2Q26: The Revenue Base Outruns the Re-rating

Palantir Technologies (PLTR US)

Memo Update No. 2 · 2Q26 Results Review · Results released 3 August 2026, valued at the 4 August close

---

I. Thesis Refresher

The central question under evaluation is whether Palantir Technologies ($PLTR)'s operational orientation, distinct in architecture from the analytical workloads that define the hyperscaler stack, produces commercial economics that compound durably rather than cyclically. The primer concluded that the conditions for that compounding were present, anchored by the bootcamp-led acquisition motion, use-case proliferation within existing accounts, IDIQ-led forward visibility in government, and a financial profile that distinguishes Palantir from any reasonable peer comparison. Memo Update No. 1 found the four quantitative conditions cleared with margin and left capital allocation and competitive positioning unresolved. The investment horizon remains 36 to 48 months from April 2026.

This memo covers the 2Q26 reporting cycle, drawing on the earnings release of 3 August 2026, the management discussion in the Form 10-Q, and the earnings call transcript. It is the second quarterly observation against the watch condition framework. The commercial engine strengthened against every condition that could be read, and the period also thinned the disclosure behind one load-bearing condition, which is why the assessment that follows separates a condition's status from whether it could be observed at all.

---

II. 2Q26 in Brief

Metric	2Q25	2Q26	Change
Revenue	$1,004M	$1,935M	+93%
US Commercial	$307M	$764M	+149%
US Government	$426M	$809M	+90%
International Commercial	$144M	$182M	+26%
International Government	$127M	$181M	+42%
GAAP operating income	$269M	$912M	+239%
GAAP operating margin	26.8%	47.1%	+2,030 bps
Adjusted operating margin	46%	62%	+1,600 bps
GAAP net income	$327M	$1,062M	+225%
Adjusted free cash flow	$569M	$1,220M	+115%
GAAP diluted EPS	$0.13	$0.41	+215%
Net dollar retention	--	157%	+700 bps vs 1Q26
Rule of 40	94%	155%	+61 pts

Source: Palantir 2Q26 earnings release, Marvin Labs

GAAP net income is the figure attributable to common stockholders.

Net dollar retention was not disclosed for 2Q25, so the change is stated against 1Q26.

Export

Management raised FY2026 revenue guidance to a midpoint of $8,154M, an 82% growth rate and an eleven-point increase over the guidance given a quarter earlier, which Glazer characterised as the largest full-year raise in company history. US Commercial guidance rose to at least $3,424M, a growth rate of at least 134%.

Three things sit underneath the headline. Cost of revenue grew 37% sequentially against revenue growth of 19%, because Palantir took on cloud hosting for a government customer, and Glazer framed that arrangement as ongoing rather than as a one-quarter effect. Stock-based compensation reached 13.7% of revenue, reversing a compression that had run for five consecutive quarters and that Memo Update No. 1 recorded at its low. An unrealised gain on the company's SpaceX holding contributed $0.03 to GAAP diluted EPS and $0.02 to adjusted diluted EPS, which places a mark-to-market movement on a private holding inside the adjusted figure as well as the GAAP one.

None of the three changes the direction of the quarter. Each changes what the operating margin expansion is made of.

---

III. Watch Condition Assessment

Status is the thesis-impact reading the framework has always used. Evidence records whether the condition could be observed at all, which this quarter forced apart.

C6 is the only downgrade driven by observed performance. C3's amber marks a load-bearing condition that can no longer be verified, and its arrows are deliberately split: the operating trend holds flat while the evidence deteriorates. US Government revenue growth accelerated, backlog growth decelerated from extraordinary rates to very high ones, and nothing in the disclosed indicators declined. An unverifiable condition is a thesis problem regardless, which is why the status moves, but a later reader should not infer an operating decline the quarter did not contain. The column disciplines C8 in the other direction, where green rests on the absence of a firing event in materials that did not include Form 4 filings. Section VI carries C4.

Cluster 1: Commercial momentum

Net dollar retention reached 157%, twenty-seven points above the condition's threshold and 700 bps above the prior quarter. Karp attributed the movement to existing customers migrating across the product stack, describing accounts that ran Foundry alone moving onto the Ontology and the sovereign AI stack. That is the primer's use-case proliferation mechanism, named by management as the driver.

US Commercial revenue grew 149% against a 60% threshold, with US Commercial TCV bookings of $2,132M setting a record at 153% growth. Underneath, customer count grew 35%, decelerating from 42% in 1Q26, while revenue per US Commercial customer roughly doubled to $1.17M from $0.63M a year earlier. Deals of at least $10M rose to 73 from 47.

The evidence indicates growth is concentrating in existing accounts, and that reading is an inference. Palantir publishes no cohort bridge separating beginning-period revenue, newly acquired revenue, expansion and churn, and new customers could also be entering at larger initial contract values. Retention, decelerating customer count and higher revenue per customer support the inference without establishing it. On that reading the mechanism is working as specified, and durability rests increasingly on how deep existing accounts can go.

Cluster 2: Government conversion

Neither the earnings release, the Form 10-Q management discussion, nor the call quantified the IDIQ ceiling. The primer's condition asks whether the $12.3B ceiling produces task order activity at a rate consistent with growth in guaranteed deal value, and the first term of that comparison is unavailable. No 2Q26 conversion rate can be computed, and the $12.3B carried in the workbook is a 1Q26 figure held as context.

The layers that were disclosed are strong in level and decelerating in rate. Total remaining deal value reached $13.1B at 83% growth, against 98% in 1Q26. Remaining performance obligations reached $4.9B at 103% growth, against 134% in 1Q26. US Government revenue grew 90%, accelerating from 84%. Revenue growth accelerated while backlog growth decelerated, and trailing-twelve-month book-to-bill fell to 2.09x from 2.25x.

A book-to-bill above 2x means the backlog is still building at twice the rate revenue consumes it, so the gap is narrowing without yet signalling a problem. What matters for this condition is that Glazer stated RPO is primarily a commercial measure, because it excludes contracts with initial terms under twelve months and obligations beyond termination-for-convenience clauses, both common across most of the government business. By the company's own description, the one forward-visibility metric disclosed this quarter is a poor proxy for the segment the condition governs.

Cluster 3: Financial profile

Stock-based compensation reached $265M, or 13.7% of revenue, against 12.4% in 1Q26. Memo Update No. 1 recorded that 12.4% as the lowest ratio in the company's reporting history. The absolute figure rose 31% sequentially, the largest sequential increase in the series.

Two pieces of evidence cut the other way. Measured year-on-year the ratio still fell, from 15.9% to 13.7%. Dilution discipline improved, with diluted share count growing 0.23% against 0.7% in 1Q26, so the compensation is being funded with materially less shareholder dilution than a year ago. Operating leverage also held everywhere else in the cost structure, with sales and marketing compressing to 17.5% of revenue from 19.5% and general and administrative to 10.1% from 11.2%.

What decides the status is forward-looking. Glazer attributed the increase to technical hiring and guided a significant further ramp in 3Q26 on new-hire seasonality. The condition asks for SBC intensity at or below current levels. Management guided a significant further increase in absolute expense, which creates a risk that intensity stays elevated for at least another quarter without establishing that the ratio will rise again, since sufficient revenue growth would lower it even on higher absolute spend. That guidance removes the basis for treating the condition as comfortably affirmed. It does not by itself establish a reversal of the primer's trajectory.

Cluster 4: International reach

International Commercial revenue reached $182M, growing 26% year-on-year against a primer threshold treating sub-10% growth across multiple years as evidence the gap is permanent. On that comparison the condition improved materially from Memo Update No. 1, which recorded the weakness hardening into deliberate strategy.

The sequential figure complicates it. International Commercial grew 2% quarter-on-quarter, against 28% for US Commercial. The year-on-year improvement is measured off a weak 2Q25 base, and one sequential quarter does not establish which reading describes the run-rate. The condition also asks specifically for recovery beyond the United Kingdom, and no jurisdiction detail was disclosed, so the part that would separate a broad recovery from a single-market one cannot be read. The status holds on evidence pointing both ways.

Cluster 5: Capital allocation

Cash, cash equivalents and marketable securities reached $9.4B, from $8.0B at 1Q26, an increase consistent with the quarter's $1.22B of adjusted free cash flow. The company describes $9.2B of that as cash, cash equivalents and short-term US Treasury securities.

Capital allocation was not raised on the call by management or by any analyst. No buyback authorisation, no acquisition, no framework. Six-month financing activity ran to $9.8M of option exercise proceeds against $1.4M of outflows, which is to say the position is managed by accumulation. Karp's 1Q26 characterisation of the stock as "somewhat undervalued", which Memo Update No. 1 flagged as a possible precursor to policy, was neither repeated nor compounded.

The criticism here is the absence of a stated policy and not the absence of spending. At 58.8x trailing revenue, declining to repurchase stock may be entirely rational, and an acquisition made to deploy a balance rather than to acquire a capability would be worse than holding cash. What management has not supplied is the strategic purpose of the balance, a minimum liquidity requirement, acquisition criteria, an intended treatment of dilution, or the conditions under which repurchases would occur. Any one of those would resolve the condition in either direction. The condition is not deteriorating, because nothing has been done that damages the position, and it is not resolving either.

Governance and key person

No leadership change, no amendment to the founder voting structure and no key-person development were disclosed in the release, the management discussion or the call. The materials do not include Form 4 filings, so the insider transaction detail Memo Update No. 1 carried has no counterpart here, and the green rests on the earnings materials alone.

---

IV. What the Price Now Embeds

At the close on 4 August 2026, the first full trading close following the results, Palantir carried a fully diluted equity value of $371.1B and, against $9.4B of net cash and no debt at 30 June, an enterprise value of $361.7B. That is 58.8x trailing twelve-month revenue of $6.16B. The share count convention is the 2,569M diluted weighted-average from the earnings denominator, which is why this reads as fully diluted equity value instead of market capitalisation. On the 2,403M shares outstanding implied by par value, market capitalisation would be $347.1B and the multiple 54.9x.

The workbook solves that enterprise value for the ten-year revenue compound growth rate it requires. Holding every valuation assumption at its 1Q26 setting, the answer is 34.4%, which carries revenue from $6.16B to roughly $118.6B by FY2036. This is a constant-assumption implied CAGR, not a forecast and not a fair value estimate. Its purpose is a controlled comparison against the prior quarter, and it is only interpretable alongside the assumptions that produce it.

Method. The model uses GAAP EBIT as a simplified proxy for pre-tax unlevered free cash flow. Stock-based compensation stays expensed and is not added back. Cash flow in each of ten years is revenue times an EBIT margin interpolating linearly from the 42.8% trailing twelve-month GAAP starting margin to a 45% terminal margin, with no separate tax, working capital or reinvestment charge. Terminal value is a Gordon growth calculation on year-ten cash flow at a 3.0% perpetual rate, discounted at a 10.53% all-equity WACC, with net cash added to the total.

Two omissions pull in opposite directions and should not be conflated. Palantir's capital intensity is negligible, with capex of $15M against $1.94B of quarterly revenue, so omitting reinvestment distorts little. The absence of a tax charge is a different matter, because EBIT is pre-tax and the omission materially overstates conversion against conventional unlevered free cash flow. Applying the 23.0% long-run rate the company itself uses for adjusted EPS raises the implied requirement from 34.4% to 38.3%. The headline figure is therefore a floor: a fully specified model asks more of the business, not less. The 34.4% earns its place as a controlled comparative against the identically constructed 1Q26 figure, and not as the output of a complete DCF.

The comparable figure at 1Q26 was 35.7%. The embedded growth expectation fell 1.3 points over a quarter in which the share price rose 9.1%.

That decomposes cleanly, changing one input at a time from the 1Q26 basis in the order shown.

Step	Implied CAGR	Change
1Q26 basis	35.7%
Share price $132.38 to $144.45	37.0%	+1.31 pts
LTM revenue base $5.23B to $6.16B	34.6%	-2.39 pts
LTM EBIT margin 38.1% to 42.8%	34.5%	-0.12 pts
Net cash $8.03B to $9.41B	34.4%	-0.06 pts
Diluted shares 2,570M to 2,569M	34.4%	-0.01 pts

Source: Palantir filings, Marvin Labs

The walk is order-dependent. Taking price first measures the re-rating before the results that prompted it.

Export

The walk is order-dependent, and taking price first measures the re-rating before the results that prompted it. Read that way one line does the work. The revenue base grew 17.8% in a quarter, which lowers the rate required from every subsequent year, and that alone more than absorbed a 9.1% re-rating. Margin, net cash and share count together moved the answer by less than a fifth of a point, because the terminal margin is fixed at 45% and a higher starting margin only changes the path toward it.

The useful anchor for judging 34.4% is Palantir's own record. Revenue compounded at 34.9% between FY2019 and FY2025, from $743M to $4.48B. The price therefore asks the company to sustain across the next ten years approximately the rate it achieved across the last six, starting from a base roughly eight times larger and against the law of large numbers rather than with it. Management guided FY2026 to 82%, so the near term is guided well above what the price requires, and the burden sits in the back half of the decade.

Backlog provides meaningful near-term visibility and cannot speak directly to the decade. Total remaining deal value of $13.1B is 1.6x guided FY2026 revenue and firm remaining performance obligations of $4.9B are 0.6x, although neither is a clean coverage ratio, because conversion timing differs across contracts and RDV includes customer options rather than committed revenue alone. What neither can do is speak to years three through ten, because a backlog is a stock that converts and replenishes while the valuation is underwriting a decade of flow. The observation worth drawing is about replenishment rather than coverage. Sustaining 34.4% requires the bookings engine to keep forming contracts at rates comparable to this quarter's for many years, and the current backlog is evidence about the engine's present output instead of a down payment on the horizon.

The valuation assumptions were not re-struck this quarter. The WACC and terminal-margin framework carry forward from 1Q26 at a 4.574% risk-free rate, a 4.33% equity risk premium and a Blume-adjusted beta of 1.375, while the current share price is sensitised around the 4 August close.

Input	Low	Base	High
Share price ($125.65 / $144.45 / $161.92)	32.4%	34.4%	36.1%
WACC (9.53% / 10.53% / 11.53%)	31.7%	34.4%	37.0%
Terminal EBIT margin (40% / 45% / 50%)	36.0%	34.4%	33.0%

Source: Marvin Labs

These are one-at-a-time movements and not combinations, so an adverse combination would fall outside the range at either end.

The base column is the 34.4% carried in the text.

Export

Across the one-variable sensitivities shown, the implied CAGR ranges from 31.7% to 37.0%. These are one-at-a-time movements and not combinations, so an adverse combination would fall outside the range at either end. Whether a decade at any of these rates is achievable is not a question this memo resolves, and the reverse-solve is not a valuation opinion. What it establishes is the size of the operating achievement the current price is underwriting, and that the achievement required got slightly smaller during a quarter in which the stock got more expensive.

---

V. Cost of Revenue and the Hosting Shift

The variable that determined this quarter's margin composition sits in no watch condition. Cost of revenue rose 37% sequentially against revenue growth of 19%, and Glazer attributed the resulting gross margin compression to costs associated with taking on cloud hosting for one of the company's government customers.

Metric	1Q25	2Q25	3Q25	4Q25	1Q26	2Q26
Cost of revenue	$173M	$193M	$207M	$216M	$216M	$297M
Cost of revenue as % of revenue	19.6%	19.2%	17.5%	15.4%	13.2%	15.3%
GAAP gross margin	80.4%	80.8%	82.5%	84.6%	86.8%	84.7%

Source: Palantir quarterly filings, Marvin Labs

Export

Four quarters of gross margin expansion reversed in one. The ratio of cost of revenue to revenue returned to roughly its 4Q25 level, and the trajectory that carried gross margin from 80.4% to 86.8% across five quarters gave back a third of its gain.

Not all of that is hosting. Of the $80.9M sequential increase in cost of revenue, $13.0M is stock-based compensation charged to cost of revenue, which rose in line with the company-wide movement described in Cluster 3. The remaining $67.9M is delivery cost, and management identified the hosting arrangement as the driver. On an adjusted basis excluding stock-based compensation, gross margin fell to 86.3% from 87.9%.

What makes this worth a standing section rather than a line item is that management framed it as a durable change in the delivery model. Glazer stated the arrangement would power faster time to value, drive greater efficiency, provide greater cost certainty to the customer, and enable expansion of that customer's future workflows. Those are the terms in which a company describes something it intends to repeat.

Gross margin is not the same as contract economics, and the evidence does not yet separate them. If this is a repeatable contracting model instead of a customer-specific arrangement, Palantir may be exchanging a portion of gross margin for faster deployment, greater account control and higher lifetime contract value, which would be a good trade at a lower reported margin. Nothing disclosed establishes whether the contract price compensates for the hosting expense, whether the effect persists as the contract scales, or whether the company is taking on infrastructure broadly or absorbing a narrower cloud-service component. The reason it still matters is that the primer's financial profile argument rests on a peer comparison in which Palantir's margin structure is the distinguishing feature, so a durable change in that structure touches the comparison even where it improves the underlying economics.

The connection to Section IV is direct and quantifiable. The reverse-solve holds terminal EBIT margin at 45%. If the delivery model resets the achievable ceiling, the valuation arithmetic moves through the margin input rather than the growth input, and the sensitivity in Section IV prices that: a 40% terminal margin raises the required revenue CAGR from 34.4% to 36.0%. A 500 bps reduction in the terminal margin assumption is worth roughly 160 bps of additional annual revenue growth for a decade.

One quarter driven by a single named customer does not establish a trend, and a one-customer accommodation that annualises into nothing cannot be excluded on this evidence. The distinction is observable in the metrics above and in no condition currently in the set.

Proposed standing overlay. The three rows above are carried in every subsequent memo as a reporting layer, holding no status and no thresholds, so that six quarters of evidence rather than one informs the restatement decision.

---

VI. Competitive and Ecosystem Developments

Memo Update No. 1 set the evidence that would move this condition: disclosed operational workload migrations away from Palantir, or named hyperscaler wins in categories where Palantir has been incumbent. None appeared, and the hyperscaler quarterly disclosures that supply the other half of the read fall outside the materials for this period.

What the quarter did show is a shift in the axis on which the company markets its differentiation. Karp opened the release by stating that demand for AI sovereignty has been unleashed and that customers' competitive advantage should never become the training data for future models. Sankar described extending the AIP stack for sovereign AI, defined as the ability to orchestrate and fine-tune models. That is a claim about control over data and models relative to foundation model providers, which is a different axis from the primer's architectural distinction relative to analytical platforms. Either the sovereignty frame is where the durable moat sits and the primer named the axis imprecisely, or it is adaptation to what enterprises are currently anxious about. This period does not separate the two.

One symmetry is worth the ledger. Memo Update No. 1 read hyperscaler vocabulary moving toward operational framing as pressure on Palantir's territory. This quarter Palantir described its own product in terms of orchestrating models, which is the vocabulary the primer assigned to the hyperscaler stack. Convergence runs both ways, and vocabulary is weak evidence of workload displacement in either direction. A Palantir partnership with NVIDIA ($NVDA) covering secure AI deployment was reported on 1 July 2026, post-period and from press instead of filings, and it is logged for the next memo.

---

VII. Thesis Standing

What strengthened. Both load-bearing commercial conditions cleared by multiples of their thresholds, through the mechanism the primer identified. Retention at 157% and revenue per US Commercial customer roughly doubling describe expansion within the installed base, which is the compounding engine the primer said would have to carry the case, and management named that mechanism when explaining the result.

What weakened or became unreadable. C6 is the only downgrade this quarter caused by observed performance, and it is recoverable if the guided expense ramp proves seasonal. C3's problem is different and more awkward. A load-bearing condition can no longer be verified against the metric its threshold names, and the metric that was disclosed is, by management's own account, a poor proxy for the segment in question. The government business shows no sign of deterioration. The ability to confirm that from disclosure has thinned.

What changed in the valuation and thesis risk. The re-rating cut the opposite way to the usual reading. A 9.1% higher share price would ordinarily raise the bar the operating business must clear, and on constant assumptions it fell, because the trailing base grew faster than the price did. The valuation still asks the company to sustain for a decade roughly the rate it managed over six years from a base eight times smaller. Two risks are live and one of them is new. Capital allocation has run two quarters without a stated policy against a balance now at $9.4B. The delivery model change in Section V is the first identified development in this coverage that could move the valuation through the margin input instead of the growth input, and Section IV prices that channel at roughly 160 bps of required annual growth for every 500 bps of terminal margin.

Proposed framework refinements

None is adopted here. All are logged for the FY2026 restatement.

C3 re-anchoring. Rewrite the condition against metrics Palantir discloses each period. Remaining deal value excluding US Commercial, derivable at $6.86B this quarter by subtracting disclosed US Commercial RDV of $6.24B from the $13.1B total, is the closest available proxy for government forward visibility, imperfect because it also contains both international segments. The IDIQ ceiling becomes context instead of a threshold term.

Delivery model addition. The Section V overlay is a candidate condition, to be decided on six quarters of evidence.

An evidence dimension for the whole book. This memo carried a second column because status alone could not distinguish a condition that deteriorated from one that became unobservable. That distinction is not specific to Palantir, and the coverage-wide framework should carry it so that disclosure quality is tracked separately from operating performance.

---

VIII. What to Watch

Management guided 3Q26 revenue of $2.160B to $2.164B and adjusted operating income of $1.292B to $1.296B, so the growth line is specified in advance and the informative content sits below it.

Two conditions need materials from outside the company's own filings. C4 requires the Microsoft ($MSFT), Amazon ($AMZN) and Snowflake ($SNOW) quarterly disclosures, and specifically any named operational workload migration or competitive win in a category Palantir has held. C8 requires Form 4 filings, which were unavailable for this period and left the insider transaction picture unread.

The primer asked whether operational orientation produces economics that compound durably rather than cyclically. Two quarters in, the compounding is visible and accelerating, the disclosure behind one load-bearing condition has thinned, and on unchanged assumptions the price asks marginally less than it asked in May. The next four prints remain the test.

---

Lifecycle log. C3 proposed for refinement, re-anchored to disclosed remaining deal value with the IDIQ ceiling demoted to context. Cost of revenue and delivery model flagged as a candidate addition, carried as a standing overlay in the interim. An evidence-availability dimension proposed for the coverage-wide condition framework. No condition resolved. No condition added or removed from the canonical set, which remains the FY2026 restatement's decision.