Palantir has just printed the best quarter of its public life: revenue up 93% year over year, a 47% GAAP operating margin, its first billion-dollar profit and net dollar retention of 157%. There is nothing wrong with this business. The entire argument is about price. Our band is $100–130 against a $125.65 close — and that close predates the print, which is the first thing to understand before reading any multiple on this page.
The quality question is settled. A fifth consecutive quarter of accelerating growth — 48, 63, 70, 85, 93 — at a $7.7bn run-rate, with a 47.1% GAAP operating margin, cash conversion above net income and a diluted share count up 0.2%. Any bear case built on “it is all stock compensation” is refuted by this quarter’s filings.
A sixth of pre-tax income is not operating income. Interest income of $77.5m plus other income of $91.8m is 15.7% of pre-tax income, and the tax provision was $15.4m on $1,081m — an effective rate of 1.4%. At the company’s own 23% long-term rate the quarter is roughly $0.32, not $0.41.
The reverse test is the strongest evidence here. It requires no view of the company at all. At the last close, a 12% discount rate and a 40% terminal cash margin, the price asks for 37% compound revenue growth for nine consecutive years — $139bn of revenue by 2035, against $8.2bn this year.
Palantir reported second-quarter results after the close on 3 August. The $125.65 used throughout this report is that day’s closing price — the last completed session, and the last price that predates the information the entire report is about. The shares were indicated sharply higher in the pre-market session on 4 August. We price against the last close because that is the single measurement convention this publication uses on every name, and because a pre-market indication on thin volume is not a price anyone was required to accept. Every multiple, gap and percentage below scales directly with it, and will be stale in a way that flatters the valuation the moment the market opens.
With that stated, the argument is simple and it is not about the business. Palantir has moved decisively from a story stock with good margins to one of the highest-quality software profit and loss accounts in the market. Revenue of $1,935.5m grew 93%, the fastest of its public life and the fifth straight quarter of acceleration. GAAP operating income of $912.0m is a 47.1% margin against 26.8% a year ago. GAAP net income crossed a billion dollars for the first time. Adjusted free cash flow of $1,220.4m is a 63% margin on capex of $14.6m. Net dollar retention reached 157%, up 700 basis points in a single quarter. The Rule of 40 reads 155.
None of that is contestable, and we do not contest it. What we do is put a price on it, and the price is where reasonable people separate. At $125.65 the market pays roughly 38x guided 2026 revenue and roughly 70x guided 2026 free cash flow. The question this report exists to answer is what has to be true for that to work, and the honest answer is that it requires the bull case to become the base case.
| Measure | Value | Note |
|---|---|---|
| Last close, 3 Aug 2026 | $125.65 | The reference price used throughout. Q2 was released after this close |
| Market capitalisation | $322.8bn | On 2,568.7m diluted shares; $302.2bn on the basic count |
| Enterprise value | $313.6bn | Less $9.2bn of cash and treasuries; no financial debt |
| 52-week range | $106.37–207.52 | 39.5% below the high |
| 50-day / 200-day average | $130.49 / $152.62 | Below both |
| Short interest | 3.25% | Of shares outstanding |
| Implied / 30-day realised volatility | ~69% / ~60% | Position sizing matters more than direction here |
| Consensus target, 32 analysts | $182.20 | Low $70, high $255 — the widest dispersion of any mega-cap software name |
That dispersion is itself the most honest single statement about this stock. A $70 to $255 spread on a $300bn company is not a failure of analysis. It is what happens when almost all of the value sits in a terminal assumption a decade out.
This is an unlevered discounted cash flow on free cash flow after stock-based compensation and after a normalised 23% tax rate — the honest economic cash flow, not the adjusted free cash flow the company guides to. Those two adjustments matter more here than on almost any name we cover: stock compensation runs at $1.06bn annualised, and the reported tax rate of 1.4% is a net-operating-loss artefact that cannot persist. The base year is anchored to the raised full-year guidance of $8.15bn. Net cash of $9.2bn is added back and the share count grows 0.75% a year.
The sliders open on the generous case, not the base case. That is deliberate: it is the most favourable set of assumptions we consider defensible — a 10% discount rate, 4.5% terminal growth and a 48% terminal cash margin — and it is the one that gets closest to justifying the current price. The three scenario cards below load the strict cases. Move the discount-rate slider first; it does more work than anything else on this page.
Terminal value is 85% of enterprise value. That share is the whole risk in this model, and no amount of care about the first five years reduces it.
Reading the inputs. The widget is a five-year constant-growth form, so the growth rates shown are constant-rate equivalents of an underlying nine-year fading path — they produce the same present value and they are not the near-term growth forecast. The base case’s real path is 52% in FY2027, averaging 32% across FY2028–30 and 15% across FY2031–35, a 23.9% ten-year compound rate to $56bn of revenue; expressed as a flat five-year rate with a Gordon tail that is 41.0%. Terminal EBITDA margin converts to the terminal unlevered cash margin the scenarios quote: 39.5% becomes 30% after tax and capex, 52.5% becomes 40%, 59.0% becomes 45% and the 63.0% opening value becomes 48%. Capex is genuinely negligible — $14.6m in the quarter — and D&A is held at 1% of revenue. One conservatism we have not credited: existing loss carryforwards mean cash taxes will run far below 23% for several years, worth a few dollars a share.
The number above is not a fact about Palantir. It is what our assumptions produce, and with terminal value at roughly 85% of enterprise value the assumption doing the work is one nobody — us included — can defend to a decimal place. Change the inputs until they are yours rather than ours and use the number that falls out. That is the whole reason the model is on the page rather than in a drawer.
Same model, different inputs. Click any card to load it. The spread is enormous — the bull is more than five times the bear — and on a business whose value is almost entirely terminal cash flows a decade out, that is the honest shape of the distribution rather than a failure of analysis.
Weighting those 25 / 50 / 25 at a common 12% discount rate and 3.5% terminal growth gives $68. That weighted case cannot be loaded as a preset — the model supports three — so it is stated here instead. We publish it and do not lead with it, because a probability-weighted average across a distribution this skewed is arithmetic rather than insight.
The discounted cash flow is one input among several, not the answer. On a business where 85% of modelled value sits beyond year five, a forward multiple on something observable carries real weight, and so does the possibility that the model cannot capture what this company becomes. The band below blends all of it.
| Method | Assumption | Per share |
|---|---|---|
| DCF, strict | The base case at a 12% WACC, 3.5% terminal growth, 40% terminal cash margin | $63 |
| DCF, generous | A 10% WACC, 4.5% terminal growth and a 48% terminal cash margin — the opening state of the model above | $113 |
| DCF, faster-growth base | FY2027 at 62% and FY2028 revenue of $19.3bn, at an 11% discount rate | $101 |
| Same, at a 10% rate with 48% terminal margins | $136 | |
| Multiple on FY2028E revenue | Base-case FY2028 revenue of $17.4bn at 15x EV/sales, discounted back at 12% | $83 |
| Same, at 20x | $110 | |
| Same, at 25x | $136 | |
| Multiple on FY2028E free cash flow | About $6.4bn at 35x, discounted back | $72 |
| Same, at 45x | $92 | |
| Probability-weighted DCF | 25 / 50 / 25 at a common discount rate | $68 |
| Street | Consensus Buy across 32 analysts — 19 strong buy, 1 buy, 10 hold, 2 sell | $70–255 |
At $125.65 the shares sit inside that band, in its upper quarter, and about 9% above the point estimate. We have deliberately set the band above the probability-weighted model output of $68. That is a judgement, not arithmetic, and it deserves to be stated plainly rather than buried: a discounted cash flow cannot price the option that Palantir becomes the default operating layer for Western government and large-enterprise AI, and that option is not worth nothing. Investors who use a 9–10% discount rate and believe in a decade of 30% compounding will reach $180 and beyond, and the arithmetic supports them if the assumptions hold. The honest way to state the disagreement is not that the model is wrong, but that the current price requires the bull case to become the base case.
More useful than arguing about what the shares are worth is inverting the model, because the answer depends on no view of the company at all. It simply states what must happen for a buyer at the last close to earn a normal return.
At a 12% discount rate, 3.5% terminal growth and a 40% terminal cash margin, the price requires revenue to compound at 37.0% a year for nine years — to roughly $139bn by 2035, generating about $56bn of annual free cash flow. Under deliberately generous assumptions — a 10% discount rate, 4.5% terminal growth and a 48% terminal cash margin — the required rate falls to 26.4% a year, to about $67bn of revenue. For scale: $67bn would make Palantir larger than SAP is today, and $139bn would make it one of the largest software companies that has ever existed. Neither is impossible. Neither is the balance of probability.
Base-case cash flows throughout, at a 40% terminal cash margin. Read the columns before the rows: this is a business where the risk-free rate is arguably a larger variable than anything management does.
| WACC ↓ / Terminal g → | 2.5% | 3.0% | 3.5% | 4.0% | 4.5% |
|---|---|---|---|---|---|
| 10.0% | $78 | $82 | $86 | $91 | $97 |
| 11.0% | $67 | $70 | $73 | $77 | $81 |
| 12.0% | $59 | $61 | $63 | $66 | $69 |
| 13.0% | $53 | $54 | $56 | $58 | $60 |
| 14.0% | $47 | $49 | $50 | $51 | $53 |
| At a 12% WACC and 3.5% terminal growth | 30% | 35% | 40% | 45% | 50% |
|---|---|---|---|---|---|
| Value per share | $50 | $57 | $63 | $69 | $76 |
Note what the two grids say together. Across the entire plausible range of discount rate, terminal growth and terminal margin, the strict model lands between roughly $47 and $97. Getting from there to the current price is not done by adjusting these dials. It is done by assuming a materially faster and longer growth path than the base case, which is exactly what the bull card and the reverse test describe.
The single most contested input deserves its own arithmetic rather than an assertion. The ten-year Treasury sits near 4.7%, an eighteen-month high, with the Fed having held at 3.50–3.75% in late July against three dissents in favour of a hike and futures pricing roughly a two-thirds chance of a 25 basis point rise in September. Core PCE runs about 3.3%. An equity risk premium of about 4.75% on Palantir’s 1.56 beta gives a cost of equity near 12%. There is no debt, so the weighted average cost of capital equals the cost of equity.
The bull objection is that 12% is punitive for a debt-free business with $9.2bn of net cash, 86% gross margins and the best retention metric in enterprise software, and there is force in that. The answer is that the risk-free rate is not a philosophical preference, and that for an asset whose value is almost entirely terminal cash flows a decade out, the direction of long rates is arguably the largest non-company variable in the thesis. That is why the slider exists and why we say to move it first.
| Metric | PLTR | ServiceNow | CrowdStrike | Snowflake |
|---|---|---|---|---|
| Market capitalisation | ~$323bn | ~$109bn | ~$209bn | ~$107bn |
| Enterprise value | ~$314bn | ~$110bn | ~$194bn | ~$105bn |
| Trailing revenue | $6.16bn | $14.73bn | ~$5.09bn | $5.03bn |
| Latest quarterly revenue growth | +93% | +24% | ~+23% | ~+33% |
| GAAP operating margin | +47% | +12% | −5 to −6% | ~−31% |
| Trailing free cash flow | $3.36bn | $4.58bn | ~$1.5bn | negative |
| EV / trailing revenue | ~51x | ~7.5x | ~38x | ~21x |
| EV / forward revenue | ~38x | ~6.6x | ~30x | ~16x |
| Trailing P/E | ~107x | ~66x | n/m | n/m |
| Forward P/E | ~86x | ~23x | ~75–90x | n/m |
| PEG, consensus | ~1.2 | ~0.95 | n/m | n/m |
| EV / EBITDA | ~94x | ~32x | n/m | n/m |
| Price / free cash flow | ~96x trailing · ~70x FY26E | ~24x | ~129x | n/m |
| Net cash | +$9.2bn | −$1.8bn | +$3.7bn | ~+$0.2bn |
| Rule of 40 | 155 | ~36 | ~48 | ~33 |
Expensive on every absolute measure and most relative ones — but the comparison is less damning than a headline 51x trailing sales suggests. CrowdStrike trades at roughly 38x trailing sales while growing 23% and losing money at the operating line; Palantir trades at roughly 51x while growing 93% with a 47% GAAP operating margin. On a growth-adjusted basis Palantir is arguably the cheaper of the two. Against ServiceNow — the closest at-scale peer, growing 24% with a 12% GAAP margin at 7.5x sales — Palantir is roughly seven times the multiple for roughly four times the growth and four times the margin. Reasonable people land on either side of that trade, and this report does not pretend otherwise.
Peer figures are last reported trailing-twelve-month financials at 3 August 2026 prices and are approximate. Palantir’s own figures throughout are computed from the Q2 2026 release and the 3 August close. What you are holding, in the terms this section describes: implied volatility near 69% against thirty-day realised volatility near 60%, a five-year beta of 1.56, and a twelve-month range from $106.37 to $207.52. Position sizing matters more than direction on a security that moves like this.
| Q2 2026 | Result | Y/Y | Comment |
|---|---|---|---|
| Revenue | $1,935.5m | +93% | Fifth consecutive quarter of accelerating growth; beat consensus by about $123m, or 6.8% |
| GAAP gross margin | 84.7% | +390bps | Adjusted 86%; pressure from absorbing cloud hosting for a government customer |
| GAAP operating income | $912.0m | +239% | 47.1% margin against 26.8% — real operating leverage, not a non-GAAP artefact |
| Adjusted operating income | $1,194.5m | +157% | 62% margin |
| GAAP net income | $1,061.9m | +225% | 55% net margin; first quarter above $1bn |
| Diluted EPS | $0.41 | +215% | Consensus $0.34–0.35. GAAP and adjusted are now the same number |
| Operating cash flow | $1,216.2m | +126% | Cash conversion above net income |
| Adjusted free cash flow | $1,220.4m | +115% | 63% margin; capex of $14.6m is a rounding error |
| Stock-based compensation | $265.2m | +66% | 13.7% of revenue, down from 15.9% |
| Diluted share count | 2,568.7m | +0.2% | Dilution has effectively stopped. Basic +1.5%; shares outstanding +1.79% |
| Net dollar retention | 157% | +~2,900bps | Up 700bps sequentially — the single best number in the release |
| Cash and treasuries | $9.2bn | — | No financial debt; $211m of lease liabilities |
| Rule of 40 | 155 | +61pts | Growth plus adjusted operating margin |
| Segment | Q2'25 | Q1'26 | Q2'26 | Y/Y | Q/Q |
|---|---|---|---|---|---|
| US commercial | $306m | $595m | $764m | +149% | +28% |
| US government | $426m | $687m | $809m | +90% | +18% |
| International commercial | $144m | $178m | $182m | +26% | +2% |
| International government | $127m | $172m | $181m | +43% | +5% |
| Total | $1,004m | $1,633m | $1,935m | +93% | +19% |
Two things deserve emphasis. First, US government re-accelerated to +90%, which undercuts the common framing that this is purely a commercial AI story riding a hype cycle. Second, international commercial grew 2% sequentially and is for practical purposes stalled at about 9% of revenue. If the sovereign-AI thesis were genuinely universal, Europe should be the most eager buyer on earth. It is not, and the reason is structural rather than cyclical: European governments and large enterprises are reluctant to let an American company own their operational data model.
Full-year revenue guidance moved from $7.650–7.662bn to $8.150–8.158bn — roughly $500m added in a single step, against a Street consensus of $7.72bn. Adjusted free cash flow guidance moved to $4.5–4.7bn. Guidance has been beaten in nine consecutive quarters, so the raised numbers are better read as a floor than a forecast. That reliability cuts both ways: a guide that is systematically conservative and systematically beaten carries less information than a guide should.
Forward metrics moved as much as revenue. US commercial total contract value bookings rose 153% to $2.13bn; US commercial remaining deal value rose 124% to $6.24bn; total remaining deal value reached $13.1bn, up 83%; remaining performance obligations reached $4.9bn, up 103%. Record deal counts: 220 deals of at least $1m, 98 of at least $5m and 73 of at least $10m. Expansion rather than customer count is doing the work — 653 US commercial customers, up 35% year over year but only 6% sequentially, against revenue up 149%.
Elevated even by Alex Karp’s standards: “otherworldly”, “the sovereign AI revolution”, and to CNBC, “forget consensus”. He committed to driving the overall business to grow at or above the US commercial rate for the next eighteen months — an unusually specific forward statement worth writing down and checking against. CTO Shyam Sankar was the analytical counterweight, arguing that customer-specific benchmarking beats generic benchmarks and noting that a standard Nemotron Ultra model outperformed frontier models on five production tasks within twenty-four hours. CFO Dave Glazer was the grounded one and volunteered the SpaceX gain disclosure without being asked.
One. A gross-margin headwind from absorbing cloud hosting for a government customer, and a Q3 expense ramp from new-hire seasonality and product initiatives. Two. Three cents of GAAP EPS and two cents of adjusted EPS came from unrealised gains on Palantir’s SpaceX holding rather than from operations. The stake was marked at 30 June; SpaceX has since fallen below its $135 listing price, roughly 49% off its post-listing high, so a reversal in the third quarter is more likely than not. Notably, no analyst on the call landed a hard question on international weakness or on the sustainability of the tax rate.
Palantir’s software sits underneath an organisation’s operations. It takes scattered, incompatible data and builds a single working model of the business — what the company calls an ontology — then lets humans and AI agents act on that model inside permissioned workflows. The distinction that matters commercially is that this is not a dashboard. It is the place where decisions are executed, which is why it is so hard to remove once installed.
| Product | What it is |
|---|---|
| Gotham | The original defence and intelligence platform — sensor fusion, targeting, mission command. Runs inside the Department of Defense, the intelligence community and allied militaries |
| Foundry | The commercial equivalent: supply chains, manufacturing, healthcare operations, financial services |
| Apollo | The unglamorous but strategically decisive deployment layer that pushes software into classified, air-gapped and edge environments. Very few competitors can do this at all |
| AIP | Launched 2023. Connects large language models to the ontology so agents can take real actions. This is the engine behind the last two years of acceleration |
Revenue comes from multi-year enterprise licences plus deployment services, sold either directly through the “bootcamp” motion — a five-day workshop in which a prospect builds a working application on its own data, converting at roughly 75% — or, increasingly, through technical partners. Contracts are typically multi-year, but most contain termination-for-convenience clauses, which is why the backlog headline needs the treatment it gets in Section 07.
| Driver | Evidence, and the caveat |
|---|---|
| US commercial land-and-expand | 653 customers, up 35% year over year, but revenue up 149%. Growth is existing accounts widening, which is why net dollar retention jumped 700bps to 157%. The caveat is that expansion within a fixed base eventually saturates |
| Defence software budgets | Roughly $10bn of contract ceiling across the Army enterprise agreement, Maven, Open DAGIR and intelligence-community programmes. Management notes trailing DoD revenue is still under 25 basis points of the Pentagon budget |
| Sovereign AI | The pitch that a customer’s proprietary advantage should never become training data for future models is landing; management says even Foundry-only customers are migrating onto the sovereign stack |
| Partner-led distribution | Explicitly named as the way to scale past the forward-deployed-engineer bottleneck. Watch this one — it is the difference between 40% and 25% growth in 2028 |
| Operating leverage | 86% adjusted gross margins across roughly 4,400 people. Revenue per employee is $1.39m trailing and $1.75m annualising Q2. Revenue grew 93% while sales and marketing grew 39% and G&A grew 20% |
| Competitor | Where they collide | Our read |
|---|---|---|
| Frontier labsOpenAI, Anthropic, xAI, Google | Enterprise agents; each won $200m Pentagon agentic-AI awards | The genuine long-term threat. Those awards put all four inside the same customer buildings Palantir occupies |
| ServiceNow | Enterprise workflow plus agentic AI; a similar “system of action” pitch | The closest at-scale peer — roughly 24% growth at a fraction of the multiple |
| Snowflake / Databricks | Data platform layer; both pushing agents | Own the warehouse; weaker on operational workflow and on accreditation |
| Anduril | Defence software and hardware; partner and rival on NGC2 | Fast-growing, private, culturally similar. The second tier of defence-technology firms is well funded now |
| Booz Allen, Leidos, CACI, Accenture | Government systems integration | Losing share to Palantir, but they hold the relationships and the headcount |
| C3.ai, BigBear.ai | Nominal “AI software” peers | Not competitive at this point; useful only as sentiment proxies |
One episode from July is worth recording because of what it reveals about the multiple rather than the business. The stock fell roughly 6% in a day when a free open-source “Palantir-style” intelligence tool appeared on GitHub — an event with no revenue impact whatsoever. The fragile thing here is the multiple, not the company.
| Dimension | Score | Assessment |
|---|---|---|
| Switching costs | 9/10 | Once the ontology becomes an organisation’s operating model, replacing it means re-engineering the business. In government, accreditation adds years to any rip-and-replace |
| Regulatory / certification barrier | 9/10 | Two decades of clearances, classified-environment deployment and Apollo’s air-gapped delivery. Very few competitors can even bid |
| Customer loyalty | 9/10 | 157% net dollar retention; 220 deals of $1m or more in a single quarter |
| Brand | 8/10 | Unmatched credibility in Western defence and intelligence, plus a genuine retail following. The same brand is a liability in parts of Europe and with some enterprise buyers |
| Cost advantage | 7/10 | $1.75m annualised revenue per employee across ~4,400 people. The forward-deployed engineer model, once a scaling constraint, is now itself being levered by AI |
| Scale | 7/10 | A $7.7bn run-rate at 47% GAAP operating margins funds research and development smaller rivals cannot match |
| IP / technology | 7/10 | The ontology plus Apollo is genuinely hard to replicate — but the differentiation is architecture, not an unassailable technical secret |
| Distribution | 6/10 | Bootcamps convert at roughly 75%, which is remarkable, yet the partner channel is immature next to Accenture- or Deloitte-scale distribution. Management named this as the constraint to solve |
| Data advantage | 6/10 | Deliberately weak by design: customer data never becomes Palantir’s training data. That is a trust moat, not a data flywheel — an important distinction that is frequently blurred |
| Network effects | 5/10 | Limited in the classic sense; the 25,000-builder Maven ecosystem and shared ontology templates create modest cross-customer effects |
Strongest in switching costs and regulatory barriers, where the government moat is close to unbreachable over a five-year horizon. Commercially the moat is real but shallower: it depends on being first into the workflow. Direction of travel is expanding in US government and US commercial, flat to shrinking internationally. The caveat is that the moat is widening fastest in the segment — government — that historically carries the lowest terminal multiple. That tension sits underneath every terminal assumption in Section 02.
We ran a deliberately hostile read of the release, the financial statements and the insider filings. Severity is scored one to ten, where one is immaterial and ten is severe.
| Concern | Sev. | Evidence, and why it matters |
|---|---|---|
| Insider selling is relentless and one-way | 7/10 | Roughly $130.6m of net open-market sales in the 90 days to mid-June with zero purchases; six executives — Karp, Glazer, Taylor, Sankar, Cohen and Buckley — filed sales on the same day, 20 May; Sankar sold 185,000 shares on 2 July near $130. All under Rule 10b5-1 plans and therefore legally unremarkable. But the unanimity, and the absence of a single insider buy over years, is a signal about where the best-informed people see risk-adjusted value |
| Non-operating income flatters EPS | 6/10 | Interest income of $77.5m plus other income of $91.8m equals $169m, or 15.7% of pre-tax income. The CFO confirmed unrealised SpaceX gains added $0.03 GAAP and $0.02 adjusted EPS. The stake was marked at 30 June and SpaceX has since fallen below its listing price, so a third-quarter reversal is more likely than not |
| A 1.4% effective tax rate | 6/10 | A tax provision of $15.4m on $1,081m of pre-tax income. Net income is not comparable to a normally-taxed peer. At the 23% long-term rate management itself uses for adjusted EPS, the quarter would have been roughly $0.32, not $0.41. Loss carryforwards are finite. Model normalised tax before applying any earnings multiple |
| Governance: three-class structure | 6/10 | Class F shares let Karp, Thiel and Cohen retain up to roughly 49.999% of voting power irrespective of economics, while insiders hold about 7.4% of shares. Outside shareholders have essentially no governance recourse — costless while execution is superb, expensive if it is not |
| Backlog metrics are softer than they look | 5/10 | The release states that TCV, ACV and RDV presume the exercise of all contract options and no termination, while the majority of contracts are subject to termination provisions including for convenience. RPO — the only non-cancellable measure — is $4.9bn, about 60% of one year’s guided revenue. The $13.1bn headline is a pipeline measure, not a backlog, and should not be treated as contracted revenue |
| Customer and programme concentration | 5/10 | US government is 42% of revenue, concentrated in a handful of programmes — the Army enterprise agreement, Maven, NGC2 — and award protests are live: a DIA solicitation was withdrawn this year after one. A single programme decision can move a quarter |
| Legal and reputational exposure | 5/10 | ICE, immigration, IRS and surveillance-adjacent work; recurring public and employee criticism; heavy political identification through Thiel and Karp. A tail risk that is hard to price and impossible to hedge, and it caps the addressable customer set in parts of Europe |
| International commercial is stalled | 4/10 | $182m, up 26% year over year but 2% sequentially, at about 9% of revenue. If the sovereign-AI thesis is genuinely universal, Europe should be a tailwind. It is not showing up yet |
| Promotional management language | 4/10 | “Otherworldly”; “forget consensus”; geopolitical framing on an earnings call. Not an accounting risk, but rhetoric of this intensity should raise, not lower, the evidentiary bar an analyst applies |
| Non-GAAP adjustments are large in absolute terms | 3/10 | Stock compensation of $265m, 13.7% of revenue, plus $17m of employer payroll taxes excluded from adjusted figures. Standard practice, cleanly disclosed and shrinking as a share of revenue — but $1.06bn a year is a real economic cost, and our model charges it |
| Dilution — largely resolved | 2/10 | Diluted weighted-average shares up 0.2% year over year; basic up 1.5%; shares outstanding up 1.79%. Worth stating plainly because the bear case still repeats the old story: the dilution objection has substantially expired |
| Receivables and revenue quality | 2/10 | Days sales outstanding about 70, against 77 in Q1'26 and 68 in Q4'25. Operating cash flow of $1.22bn against $1.06bn of net income. Cash conversion is excellent and there is no evidence of aggressive recognition |
| Debt | 1/10 | No financial debt, $9.2bn net cash, only $211m of lease liabilities. A genuine fortress balance sheet |
There is no accounting-integrity problem here. Cash follows earnings, the balance sheet is pristine and disclosure is above sector average. The flags that matter are narrower and specific: earnings quality, since roughly a sixth of pre-tax income is non-operating and the tax rate is unsustainably low; governance and insider behaviour; and the gap between headline pipeline metrics and contractually enforceable backlog. Model normalised tax and strip out investment marks before applying any multiple to this company.
| Area | Score | Assessment |
|---|---|---|
| CEO track recordAlex Karp | 5/5 | Built an In-Q-Tel-backed startup into an $8bn revenue run-rate at 47% GAAP operating margins, and called the enterprise-AI shift earlier and more decisively than any large-cap peer. Whatever one makes of the rhetoric, the operating record is close to unimpeachable |
| CFO credibilityDave Glazer | 4/5 | Guidance beaten nine consecutive quarters. Volunteered the SpaceX EPS contribution unprompted and flagged the Q3 expense ramp and gross-margin headwind — the behaviour of a CFO managing expectations honestly |
| Guidance accuracy | 4/5 | Systematically conservative and systematically beaten. Reliable — but the pattern itself means the guide carries less information than it should |
| Capital allocation | 4/5 | Almost no M&A, a real virtue in this sector, and $9.2bn of net cash earning interest. The blemish is strategic investments such as SpaceX that inject non-operating volatility into reported EPS |
| Acquisitions | 4/5 | Effectively none of consequence. No goodwill, no integration risk, no serial-acquirer accounting |
| Dilution management | 4/5 | Materially improved: diluted count up 0.2% and stock compensation down to 13.7% of revenue from 15.9%. Credit where a real problem was genuinely fixed |
| Transparency | 3/5 | Disclosure of retention, RPO, deal counts and segment detail is above sector average. Against that, TCV, ACV and RDV are proprietary definitions assuming full option exercise and no termination, and the headline metrics lean heavily non-GAAP |
| Board quality | 3/5 | Competent and founder-aligned. The Class F structure means board independence is structurally limited |
| Insider ownership and alignment | 2/5 | Thiel holds roughly 13.7m shares and Karp about 6.4m Class A plus 52.3m Class B — meaningful skin in the game, but the direction of travel is one-way. Six executives filing sales on the same day, with no open-market purchases in years, is not the behaviour of managers who think the stock is cheap |
| Buybacks and capital returns | 2/5 | No meaningful return of capital. Defensible at this multiple — repurchasing stock at 38x forward sales would be poor allocation — but the cash pile is now large enough that shareholders deserve a stated policy |
| Compensation | 2/5 | Historically among the most generous packages in software; $1.06bn of annualised stock compensation across roughly 4,400 employees |
| Communication style | 2/5 | Charismatic and, for a subset of customers, commercially useful — but it is promotion, and it should raise rather than lower an analyst’s scepticism |
Do they act like long-term owners? In the operating business, unambiguously yes: no empire-building acquisitions, no chasing revenue at the expense of margin, a willingness to walk away from bad contracts, and a genuine decade-long investment in a product architecture that only recently paid off. In their personal capital allocation and in governance, less so: the voting structure insulates them from accountability, and the steady liquidation of stock by every senior executive sits awkwardly beside the public conviction. Both things are true at once, and an investor should hold both.
| Catalyst | Timing | What it tests |
|---|---|---|
| SpaceX mark-to-market reversal | Reported Nov | Mechanical. The 30 June mark added $0.03 to GAAP EPS; SpaceX now trades below its $135 listing price, roughly 49% off its post-listing high. Booked in Q3 |
| September FOMC | Mid-Sept | The single biggest exogenous variable. Futures price roughly a two-thirds chance of a 25bp hike; a ten-year at 4.7% is what makes a 38x-sales multiple fragile. A dovish shift re-rates long-duration growth hard in the other direction |
| Analyst target revisions | This week | The $182 consensus likely moves higher after the raise; some houses will move the other way on valuation |
| US federal FY2027 budget | From 1 Oct | New programme starts against the risk that a shutdown or continuing resolution slips awards between quarters |
| US midterm elections | 3 Nov | Continuity of defence and immigration spending. A change in control could pressure ICE-adjacent contracts and reopen scrutiny of sole-source awards |
| Q3 2026 results | ~2 Nov | Whether Q3 lands above $2.20bn against management’s own $2.16bn guide, and whether US commercial holds above roughly 120% year over year. First sequential deceleration in US commercial is the thing to watch |
| FY2027 guidance | Early Feb 2027 | The single highest-information event on the calendar. A guide implying more than 55–60% growth validates much of the bull case; under 45% breaks the acceleration narrative, and the multiple with it |
| NGC2 next phase; Golden Dome | H2'26–H1'27 | Multi-billion ceiling expansion against the risk that Anduril or a prime takes the lead integrator role |
| European / NATO sovereign-AI frameworks | H1 2027 | Would fix the one clear weakness in the model. Continued EU resistance keeps international flat |
| Lapping the 93% comparison | From Q2 2027 | Arithmetic alone forces sharp headline deceleration. The market’s reaction to the first sub-50% quarter is the key risk event of 2027 |
| Partner-channel scaling | Through 2027 | Management’s stated answer to the delivery bottleneck. Success here separates a 40% 2028 from a 25% one |
| Capital returns | Unscheduled | With $9.2bn net cash and $4.5–4.7bn of annual free cash flow, a first buyback or dividend authorisation becomes plausible |
BULLStart with the number nobody can wave away. Net dollar retention at 157%, up 700 basis points in a single quarter, with customer count up only 6% sequentially. That is not a sales push — that is customers who already bought expanding across departments because the thing works. Fifth consecutive quarter of accelerating growth at a $7.7bn run-rate. Nobody at this scale has done this.
BEARNobody at this scale has done it yet, and that is precisely the point — you are extrapolating an unprecedented series. Ninety-three percent is measured against a quarter when revenue was $1bn. Management’s own guidance implies 73% in Q4. The deceleration has already begun; it is simply happening at a level that still looks spectacular.
BULLDecelerating from 93% to 73% while raising the full year by $500m is not a problem, it is sandbagging — the ninth consecutive quarter of it. And this quarter it was the government business that re-accelerated, to 90%. Trailing DoD revenue is still under 25 basis points of the Pentagon budget. The runway is barely touched.
BEARThen explain international commercial: plus 2% sequentially. If sovereign AI is the tidal wave management describes, Europe should be the most eager buyer on earth. It is not, because European governments do not want an American company owning their operational data model. Nine percent of revenue is effectively ex-growth, and that is exactly the segment that tests whether the thesis generalises beyond the American state.
BULLYou are arguing about the weakest tenth of the business while ignoring that the strongest 81% compounds at 115%. And the profitability is real: 47% GAAP operating margin, $1.06bn of GAAP net income, $1.22bn of free cash flow at a 63% margin, Rule of 40 at 155. The dilution complaint is dead — diluted shares up 0.2% year over year.
BEARI concede the dilution point entirely; bears still making it are not reading the filings. But your earnings figure is dressed up. Fifteen point seven percent of pre-tax income is non-operating — interest income plus an unrealised mark on a SpaceX stake. Three cents of that $0.41 is a mark on a holding that has since fallen below its listing price. And the tax rate is 1.4%. Apply the company’s own 23% long-term rate and the quarter is 32 cents, not 41.
BULLThose are timing items, not quality problems. Cash from operations was $1.216bn against $1.062bn of net income. The cash is real, it is arriving now, and it is unaffected by how you book a tax shield.
BEARThen let us go to price, because that is the only argument that matters. Reverse-engineer it at a 12% discount rate and the last close demands 37% annual revenue growth for nine straight years — $139bn of revenue by 2035. Even granting a 10% discount rate, 4.5% terminal growth and a 48% terminal cash margin, you still need 26% a year to $67bn. That would make Palantir bigger than SAP.
BULLYour discount rate is doing all the work. Twelve percent for a debt-free business with $9.2bn of net cash, 86% gross margins and the best retention metric in enterprise software is punitive. Use nine or ten and the arithmetic looks entirely different — which your own band concedes.
BEARThe ten-year is at 4.7% and the market prices a two-thirds chance the Fed hikes in September. The risk-free rate is not a philosophical preference. And on relative value, ServiceNow grows 24% with a 12% GAAP margin at 7.5 times sales. You are paying seven times the multiple for four times the growth — a good trade only if the growth persists far longer than history says growth of this magnitude persists.
BULLHistory also says the market was wrong about this company at $10, $25, $60 and $100. And every senior insider selling on a pre-scheduled plan tells you nothing about 2030.
BEARNo — but six executives filing on the same day, and not one open-market purchase by anybody in years, tells you what the people with the best information do with theirs.
CHAIRThe bull wins the business argument, decisively. The evidence on operating quality is not seriously contestable: accelerating growth at scale, 157% net dollar retention, 47% GAAP operating margins, cash conversion above 100% of net income, a fortress balance sheet, and a dilution profile that has genuinely been repaired. The bear wins the price argument, also decisively, and the reverse test is the strongest single piece of evidence in the room because it depends on no view of the company at all — it simply states what must happen for a buyer here to earn a normal return. Roughly 26 to 37% compound growth for nine years is the requirement. Palantir might do it. It is not the balance of probability. The two sides are not arguing about facts; they are arguing about which set of facts is load-bearing. Our band spans the disagreement rather than pretending to resolve it, and the last close sits inside it.
The durability of US commercial growth once the base is $3.4bn rather than $1.4bn; whether frontier labs commoditise the application layer or become distribution for it; the terminal multiple, where 6x versus 18x sales a decade out changes fair value by a factor of three; and the path of long rates. Nobody in the debate above knows any of these, and anyone who tells you otherwise is selling something.
One. Q3 US commercial sequential growth — does it hold above roughly 20% quarter on quarter? Two. The size of the SpaceX reversal in Q3 GAAP EPS. Three. International commercial sequential growth, where a second flat quarter would matter. Four. The Q3 gross margin given the government cloud-hosting headwind. Five. The FY2027 revenue guide in February, which carries more information than everything else on the calendar combined.
Imagine a huge organisation — an army, a hospital network, a carmaker — where the information lives in fifty different systems that do not talk to each other. Palantir’s software stitches all of it into one live picture of how the organisation really works, then lets people, and now AI assistants, make decisions and take action on top of that picture. Their word for that unified picture is the “ontology”. Think of it as turning a filing cabinet full of paper into a live control room.
Big multi-year software contracts. Roughly half from the US government — the Pentagon, intelligence agencies, and increasingly civilian departments — and roughly half from large companies. Their clever sales trick is a five-day “bootcamp” where a potential customer builds something real with their own data. About three-quarters of the people who do one end up buying.
Because the numbers just went a bit silly. Revenue grew 93% versus a year ago, and it has sped up five quarters in a row, which almost never happens at this size. Existing customers are spending 57% more than they did a year ago. The company keeps 47 cents of every revenue dollar as operating profit, has $9.2 billion in the bank, and owes nobody anything.
Three things, mainly. Growth slows faster than expected once the year-ago comparisons get hard — and they get hard from next year onward. The big AI labs build competing tools and squeeze Palantir’s territory; all four of them just won Pentagon contracts of their own, so they are inside the same buildings now. Or interest rates keep rising, which hurts expensive growth stocks more than anything else. There is a quieter one too: 42% of revenue depends on government contracts, and governments change their minds and their budgets.
Very. Over a billion dollars of profit in a single quarter. Just note two things. The tax bill is currently tiny — 1.4% — because of old losses being carried forward, and that will normalise. And about three cents of the 41-cent quarterly profit came from an investment in SpaceX rising on paper, not from selling software. SpaceX has since fallen, so expect that three cents to come back out next quarter. Neither of those is dishonest. Both are disclosed. They just mean the headline profit is not quite the profit from the actual business.
This is the honest sticking point, and it is worth being blunt. You are paying about $38 for every $1 of revenue the company expects to make this year. A typical excellent software company trades at $7 to $20. Our own work puts fair value at $100 to $130 a share against a closing price of $125.65 — so the price sits inside our range, near the top of it. That does not mean it will fall, and it does not mean it will rise. It means you are paying up front for years of success that has not happened yet, and you should know that is what you are doing.
One last thought, and it is the most useful habit in this whole report. Notice how much of the argument here is about things nobody can know: what AI spending does in 2030, what the right discount rate is, whether the big labs eat this market. Published price targets on this exact stock range from $70 to $255. That spread is not a failure of analysis. It is an honest reflection of genuine uncertainty, and it is why this report gives you a range and a model you can change rather than a single number you are asked to trust.
| Status | What it covers |
|---|---|
| Verified from Palantir | All Q2 2026 and Q2 2025 income statement, balance sheet, cash flow and non-GAAP figures; segment revenue across Q2'25, Q1'26 and Q2'26; net dollar retention, customer counts, deal counts, TCV, ACV, remaining deal value and remaining performance obligations; stock-based compensation, share counts, interest and other income, and the tax provision; Q3 and full-year 2026 guidance including the revenue and adjusted free cash flow raise; management and analyst commentary from the 3 August call, including the SpaceX EPS contribution, the government cloud-hosting gross-margin headwind and the flagged Q3 expense ramp; Form 4 insider filings from May to July 2026. |
| Our estimate or judgement | All enterprise value and valuation multiples; every input and output of the discounted cash flow, including the constant-rate growth equivalents, the terminal cash margins, the discount rate build-up, both sensitivity grids and both reverse tests; the $100–130 band and the $115 point estimate, including the explicit decision to set the band above the $68 probability-weighted model output; all scores in Sections 06, 07 and 08; and the reading of international commercial weakness as structural rather than cyclical. |
| Still open | Q2 2026 figures are drawn from the earnings release, the condensed consolidated financial statements filed as an exhibit to Form 8-K on 3 August 2026 and the earnings call, and will be reconciled against the Form 10-Q when it files. Peer market capitalisations and multiples are third-party aggregator figures at 3 August 2026 prices and are indicative rather than a precise like-for-like screen. Macro figures — the ten-year yield, Fed pricing and core PCE — are as reported in market commentary dated 3 August 2026. |
Method. All valuation work uses a diluted share count of 2,568.7m, grown at 0.75% a year, and unlevered free cash flow after stock-based compensation and after a normalised 23% tax rate. Company “adjusted” figures are used only where explicitly labelled. The share price used throughout is the $125.65 close of 3 August 2026, the last completed session; Q2 results were released after that close, and every multiple in this report scales directly with the price.