh entry waits for that event. It falls to C on a close below $192. The grade says when, never what Atlassian is worth, and it moved no number here.

One change to drivers.json. Running the odds stopped on the bottom of the acquisition driver's solve range: at 0% of revenue the model's net reinvestment turns negative and the method cannot value it. The range the solver searches was narrowed from 0-10% to 1.2-10%; the distribution the simulation draws (2-6%, median 4%) is unchanged, and at 1.2% the value is $181, still short of the price, so the answer is the same either way.

6. The bear case: the threat lands

In the bear, agents take on work in Jira, Confluence and the service desk through Atlassian's own interfaces and the labs' tools; teams shrink and seats fall 2-3% a year from FY2030; price per seat grows 2% a year as Microsoft, Google and the labs bundle AI free; and a quarter of Data Center customers leave at end of life. Revenue stalls near $8.1B from FY2029. Atlassian responds as it did in March 2026 by cutting cost: GAAP margin reaches 14.1% on a reported non-GAAP 27.5% and stock pay 12.5%. Its long fade starts from shrinking profit, and the outside view takes the software group's 25th-percentile outcome (28.6x). Fair value $63.

Why the other side could be right, for a lower value. Our central seat path assumes a slowdown, not a reversal, and the only evidence on seats is management's word. If every later point of stock pay cut is replaced by cash, rather than our 55%, the central falls to $143. The 55% is the FY2027 share carried forward; the company has not said what later cuts will cost in cash, and the other side of that assumption is section 5's Hold at $205. The Data Center pull-forward flatters FY2026, and subscription ARR growth is guided down five points.

Why the other side could be right, for a higher value. Cloud growth rose to 31%, retention rose, and Atlassian beat its revenue guide in each of the last eight quarters. If agents create more tracked work and Atlassian charges for it (the bull), revenue reaches $15.0B and the value $373. One reviewer's combination of the arguments for a higher value (no cash replacing stock pay, metered Rovo adding a point of price a year and a 5% risk-free rate) reaches $218, still below the price.

The de-rating path. If our central forecast holds but the market keeps paying the bear's exit multiple (18.2 times FY2031 GAAP operating profit, all in), the stock is worth $118 today (output.txt). What would keep the multiple there: years of flat or falling disclosed seats, or net revenue retention below 110%; a lab or platform shipping a tracker or service desk that customers adopt instead of Jira; or Rovo usage raising Atlassian's inference cost faster than its prices. Without that evidence, a growing, cash-generating Atlassian at 18.2x would be a de-rating the forecast does not explain.

7. Next results preview: Q1 FY2027, reported 2026-11-05 after the close

Date: Atlassian's release of 2026-10-05 on Business Wire, read through a search summary (the page was blocked); call at 2:00 p.m. Pacific. Each forecast is the company's guide plus its median beat over eight quarters; ranges run from its smallest to its largest beat (output.txt).

The figure that moves the value is cloud growth, the visible part of seats times price. Each $1B of FY2031 revenue above ours is worth about $20 a share; a sustained extra point of seat growth adds about $7.

8. Catalysts and falsifiers

• 2026-11-05, Q1 FY2027 results. Cloud growth below 30% (the guide plus its smallest beat in two years) or any statement that seats are shrinking moves weight to the bear.
• Seat or agent pricing disclosure. A paid-seat count, or a price for agent work (Rovo credits sold beyond the bundles, Flex commitments), tests the central's price-per-seat path.
• March 2028 and March 2029, Data Center deadlines. More than 20% of Data Center ARR lost falsifies our 10%.
• Cash replacing stock pay (the call's hinge). The 2026-11-05 print (cash pay in operating expense against the stock-pay decline) and the FY2028 guide (August 2027) show whether later stock-pay cuts cost cash. A FY2028 non-GAAP margin guide that rises by most of its stock-pay cut, so that less than a quarter of it returns as cash, moves the call to Hold ($205 at none).
• Stock pay above 17% of revenue in FY2028 falsifies the margin path (about $9 a share per 2 points). Whether the cut shows up as cash pay (FY2027: 3.0 of 5.4 points) or as margin is worth $62 between its ends.
• Labs' products. Anthropic, OpenAI or a cloud platform selling an issue tracker, wiki or service desk to enterprises, or Atlassian losing its agent integrations, moves weight to the bear.

9. What the reviewers found and what we did

Two reviewers read the blind draft after the market pass: Codex, auditing the model, and Opus, arguing for a higher value. Each change below rests on a dated company document, not the price, and is reproduced one step at a time from the blind $201 in review_answers.txt. Codex reran after the call changed (rows 11-14). Every change moved the value away from the price, and together they turn Hold into Avoid.

| 11 | Codex, round 2 | Avoid stands, but it depends on extending the 55% cash replacement past FY2027, which the letter quantifies for FY2027 only | Disclosed in the decision block, the opening, the bear section and as a falsifier. With no later replacement $204.56; below 23.93% replacement the call is Hold | none | | | 12 | Codex, round 2 | The return windows mixed measures: FY2021 used $67.6M operating income where restated US GAAP is $141.4M | Fixed. Rebuilt on US GAAP from EDGAR XBRL (FY2021 on), six windows, median −3.6% | Median 0.3% → −3.6%; floor still binds | $0 | | 13 | Codex, round 2 | FY2031 central non-GAAP margin is 30.79%, not 30.3% | Fixed in the text (bear 27.6%, bull 33.6% corrected too) | 30.3% → 30.8% | $0 | | 14 | Codex, round 2 | The combined upside case is $218.37, not about $216 | Fixed (review_answers.txt) | $216 → $218.37 | $0 |

Net: $200.64 → $170.65 central, $74.64 → $63.17 bear, $445.97 → $372.54 bull. Codex's own estimate of finding 3 was −$8.60 on the blind model and Opus's of finding 9 +$12.75; our figures are the step on the revised model.

10. Sources

• Atlassian 10-K FY2026, filed 2026-08-14; 10-Q Q3 FY2026, filed 2026-05-01; 8-Ks of 2025-09-04, 2025-09-08 (end-of-life blog and investor deck), 2025-09-18 (DX), 2026-03-11 (restructuring), and the earnings releases of 2024-08-01 to 2026-08-06. SEC EDGAR, fetched 2026-10-26, masked for price before reading (data/). Primary.
• Q4 FY2026 shareholder letter, 2026-08-06, EDGAR exhibit images read by OCR. Primary.
• Q3 and Q4 FY2026 earnings calls and Investor Day 2026-05-06, Market data/Quartr transcripts. Secondhand transcripts of primary events.
• Anthropic, PBC Form S-1, filed 2026-10-26, and our read of it (../2026-10-26-anthropic-s1/report.md). Primary.
• Market data statements, annual and quarterly, fetched 2026-10-26, checked against the 10-K; market data fiscal-year-end history for the outside view (data/exit-anchor.json) and peer betas (data/peer-betas.json).
• Damodaran implied equity risk premium and risk-free rate, 2026-10-01.
• Results date: Business Wire, 2026-10-05, via a search summary. Secondhand.
• Atlassian usage-based pricing page, atlassian.com/licensing/usage-based-pricing, read 2026-10-26. Primary.
• Model: model.py (output.txt), exit_choices.py (exit_choices.txt), hist_returns.py (hist_returns.txt), review_answers.py (review_answers.txt), drivers.json, assumptions.md.

This report was written with Claude, a model made by Anthropic, whose threat to Atlassian it assesses: a conflict of interest.