Palantir

IThe business

Palantir sells software that puts an organization’s data, its rules and its AI models into one system. Governments use it to plan and run operations. Companies use it in industries from airlines to insurance to healthcare.³ It sold to 1,049 customers in the twelve months to June 2026 and employed 4,429 people at the end of 2025; 2025 revenue per employee was $1.01M C (calculated).⁴³

The company has four platforms. Gotham serves defense and intelligence agencies. Foundry is the data operations platform for everyone else, and many agencies use it too. Foundry holds the Ontology, a model of the customer’s people, assets and processes. AIP connects third-party language models to that model with access controls and audit logs. Apollo delivers all of it into any environment.³

The sale starts with a bootcamp. Palantir builds a real workflow on the customer’s own data in days, often at its own expense, and grows from there.³ The 10-K says the speed of experiment is what “makes the software stick.” Growth then comes from inside the account. In 2025, $774M of the $833M increase in government revenue and $425M of the $777M increase in commercial revenue came from customers already on the books a year earlier.³

In the twelve months to June 2026 gross margin was 84.8%, research and development 10.4% of revenue, sales and marketing 20.1%, and capital spending 0.7% C (calculated).⁴ Property and equipment net was $61M.

Revenue grew from $595M in 2018 to $4,475M in 2025, 33.4% a year, and to $6,156M in the twelve months to June 2026.⁷³⁴

Who pays

Government customers were 52% of revenue in the first half of 2026 and commercial customers 48% C (calculated).⁴ The U.S. government alone was 42% of revenue, $1,496M of $3,568M C (calculated).⁵²⁷ No customer was above 10% of revenue. The top three together were 16% and have been customers for fifteen years on average. The top twenty paid $124M each over the twelve months, up 67%; that is about $2.5B, 40% of revenue, against about 44% a year earlier, and the set is ranked afresh each period C (calculated).⁴ One customer was 27% of accounts receivable at 30 June 2026 and 25% at the end of 2025. The filing does not name it or say why it holds that share while no customer is above 10% of revenue, and I cannot tell timing from a slow payer.⁴

The 10-K describes the target: large, hard projects that others have failed at. It gives the example of a U.S. military department that spent more than $1B on an enterprise system that was never delivered.³

We actively seek them out.

Where the growth is

The growth is American. In the first half of 2026 revenue from U.S. customers was $2,855M, up 110% and 80% of the total. Revenue from the rest of the world was $713M, up 35%.⁴ In the second quarter U.S. commercial revenue was $764M, up 149%, and U.S. government revenue $809M, up 90%.⁵ On the call the company put international commercial revenue at $182M, up 26%, and international government revenue at $181M, up 42%; those figures are in a third-party transcript.³⁹ U.S. commercial revenue was $335M in 2022 and $1,465M in 2025, 4.4 times as much C (calculated).¹⁴²⁶

U.S. commercial customers numbered 653 in the second quarter, up 35%, according to the chief financial officer on the call; the figure is in a third-party transcript and not in the filings.³⁹ Customers rose 24%, from 849 to 1,049, over the year to June 2026 while revenue rose 79%, so revenue per customer rose from about $4.1M to $5.9M, up 45% C (calculated).⁴³ The company counts agencies of one government as separate customers, which blurs the customer count.⁴

Backlog

$2.1B of the next twelve months’ revenue is in non-cancelable backlog, 43% of the $4.9B of remaining performance obligations at June 2026 C (calculated).⁴ My year-one revenue is $10.4B, so about 20% is contracted and 80% depends on renewals, expansions and new wins E (estimate). The ratio understates coverage, because the figure leaves out contracts of twelve months or less.⁴ Remaining performance obligations were $1.7B in December 2024.¹² The contracts closed in 2025 are expected to bring revenue for four more years on a dollar-weighted basis, if every option is exercised.³ U.S. commercial remaining deal value, a company measure that assumes every option is exercised, was $6.2B.⁵ Most contracts can be ended for convenience, and the U.S. federal government cannot exercise options more than a year ahead.³ The company excludes $12.3B of government IDIQ awards whose funding is not fixed.³

Palantir reports no revenue line for AIP. U.S. commercial revenue, customers and contract value are the visible markers, and any AIP figure would be my estimate.⁵

IIThe moat

I think Palantir’s moat is switching costs: an embedded system of record for an organization’s operations, backed by the accreditations that government buyers require. It is wide in government and young in commercial. The evidence so far points to widening, and I have not seen how it holds in a downturn.

The product holds the customer’s logic, not only its data. Once an air force or a bank has built its workflows on the Ontology, a rival has to rebuild the model of the operation. The top three customers have stayed fifteen years on average, and growth comes mostly from accounts already on the books.⁴³

In government a second barrier sits on top. The 10-K names FedRAMP and Impact Level 2, 4, 5 and 6 authorizations among those the business must keep.³

Evidence it is widening, and what it does not prove

Standardization in government. The U.S. Army agreed in July 2025 to fold 75 contracts into one enterprise agreement worth up to $10B over up to ten years. The Army calls that a ceiling, not an obligation.³⁶ That shows standardization. It also gives the buyer one negotiation, and I have no evidence of the price paid. The Defense Department raised the ceiling on the Maven Smart System contract by $795M in May 2025, to nearly $1.3B through 2029, and NATO’s agency finalized a contract for the system on 25 March 2025 at an undisclosed value, reported on 14 April.³⁸³⁷

Contracted revenue. Remaining performance obligations rose from $1.7B to $4.9B in eighteen months, 2.9 times, while revenue rose 2.1 times C (calculated).¹²⁴

Evidence it is narrowing, or untested

Third-party models. The 10-Q describes AIP as secure connectivity to third-party language models and does not describe a model that Palantir owns.⁴ A customer can reach the same models without Palantir. What Palantir sells is the model of the operation around them.

Clouds reach the same accounts. On its 29 July 2026 call Microsoft reported 100,000 Foundry customers with revenue more than doubling, over 40,000 paid Fabric customers, and nearly 40 million agents registered in Agent 365 across tens of thousands of companies.⁵¹ Those counts are for other products and are not comparable with Palantir’s 1,049 customers. They show that a cloud with a contract already in place can offer a platform for building agents on a customer’s data to the same buyers.

Cloud dependence. In March 2026 Palantir committed to spend at least $5.6B on a third-party cloud host through February 2036.⁴

Outside the United States. International commercial revenue grew 26% in the second quarter, about one-sixth of the U.S. rate C (calculated).³⁹⁵

Where the moat did not hold. Palantir signed contracts with companies it invested in during the SPAC boom. Their total value, the revenue recognized plus the remaining deal value, fell from $493M in December 2022 to $326M in June 2026, because the value of contracts with investees that went bankrupt or ended their contracts is left out. Palantir has recognized $325M of the $326M, and $3M of it in the first half of 2026.¹⁰⁴

The moat protects an account once Palantir has won it: existing customers supplied 98% of the second-quarter rise in government revenue ($428M of $437M) and 82% of the rise in commercial revenue ($407M of $495M), and new wins added the rest C (calculated).⁴

IIIThe competitive advantage

Palantir grows 79% at a 43% GAAP operating margin; no company in the tables below does both. I think the cause is the order of the sale: Palantir builds a working workflow on the customer’s own data before the contract, at its own expense, and a rival that waits for a signed contract starts behind.³ The 10-K names no competitor. It lists large enterprise software companies, government contractors, system integrators and customers’ own builds.³ I compare Palantir with the contractors on that list and with the software companies it is priced against.

Start with the contractors, because the U.S. government was 42% of Palantir’s revenue in the first half of 2026.⁵²⁷

Palantir grew 79% in the twelve months to June 2026 at a GAAP operating margin of 43%. Booz Allen’s revenue fell 6% at a 9% margin. Leidos grew 4% at 12% and CACI 10% at 10% C (calculated).^4464748 The market pays 1.2 to 1.9 times revenue for the contractors and 70.4 times for Palantir on basic shares, or 75.3 times on the diluted shares I use in the valuation. For operating income the multiples are 11 to 20 times for the contractors and 164 times, or 176 times diluted, for Palantir C (calculated).⁵⁶⁴⁴⁹ Together the three contractors took in $38.3B of revenue in the last twelve months and have an enterprise value of $54B. Palantir took in $6.2B and has an enterprise value of $433B on basic shares and $464B on diluted shares C (calculated).^446474849 A trailing multiple says little about a company growing 79% to 93%. On the 2026 revenue guide of $8.154B the diluted enterprise value is 57 times, and on my year-one revenue 45 times C (calculated).⁵⁶⁵

Among software companies the picture is more balanced. Snowflake, ServiceNow, Datadog and CrowdStrike grow 22% to 32% and trade at 11 to 48 times revenue.^40414243 Add growth to GAAP operating margin and Palantir scores 122, against 34 for ServiceNow, 32 for Datadog, 22 for CrowdStrike and 9 for Snowflake C (calculated).^440414243 Divide each revenue multiple by its growth rate and the scores are ServiceNow 0.50, Snowflake 0.65, Datadog 0.74, Palantir 0.89 and CrowdStrike 1.98 C (calculated). Palantir is second-highest of the five, and the measure takes today’s growth as given, which is the thing in question.

Where Palantir wins

It wins large, hard projects. The 10-K says AIP bootcamps deliver real workflows on actual customer data in days, and it calls them shorter, more cost-effective programs than the months of pilots its sales model once required. It also says the sales cycle often lasts six to nine months, and that bootcamps may result in no or minimal future revenue.³ The other half of the pitch is trust. Government customers choose a supplier that holds the authorizations, and the proxy statement says Palantir has worked with the U.S. government through multiple administrations. The top three customers average fifteen years.⁶⁴

Where it loses

Palantir discloses no win rate. I found no sourced loss to a named rival, so this section rests on the 10-K’s risk language: multi-award IDIQ contracts can create pricing pressure and require repeated bids, and software that customers build for themselves is a rival.³ The one case of contract value falling is the SPAC-investee contracts, under the moat.¹⁰⁴ Databricks and Anduril, two private names often set beside Palantir, file no accounts that I could source, so I leave them out of the numbers.

Add revenue growth to GAAP operating margin and Palantir scores 122, against 34 for ServiceNow, the next highest of the software peers C (calculated).^440414243

IVManagement

Palantir has had the same leadership for most of its life. Alexander Karp has been chief executive since 2005 and a director since 2003. Stephen Cohen is president and a director. Peter Thiel is chairman. Shyam Sankar has been chief technology officer, David Glazer chief financial officer and Ryan Taylor chief revenue officer and chief legal officer. The five executive officers have been with the company 13 to 23 years.⁶

Who holds the votes

Class F is the control device. It tops up the votes of Karp, Cohen and Thiel, acting by majority, to 49.999999% of the total while they and their affiliates hold at least 100 million equity securities.⁶⁴ Together they hold about 9% of the shares, counting options they can exercise within 60 days C (calculated).⁶ Karp’s 57.9M record shares are 2.4% of the shares outstanding and 11.4% of the votes C (calculated).⁶⁴ In 2026 all three shareholder proposals failed, and the advisory vote on pay passed with 85% of the votes cast, counted with the Class F top-up.³⁰

Pay and incentives

Karp’s pay in 2025 was $8.6M, and $7.5M of it was other compensation: $2.5M of personal security, $2.0M of personal travel on chartered and owned aircraft, a $2.4M antitrust filing fee that the company paid on his behalf, and $0.6M of tax services, health cover and umbrella insurance. He has had no new equity award from 2023 to 2025.⁶

His incentive is the 2020 grant: 141M options at $11.38 that expire in August 2032 and vest 2.5% a quarter, plus restricted stock. At 31 December 2025, 77.55M options and 21.45M restricted shares had not vested. Three quarterly vestings followed, on 20 February, 20 May and 20 August 2026, which leaves about 67.0M options and 18.5M restricted shares. At the Q4 2026 close those carry $11.6B of option spread and $3.4B of restricted stock, on top of $10.7B in shares he holds of record C (calculated).⁵⁶⁶³² Pay “actually paid” under SEC rules, which marks those awards to market, was $11.09B in 2025.⁶

His stake ties him to the share price more tightly than any bonus formula could. Of his 6.4M record Class A shares, 3.0M are pledged as collateral for personal debt: 47% of that class and 5% of all his record shares, about $550M at the close C (calculated).⁶

Karp flies on non-commercial aircraft, which the proxy supports with an independent security study. He beneficially owns one, and the company either reimburses him or buys hours from the charter company that operates it, at rates that an independent review found at or below market.⁶ The company incurred $17.2M of expense on that aircraft in 2025, for personal and business travel by Karp and business travel by other staff, and the proxy counts $2.0M of personal travel on chartered and owned aircraft in his pay.³⁶

Shares sold since January 2025

Form 4 filings show what the leaders sold. Karp sold 3.03M Class A shares for $430M, an average of $141.94, between February 2025 and August 2026. $402M of that, 2.85M shares, was automatic selling to cover tax withholding when his restricted stock vested. The other 180,000 shares, $28M, were sold under trading plans.³² The plan he adopted on 12 March 2026 allows up to 7.08M shares through 12 December 2026, about $1.30B at the close. The Form 4 of 20 August 2026 reports 90,000 shares sold under it, and I found no other sale under it C (calculated).²⁸³²

Thiel’s holding company completed its plan on 2 March 2026, selling 2.0M shares for $290M. Cohen sold 6.43M shares for $648M, $368M of it under a plan in February and March 2025. Sankar sold 2.46M shares for $324M. The officers and directors who sold took in $1.90B, $856M of it from sales to cover tax C (calculated).³²

On guidance, management has under-promised: revenue beat the top of its range in each of thirteen quarters, as the table under Financials shows.⁵ The other officers are paid in stock. Glazer and Taylor each received about $28M in 2025, mostly restricted stock and stock appreciation rights struck at $150 that cannot be exercised until late 2033.⁶

The board

Seven directors serve, four of them independent. Thiel, a co-founder, is chairman. Directors are elected annually.⁶

Capital allocation

Palantir keeps its cash. It has no debt, pays no dividend and has made no material acquisition. Cash and Treasuries reached $9.2B by June 2026.⁴³ From 2021 to June 2026 the company put $450.5M into SPAC and PIPE investments and $139M into buybacks C (calculated).¹⁰¹¹³

The SPAC and PIPE investments of 2021 and 2022 cost $450.5M. Note 4 records $72.8M of net unrealized losses on these holdings in 2021 and $159.0M of unrealized plus $113.1M of realized losses in 2022, in other income. Some investees went bankrupt or ended their contracts. The company said in 2023 that it did not expect new SPAC investments.¹⁰

The buyback was small. The board authorized $1.0B in August 2023, the company spent $139M in 2024 and 2025, and the program ended in January 2026.¹¹³ Stock pay was $3.2B from 2021 to 2025, so repurchases offset 4% of it C (calculated). The share count rose 18.5% from December 2021 to June 2026, 3.8% a year C (calculated).⁷⁴ Palantir also paid $299M of employee taxes in 2024 and 2025 to settle awards net.³

The business needs little capital: capital spending was 0.7% of revenue.⁴ What management has not shown is what it will do with $9.2B and a free cash flow that now exceeds $3B a year.

Follow the pay. Karp’s 2020 options cost $11.38 and run to August 2032, so they pay him when the share price rises, not for how the $9.2B is used.⁶ His $10.7B in shares is large, but I can’t tell from the filings how much he paid for in cash. Since 2021 the company spent $139M on buybacks while the share count rose 18.5%, and 94% of what Karp sold was automatic selling to cover tax C (calculated).⁴³²

Karp’s stake ties him to the share price, with $10.7B in shares and $11.6B of unvested option spread, and the thin part of the record is capital allocation: $450.5M of SPAC and PIPE investments and $139M of buybacks since 2021, against $9.2B of cash C (calculated).^566101134

VFinancials

Revenue growth has come with rising margins. Revenue grew 56% in 2025 and 79% in the last twelve months. Over the last year operating income rose $2.07B on a $2.72B rise in revenue, a 76% incremental margin C (calculated). The year-earlier base includes the fourth quarter of 2024, when stock pay rose to $282M; on adjusted operating income, which leaves stock pay out, the incremental margin is also 76% C (calculated).^4521222324 The free cash flow margin went from 10% in 2022 to 47% in 2025 and 55% in the last twelve months, before stock pay C (calculated). The operating margin went from −8.5% in 2022 to 31.6% in 2025 and 42.8% in the last twelve months.⁷³⁴

Contribution margin, revenue less cost of revenue and sales and marketing before stock pay, rose from 14% in 2018 to 66% in 2025 and 74% in the first half of 2026. Government was 72% and commercial 75%. It leaves out research, administration and stock pay.⁸³⁴ Revenue per employee rose from $0.45M in 2020 to $1.01M in 2025, revenue over year-end staff C (calculated).⁸³ These measure operating leverage, not switching costs.

The quarter

Revenue in the second quarter of 2026 was $1,935M, up 93%. GAAP operating income was $912M, a 47.1% margin. Adjusted operating income, which leaves out stock pay and its payroll tax, was $1,194M, a 61.7% margin.⁵ The company guides 2026 revenue to $8.150B to $8.158B, up 82% at the midpoint, raised from $7.182B to $7.198B in February and $7.650B to $7.662B in May, and adjusted free cash flow to $4.5B to $4.7B.⁵²⁶²⁷

Palantir’s guidance has run below its results. Revenue beat the top of the range in each of the last thirteen quarters, by 0.2% to 8.7%, and by 5.7% to 8.7% in the last four.^272625245 A guide that has been beaten thirteen times is a floor.

Earnings quality

Net income of $3,017M is higher than operating income of $2,635M because of interest income, marks on holdings and a tax rate of 1.3%. At a 23% tax rate on operating income and interest, earnings are $0.86 a share against the reported $1.17, and the price is 213 times them C (calculated).³⁴⁵

The tax rate is low for a reason that will end. Federal loss carryforwards grew from $5.5B to $9.0B in 2025, and the company paid no federal income tax. Excess deductions on stock pay made most of the difference. The 10-K says that after those deductions the company was in a cumulative tax loss in the United States over the last three years, despite positive pretax income.³ I model a cash tax rate of 2% until a $28B shield is used, then 23% E (estimate). The shield is the $9.0B of carryforwards plus about 60% of the $32.2B spread on options and unvested restricted shares at the Q4 2026 price E (estimate).³⁴

That spread is measured at the market price. At my base value of $83 it is $13.7B and the shield about $17B, which lowers the base value by $0.7; with no shield at all the base value is $81 E (estimate). Of the $9.0B of federal losses, $8.3B carries forward indefinitely, and the tax code lets losses from 2018 on offset only 80% of taxable income.³⁵² I model 2% in shield years. A cash rate of 4% would cost $0.2 a share and 6% $0.4 E (estimate).

From profit to cash

Free cash flow was $3,358M in the twelve months to June 2026, a 55% margin before stock pay. After stock pay it was $2,523M, 41% of revenue.³⁴ The company’s own measure, adjusted free cash flow, added back the payroll tax on stock pay and was $3,476M.⁵

The balance sheet

Cash and Treasuries were $9.2B at 30 June 2026, with no borrowings and a $500M revolver, undrawn, that matures in March 2027.⁴ Receivables were $1,485M. Deferred revenue and customer deposits added $1.1B of cash received in advance.⁴

VIValuation

How this goes badly wrong

Working back from what would hurt an owner most, I see three ways.

1. Growth fades toward a contractor’s. Palantir grew 79% in the last twelve months.⁴ If growth in years two and three comes in at 25% and 15% instead of my 45% and 34%, the base value is $63 E (estimate). This is the most likely, and my bear case, $26, is built on it.
2. The government turns. The U.S. government was 42% of revenue in the first half of 2026 and can end contracts for convenience.³⁵²⁷
3. The margin slips. I have the operating margin at 49% to 52%, above Microsoft’s 46.8%.⁵⁰ At 46.8%, the base value is $77 E (estimate).

What an owner can take out

Reported net income in the last twelve months was $3,017M. An owner can spend only what is left after stock pay and the capital needed to stand still, and that capital is small, 0.7% of revenue.⁴ Owner free cash flow was $2,523M, and $2,260M without interest on the cash. That is 25% below net income C (calculated). The tax rate is 1.3% and will rise. At 23%, earnings are $0.86 a share against the reported $1.17 C (calculated).⁴

What I can’t judge

I can’t know whether Palantir becomes the default operating layer for large governments and companies. That is most of the gap between $26 and $235, and the filings can’t narrow it. Nor can I know how much stock pay the margin needs; $1.28B on existing awards is still to be expensed.⁴ So the $83 is the middle of a wide range, and 47% of it arrives after year fifteen.

What the price implies

At $184.11 the market values Palantir at $473B on 2.57B diluted shares, and the enterprise value is $464B.⁵⁶⁴ The shares fell 6.3% on 30 October, from $196.42, in a market sell-off: the Nasdaq Composite lost 2.9% and the Nasdaq-100 3.5%.^53545658

Palantir made $2.5B of owner free cash flow in the last twelve months. That is a yield of 0.5% on its market value, against 5.00% on a ten-year Treasury.⁴⁵³³ At last year’s rate it would take 187 years of owner free cash flow to earn back the market value C (calculated).

Solve for one growth rate and the price needs free cash flow after stock pay to grow 27.5% a year for fifteen years, from $2.26B to $87B, 38 times today’s.⁵⁶³⁴ If the same price must be earned in ten years, the rate is 38.7%. A constant rate is a crude shape, so the next table turns it into revenue.

What the price assumes

Value depends on two things, revenue in year ten and the margin on it. The grid builds each cell with the same model.

At my base-case margin the price needs $122B of revenue in year ten. At a 45% free cash flow margin it needs $103B, and at 50% $94B E (estimate). Microsoft’s revenue for fiscal 2026 was $331.8B.⁵⁰ ServiceNow, Snowflake, Datadog and CrowdStrike together took in $29.5B in the last twelve months.^40414243 The price asks Palantir to reach 37% of Microsoft’s fiscal 2026 revenue in ten years, from 2% today C (calculated).

The price is the value of $122B of revenue in year ten at my margins; my base case reaches $53B.

That is 34.8% a year for ten years, the same as my bull case. The price needs my bull case’s ten-year revenue path: that path, at my base-case margins, is worth $184 a share E (estimate).³⁴

The discount rate

I use 10%. That is close to the 10.4% that the capital asset pricing model gives for Palantir’s 60-month beta of 1.28, built from a 5.00% ten-year Treasury yield and Damodaran’s 4.20% implied equity risk premium C (calculated).³³³⁴³⁵ The beta moves a lot: 0.75 over 36 months, 1.28 over 60 and 1.89 since the September 2020 listing, which gives 8.2%, 10.4% and 12.9% C (calculated).³⁵ The other analyses in this section do not share one rate: Microsoft’s uses 9% and NVIDIA’s 10%. I use 10% here because it sits close to this company’s own 60-month rate, and I put the rest of the risk into the cash flows, where it is visible. At the 10.4% my own beta gives, the base value is $78 E (estimate).

Three cases

I build each case from revenue down to free cash flow after stock pay: revenue, a GAAP operating margin, a cash tax rate, capital spending of 1.0% of revenue against depreciation of 0.5%, and working capital at 10% of each year’s revenue increase, against about 12% in the last twelve months C (calculated).³⁴ Stock pay is a cost inside the margin, as under GAAP, so the 2.58B diluted shares, counted at the Q4 2026 price, are the whole claim on the cash. Year one runs from July 2026 to June 2027. It starts from the guided second half of 2026, $4.59B at the midpoint, plus an assumption for the first half of 2027.⁵ Cash flows fall at year end and values are dated 30 June 2026, the last balance sheet; mid-year timing adds about $4 to the base value and dating from Q4 2026 about $3 E (estimate).

The base case tracks the guide in year one and then decelerates to a normal curve.

• The second half of 2026 is $4.79B, 4.5% above the guide midpoint and 4.4% above the top of the range. Palantir has beaten the top of its range by 5.7% to 8.7% in each of the last four quarters, so this is a smaller beat than any of them E (estimate).^272625245
• The first half of 2027 is 17% above that, $5.6B, up 57% on the first half of 2026. Year one is $10.4B, up 69% E (estimate). The half-over-half growth halves in one step: the second half of 2026 is 34% above the first half in this case, 28% at the guide midpoint. With the guide’s own third quarter, the first half of 2027 needs about 7% growth a quarter, against 16% to 19% in each of the last three E (estimate). Held at 34% instead, year one is $11.2B, the same as the bull case’s $11.2B, and the base value is $89 E (estimate).
• Growth is 45% in year two and 34% in year three, then falls to 20% in year five, 8.5% in year ten and 4.5% in year fifteen. Revenue reaches $30.6B in year five, $53B in year ten, 8.6 times the last twelve months, and $70B in year fifteen E (estimate).
• The operating margin is 49% in year one, 52% in years three to five and 50% from year nine, against 42.8% in the last twelve months and 47.1% in the second quarter. Stock pay is inside it E (estimate).⁴⁵ That is above Microsoft’s 46.8% in fiscal 2026, and it assumes stock pay and other costs grow more slowly than revenue.⁵⁰ Held at 46.8%, the base value is $77 E (estimate).
• Cash tax is 2% until the shield is used in year four, then 23% E (estimate).

The bear case has growth fade toward a government contractor’s. Government budgets tighten and AI platforms from model makers and clouds cut prices. Revenue grows 55% in year one, 25% in year two and 15% in year three, reaches $21B in year ten, and the operating margin settles at 32%, below today’s 43% E (estimate). Free cash flow in year ten is $5.1B.⁴⁵

The bull case has Palantir become the default operating layer for large governments and companies. Revenue grows 82% in year one, 60% in year two and 50% in year three, reaches $122B in year ten, 20 times the last twelve months, and the operating margin rises to 58%, above any GAAP margin the company has reported E (estimate). Free cash flow in year ten is $52.4B.⁴⁵ Weighted 25%, 50% and 25%, the three cases give $107 a share E (estimate), against a price of $184.11.

Where the value comes from

Of my $83 base value, $12 comes from the first five years of cash flow, $15 from years six to ten and $13 from years eleven to fifteen. The remaining $39, 47% of the total, arrives after year fifteen, and $4 is cash E (estimate).⁴⁵ A growth company’s value sits late.

How much of the price is the business as it stands? Take the second-quarter run rate: $7.7B of revenue at a 47% operating margin, taxed at 23%, growing 2.5% a year, plus the cash. That is worth about $19 a share, 10% of the price E (estimate).⁵⁴ The other $165 is growth that has not happened yet.

What closes the gap

One change closes the gap. The bull case’s revenue path, on my margins, takes the base value from $83 to $184. Its margins alone add $12. A 9% discount rate adds $17. Each is measured alone and they do not add: the bull revenue path and the bull margins together give $212.⁵⁶

Two points less growth in every year from year two would cost $13 a share, and two points more would add $16 E (estimate). Five points less operating margin in every year costs $8, and holding the margin at Microsoft’s 46.8% costs $6 E (estimate). If the second half of 2026 comes in at the guide midpoint and the first half of 2027 grows only 10%, the base value is $77; if the first half of 2027 grows 34%, it is $89 E (estimate). The tax shield matters less: with no shield at all the base value is $81 E (estimate).⁵

The margin of safety

• Price$184.11Close, Q4 2026 [56]
• Base value$83My estimate E (estimate)
• Margin of safety below$5830% under my base value C (calculated)

My rule is 25% below the base value for a business with net cash and no customer above a tenth of revenue, 30% when one cycle or two customers drive results, and 35% when fixed charges such as leases sit ahead of the shares. By its terms Palantir fits the 25% tier, which gives $62. I use 30% because the U.S. government was 42% of revenue in the first half of 2026 and can end contracts for convenience, and because 47% of my base value arrives after year fifteen C (calculated).³⁴⁵²⁷ The rule on the NVIDIA page, a price at which the bear value is three quarters of the price, would give $35. At 30% the price is $58, 28 times my year-one earnings of $2.08 at a 2% cash tax rate, 35 times the $1.64 they are at 23%, and 13.5 times my year-one revenue E (estimate). The price sits 217% above the margin-of-safety price.

At $58 my base case returns 12.4% a year and my bear case 5.9%, held for their cash flows; the bear case is 0.9 points above the 5.00% Treasury yield. The bear value is $26, 55% below the price, so the discount protects against a base case that is 30% too high, not against the bear case. The lowest close of the past year, $107.27 on 25 June 2026, was 29% above my base value C (calculated).¹²³³

Where this could be in 5 and 10 years

The base works to year 5 if revenue reaches $30.6B by June 2031, on growth of 45% and 34% in years two and three, with an operating margin near 52%. It works to year 10 if revenue reaches $52.8B at a 50% margin, which leaves 37% of revenue as free cash flow after stock pay. Capital spending is 1.0% of revenue and working capital 10% of each year's revenue increase. The base value rises from $83 to $108 and $136 because the cash flows come nearer each year.

The bear and the bull differ on growth and margin. The bear grows 25% and 15% in years two and three, its margin slips to 32%, and year-10 revenue is $21.4B. The bull grows 60% and 50%, its margin reaches 58%, and year-10 revenue is $121.9B.

The most likely way the base is wrong is growth in years two and three. If they come in at the bear's 25% and 15% and the rest stays as in the base, the value is $63 today, $80 at year 5 and $100 at year 10 E (estimate).

These are estimates of value, not forecasts of the share price. The market price can sit above or below value for years. The base return of 0.0% a year is lower than the 6.6% in the table above because it ends at my year-10 value, which assumes a 10% return from there C (calculated).

The same dollar has other uses. My base return here is 0.0% a year, against 5.00% on the ten-year Treasury.³³ The closest comparable on this site is ServiceNow, enterprise software growing 22%, where my base return is 11.4% a year.⁴¹ Palantir’s price is 122% above my base value and ServiceNow’s is 11% below.

How good the business is and how good the price is are separate questions. The first I can see in the margin and the growth. The second depends on what has not happened yet: about $19 of the $184.11 is the business as it stands E (estimate).

The price needs the bull revenue path, which I weight at 25%, and a margin that stays above Microsoft’s 46.8%, which I put at two chances in three. Multiplied, that is about one in six E (estimate).

VIIThe case for the price

What most people miss

The $4.9B backlog at the top of this page is mostly commercial work. It skips contracts of twelve months or less and anything a customer can cancel for convenience. The CFO said on the August call that both are common in government, 52% of first-half revenue.³⁹⁴ He put a wider count, total remaining deal value, at $13.1B. The 10-Q warns that many of those contracts can be ended early.³⁹⁴

The order book is also well ahead of sales. Palantir signed $2.13B of U.S. commercial contracts in the second quarter, 2.8 times that business’s revenue.⁵ Those contracts run for years, so the ratio flatters the near term.³⁹

The bull case

Palantir becomes the system large organisations run on, in a market far bigger than the one I measure it against. Alex Karp said on the 3 August call that he is driving the business to grow at or above the U.S. commercial rate, then 149%, for eighteen months, “a very high goal.”³⁹ The 2026 guide is 82%.⁵ TipRanks reports that JPMorgan’s Brian Essex, at $262 for December 2027, points to customers wanting control of their AI and data.⁶⁴ Rosenblatt’s John McPeake, at $260 since 26 October, calls Palantir a de facto enterprise operating system, in a market likely “more than an order of magnitude larger” than earlier estimates.⁶³ Both targets sit above my $235 bull value.

In April 2025, Cathie Wood wrote on X that Karp believes Palantir will become the largest pure-play enterprise AI software company: “I believe him.” That is Benzinga’s report; I haven’t seen the post.⁶⁶ In March 2026, ARK’s Brett Winton forecast enterprise software revenue rising from $1.4T to more than $7T in 2030, and named Palantir among the platforms companies may lean on.⁵⁹ ARK’s 13F for 30 June listed 3.21M shares, 2.4% of its reported value,⁶¹ and its ETFs trimmed about $21M in August.⁶² Chamath Palihapitiya’s view is secondhand. The Motley Fool’s Adam Spatacco wrote in February that Palihapitiya argued on All-In that Palantir carries no premium, because many software names it is compared with aren’t true competitors. I found no transcript.⁶⁵

Against reality

Here is the path. The third quarter comes in near $2.35B and 2026 ends near $8.52B. The first half of 2027 grows 75%. Growth then runs 60% and 50% and stays on my bull line to $122B by June 2036. The operating margin reaches 49% to 52%. If they do all that, the price is possible E (estimate).

The near end is close. Revenue grew 93% last quarter, and the price needs 34.8% a year for ten years.⁵ Palantir has beaten the top of its range by 5.7% to 8.7% in each of the last four quarters, and $2.35B needs the top of that.^272625245

The far end is the gap. $122B is 20 times the last twelve months, and my base case has $53B E (estimate). A fast start doesn’t close it: double revenue to $12.3B in year one, fade at my base rates, and the value is $97 E (estimate). The bull path’s $44B around 2030 would be about 0.6% of $7T, against 0.44% of $1.4T today.⁴⁵⁹ If software grows at the 14% ARK’s Frank Downing gave for 2025, it would be 1.6% E (estimate).⁶⁰

What they have to hit

Even if U.S. commercial and government each hit the lines below, the first half of 2027 ends about $0.57B short of $6.24B. The rest has to come from faster growth or from abroad E (estimate).⁴⁵

Where I come down

The backlog gap and the U.S. commercial bookings moved me on year one. My base case needs about 7% growth a quarter in the first half of 2027, against 16% to 19% lately. That is the piece most likely to prove too low E (estimate).²⁷²⁶⁵ ARK’s $7T did not move me, because a market forecast says little about the share Palantir wins or the margin it keeps. A third quarter near $2.35B, U.S. commercial growth above 100% and a margin above 50% with stock pay under 11% would raise my base value.

VIIIRisks, and what would change my view

1. Growth fades faster than my base case. Valuation is the main risk. Two points less growth a year costs $13 a share, and the bear case is worth $26 E (estimate). The company has guided growth only for 2026.⁵
2. Government budgets and terminations. Government was 52% of revenue in the first half of 2026. Many contracts can be ended for convenience, federal options cannot be exercised more than a year ahead, and the $12.3B of IDIQ awards carries no guaranteed funding.⁴³
3. Model makers and clouds selling to the same buyers. Microsoft reports 100,000 Foundry customers and nearly 40 million agents registered in Agent 365, and the 10-K lists customers’ own builds as a rival.⁵¹³
4. Price competition on multi-award contracts. The 10-K says IDIQ contracts can create pricing pressure. Palantir connects to third-party models and has committed $5.6B to a cloud host.³⁴
5. Stock pay and dilution. There are 149.7M options at $10.06, 33.3M unvested restricted shares and 13.7M appreciation rights outstanding. I count stock pay as a cost, so a rise in dilution beyond it would lower my value.⁴
6. Founder control. The founders hold about 9% of the shares and up to 49.999999% of the vote. Karp’s plan allows sales of up to 7.08M shares by 12 December 2026, and 90,000 had been sold under it by 20 August.⁶²⁸³²
7. Key person and succession. The 10-K says the business is highly dependent on its management, particularly Karp, and that his experience would be hard to replace. He has been chief executive since 2005. The proxy lists succession planning among the board’s duties, and the filings I reviewed name no successor.³⁶
8. Politics and reputation. The 10-Q describes activist protest, customer scrutiny and possible loss of business from work seen as controversial.⁴
9. International weakness. International commercial revenue grew 26% in the second quarter against 149% for U.S. commercial, and 80% of first-half revenue came from the United States.³⁹⁵⁴
10. Tax and investments. The tax shield ends in year four of my base case, and a federal cash tax in 2027 would cost about $2 a share E (estimate). $351M of holdings are marked through other income, which was $160M in the first half of 2026.⁴
11. Concentration. The top three customers were 16% of revenue and one was 27% of receivables.⁴
12. Gaps in what I can verify. Net dollar retention, the U.S. commercial customer count and the 2026 international figures are from a third-party transcript. There is no revenue line for AIP. I found no sourced revenue or earnings consensus. I summed the insider sales from Form 4s filed to 19 October 2026 myself.³⁹³²

What would raise my base value: Q3 2026 revenue well above the guide, U.S. commercial growth staying above 100% through 2027, an operating margin above 50% with stock pay under 11% of revenue, and international commercial growth above 50%. A Q3 at the top of the recent beat range, with the second half then at +8% on the guide midpoint and the first half of 2027 at +26%, moves the base value to $89 E (estimate); Q3 is due after the close on 2 November, with the call at 5:00 PM ET.³¹⁵⁵ What would lower it: any quarter that does not beat the top of the guide, a Q3 under $2.26B, U.S. commercial growth under 60%, government growth under 30% for two quarters, or federal cash taxes beginning in 2027.⁴⁵