Palantir vu par un propriétaire de long terme

Chercher les quintuplements sur dix ans parmi les grandes valeurs de croissance — en pressant la question du potentiel : « Peut-elle devenir bien plus grande ? »

• QHow high is its market ceiling? Is it growing a slice of an existing pie, or creating an entirely new market?8/10

  R

  Conclusion: The ceiling is very high, but Palantir is both "growing a slice of an existing pie" and "using AIP to pry open a new, not-yet-defined pie" — it does both, and the boundary of the latter cannot yet be precisely drawn by anyone today. This is exactly the source of justification for its high valuation, and also its most fragile assumption.

  First, how big is the pie itself. The Baillie Gifford-style line of inquiry is not "is the industry big," but "how large a pool can this company keep capturing share in." The enterprise data + AI application layer where Palantir operates is one of the few truly trillion-dollar arenas that is still accelerating. Gartner forecast in January 2026 that global AI spending will reach about $2.5 trillion in 2026, up 44% year over year, with AI software rising from about $282.9 billion in 2025 to about $453.2 billion in 2026, and AI model spending growing at an even higher rate of about 110%. The $2.52 trillion and +44% cited in the report are consistent with this; Gartner subsequently raised the figure in May to $2.59 trillion, +47%. It must be honestly noted: this is the entire AI pool, and the "serviceable market" Palantir can actually reach is far smaller than this. Applying the whole industry's TAM directly to a single company is the most common overestimation tactic in growth narratives.

  The "growing the existing pie" part is the most solid. In government/defense data integration, Palantir is capturing existing share — it is replacing traditional government IT contractors, custom systems integrators, and scattered internal systems. In 2026Q1, U.S. government revenue grew 84% year over year to $687 million, and in July 2025 the U.S. Army consolidated 75 contracts into a single enterprise service agreement with a $10 billion ceiling and a term of up to ten years — essentially concentrating existing defense software procurement share toward Palantir. This is real, quantifiable displacement of incumbents, but the ceiling is limited by the total size of government budgets and is not infinitely extensible.

  What truly determines the "five-fold imagination" is the "creating a new market" part — AIP connecting large models into real business systems to directly produce business actions. The explosive power of this part is reflected in 2026Q1 U.S. commercial revenue growing 133% year over year to $595 million, and U.S. commercial TCV adding $1.176 billion in the quarter, up 45% year over year. The report judges that "Palantir is expanding from a data platform into an AI operating system that can directly produce business actions" — this "enterprise AI execution layer" did not previously exist as a formed market, and whoever can define it captures the pricing power. But the other side of a "new market" is that the boundary is undefined: today no one can prove that this new pie ultimately goes to Palantir, or whether it gets eaten by the native capabilities of Microsoft, Oracle, and the hyperscale cloud providers.

  The Baillie Gifford-perspective conclusion: the ceiling height is sufficient to support the arithmetic possibility of "five-fold in ten years" — the existing government pie provides a certain foundation, and the AIP new market provides the upside (blue-sky) imagination. But the "reachable height" of this ceiling depends heavily on an unproven assumption: that Palantir can hold the dominant position in value distribution over the long term, in a new market it is creating but that also attracts every giant to rush in. That the market pool is big enough is not the question; the question is how large a slice Palantir can take and for how long.

  11 juin 2026
• QCan its revenue at least double over the next five years? Is the growth driven mainly by volume, price, or new business?8/10

  R

  Conclusion: Doubling revenue within five years is almost a "low bar" — at the current growth rate, Palantir is highly likely to double within two to three years, and tripling to quadrupling in five years is within reasonable extrapolation. Growth is led by "volume" (new customers + existing-customer expansion), "new business" (AIP) is the engine of the ramp, and pure "price" (raising prices alone) contributes the least.

  First, let's be clear about how difficult doubling actually is. The Baillie Gifford ten questions ask "can it at least double in five years," which is a real test for most mature companies, but for the current Palantir it is an almost certain lower bound. 2026Q1 revenue was $1.633 billion, up 85% year over year, and the company raised its full-year 2026 guidance to $7.650–7.662 billion, corresponding to about 71% growth. Relative to FY2025's $4.475 billion, 2026 alone approaches +71% — meaning Palantir clears most of the "double in five years" bar in a single year. The real suspense was never "can it double," but "how steeply this 80%+ curve will revert."

  Growth breakdown — "volume" is the absolute mainstay. Palantir's ramp comes first from customer-count expansion and existing-customer add-ons, not from unit-price increases:

  • Customer count (the first source of volume): 2026Q1 total customers were 1,007, up about 31% year over year. The report discloses 780 commercial customers and 615 U.S. commercial customers in the same period, with new logos continuing to pour in.
  • Existing-customer expansion (the second source of volume, and the most critical): 2026Q1 net dollar retention was 150% — meaning a customer base from a year ago, counting no new customers at all, contributes 1.5x revenue by itself this year. This is expansion of "volume" (more seats, more use cases, more workflows) rather than "price" (charging more for the same thing), and is the highest-quality part of Palantir's growth.

  New business (AIP) is the engine of the ramp, not an independent third leg. An honest distinction is needed: AIP is not a new arena parallel to the core business, but a product upgrade that lets the existing data platform "directly execute business actions," thereby lifting per-customer value and penetration depth. Its power flows directly into the "volume" above — U.S. commercial revenue +133% year over year and U.S. commercial TCV +45% year over year are almost entirely AIP-driven new signings and expansions.

  The "price" contribution is the smallest and hardest to isolate on its own. Palantir does have pricing power (82–87% gross margin, embedded in critical processes), but the revenue growth shows almost no component of "the same contract simply raising prices"; its "price" is expressed more as "selling into higher-value use cases," which still falls under volume/structural upgrade rather than price hikes.

  Honest risk note: doubling in five years is a lower bound, not the issue; the issue is the speed of the slope's reversion. The report's counterpoint puts it well — if growth falls back from 85% to 20%–25% and operating margin reverts to 25%–30%, then even if revenue still doubles, the current P/S (TTM) of about 43.5x pricing may not hold. In other words, whether doubling can be achieved is not the bet; "how gently growth decays while doubling" is. From a Baillie Gifford perspective, this company is textbook-strong on the "volume-driven + new-business engine" criterion and deserves a high mark; the deduction comes only from how much future growth it has pulled forward into today's stock price.

  11 juin 2026
• QFive years from now, what will take over as the next growth engine? Does this "second curve" exist today?6/10

  R

  Conclusion: The second curve already exists today, and it is precisely the main engine of current growth — it is the AIP-driven U.S. commercial / enterprise AI deployment business. But to be honest: what Palantir truly lacks is not a "second curve," but a clearly visible "third curve." The engine taking over today has already ignited; the one that will take over again five years from now is not yet clearly in view.

  Why the second curve is no longer a "future tense" but a "present continuous." When Baillie Gifford asks whether this curve "exists today," the answer for Palantir is almost the reverse: government/Gotham is the first curve, while the commercial/AIP second curve not only exists but is already running faster than the first. Look at the structural evolution:

  • FY2025 government revenue was $2.402 billion and commercial revenue $2.073 billion, with the revenue mix having shifted from the early "government-dominant" toward balance.
  • 2026Q1 U.S. commercial revenue grew 133% year over year to $595 million, far higher than the same period's U.S. government growth of 84%.
  • U.S. commercial TCV added $1.176 billion in the quarter, +45% year over year, building a reservoir for subsequent revenue.

  In other words, Palantir has already completed the "raised by government, taken over by commercial" handoff of curves — something many one-trick companies never accomplish in their lifetimes, and it deserves a high mark under the Baillie Gifford framework.

  What keeps the second curve sustainable. AIP's moat is "connecting large models into the customer's real business systems to directly produce actions," bringing high switching costs — 2026Q1 net dollar retention of 150% shows extremely strong existing-customer add-ons. It is not telling a story; it is an engine already collecting cash.

  But what Baillie Gifford should really probe is "where is the third curve" — and here a deduction is unavoidable. When the AIP dividend slows at the margin three to five years out, what takes over next? The candidates visible in the report are still very vague: first, geographic expansion (growth is currently highly concentrated in the U.S., 2026Q1 U.S. revenue $1.282 billion, +104% year over year, with international commercial relatively lagging), and whether it can be replicated overseas is an open question; second, Bootcamp/platformization turning delivery from "labor-intensive" into "replicable at scale," thereby opening up a longer tail of customers; third, agents/automation pushing the "AI operating system" into a deeper execution layer. These all remain at the level of "direction" rather than "an already-realized curve," and cannot be pointed at on the financials the way AIP can be ("it's already here").

  The biggest dependency and risk of this second curve. AIP's success essentially "stands on someone else's model" — it depends on foundation large-model capabilities, inference cost, and cloud platforms. The report nails it: what it most needs to defend is not "whether AI will come," but "once AI is widely adopted and the underlying capabilities are commoditized, can it still capture enough of the value distribution." If foundation models and cloud-native agent capabilities become highly commoditized in the future, this second curve's pricing power will be weakened, and the third curve that could truly hedge this has not yet formed today.

  The Baillie Gifford-perspective conclusion: the second curve already exists and is strong (high mark on this criterion), but on the question of "who takes over five years from now," what Palantir offers today is still a set of possibilities rather than one definite growth pole. It has a present-day engine; what it lacks is a visible next leg — which is exactly what high valuations most need yet are hardest to prove in advance.

  11 juin 2026
• QWhat is its core competitive advantage? Will this moat widen or narrow over the next three to five years?7/10

  R

  Conclusion: Palantir's core moat is "high switching costs + deep process embedding + government security/compliance credentials," not the classic network effect or data monopoly. Over the next three to five years it is highly likely to be "stable to slightly widening" (AIP deepening embedding), but this is a "real yet capped, and surrounded by strong rivals" moat, which cannot be credited as "risk-free widening."

  Where exactly the moat comes from — strong and weak parts must be distinguished, not blanket-elevated. The report breaks the moat down item by item, and this honesty matters:

  • Switching costs (the strongest item, a real moat): Palantir embeds the data ontology, permission model, business processes, and frontline/factory/hospital/battlefield operations into the customer's organization. 2026Q1 net dollar retention of 150% is hard evidence of embedding depth — the deeper a customer uses it, the harder it is to pull out, and the migration cost is not just technical but also includes permissions, processes, and organizational habits. This is the part of Palantir that most resembles a "moat."
  • Government/compliance barriers (the second strongest, scarce and hard to replicate): security certifications, classified-handling capabilities, track record of past performance, and tender qualifications are themselves barriers. The U.S. Army consolidated 75 contracts into an enterprise agreement with a $10 billion ceiling and a term of up to ten years, and 2026Q1 U.S. government revenue grew 84% year over year, showing this barrier has real monetization power.
  • The "weak items" that must be stated honestly: the report explicitly acknowledges — the network effect is weak to moderate (not "more users, more valuable," but "the deeper the use, the harder to switch"); the data advantage is limited (Palantir does not monopolize the world's raw data, and the "data moat" cannot be exaggerated); the scale advantage exists but is not absolute (far smaller than Microsoft/Oracle/AWS, with the advantage only in specific complex scenarios). Laying out these weak items honestly is what prevents this moat from being mis-rated as top-tier.

  Has the moat been converted into profit — this is a plus. A moat alone does not count; what matters is whether it can price. Palantir's FY2025 gross margin was 82%, and 2026Q1 gross margin about 87%, with 2026Q1 GAAP operating margin of 46%. High gross margin plus rapidly rising operating margin shows that switching costs and compliance credentials have truly converted into pricing power and profit, not merely lived on a slide deck.

  Over the next three to five years: widen, stay stable, or narrow? Judged as "stable to slightly widening," with reasons for both sides:

  • Forces widening it: AIP upgrades Palantir from a "data platform" to an "AI operating system that can directly produce business actions," which further raises per-customer value and embedding depth — the more execution links a customer hands over to it, the harder it is to extract. This is the direction of the moat naturally deepening.
  • Forces suppressing the widening (equally real): precisely because this pie is attractive, Microsoft, Oracle, AWS, Google Cloud, Snowflake, and Datadog are all converging toward "data + AI + application workflows." The report repeatedly warns — Palantir running ahead now does not mean it is differentiated forever; if foundation models, agent frameworks, and cloud-native capabilities become highly commoditized, its pricing power will be eroded. So this is "widening like rowing against the current," requiring continuous R&D investment (FY2025 R&D expense of $558 million) to hold, not collecting rent while lying down.

  The Baillie Gifford-perspective conclusion: this moat is real and already monetized on "switching costs + government compliance" (an above-average mark), and AIP provides a reasonable path toward "slight widening"; but it is not the classic, self-reinforcing, ever-widening top-tier moat of a network effect, and it continually faces the double squeeze of giants' flanking integration and capability commoditization. Within three to five years it is highly likely to hold and slightly deepen, but "it will widen" should be discounted when credited, and absolutely cannot be given a full mark as risk-free widening.

  11 juin 2026
• QIf its core business is disrupted, does it have the gene to reinvent itself? How does it treat mistakes and bad news?6/10

  R

  Conclusion: Palantir has a decent "self-reinvention" track record — it has already successfully completed one major pivot (from a government project-based company into an AIP-driven commercial platform), which is real evidence of a self-reinvention gene; but it has never truly endured a life-or-death moment of "having its core disrupted," so this gene is "half-validated." In treating bad news, its attitude is tough and even combative, which has both merits and hidden risks.

  First, the "empirical evidence" of self-reinvention — it really did pivot, and pivoted beautifully. The "self-reinvention gene" Baillie Gifford asks about cannot rely on declarations; it must rely on history. Palantir's most persuasive point is that it did not get trapped to death in its original identity as a "government data contractor."

  • Pivot at the business-model level: starting from early Gotham (defense/intelligence), the company productized its capabilities into Foundry (enterprise operations), then used AIP to connect large models into real business systems, completing the leap of "raised by government, taken over by commercial." The result is FY2025 government revenue $2.402 billion and commercial revenue $2.073 billion being roughly balanced, with 2026Q1 U.S. commercial revenue up 133% year over year.
  • Pivot at the financial-model level: from long-term losses into "high growth + high gross margin + strong cash flow." Operating margin rose from -27% in 2021 to 32% in FY2025 and 46% in 2026Q1; FY2025 free cash flow was about $2.101 billion.

  A company that can swap out both its "customer mix" and its "profit model" within ten years is clearly not a rigid body — it has the muscle memory for reinvention. This criterion deserves an above-average mark under the Baillie Gifford framework.

  But this gene is "only half-validated" — it has not truly weathered disruption. The honesty of the report lies in pointing out Palantir's most core fragility: AIP's success "stands on someone else's model," depending on foundation large-model capabilities, inference cost, and cloud platforms. In other words, Palantir's reinventions to date have all been "proactively offensive" (I upgrade while the trend is favorable), not "passively survival-driven" (my core got taken out, forcing me to rebuild). The scenario that truly tests the self-reinvention gene — for example, foundation models and cloud-native agent capabilities becoming highly commoditized, turning the "AI execution layer" into a standard offering from the big players — that blade has not yet truly come down. We therefore do not know whether, when Palantir's differentiation is pulled out from under it, it will pivot nimbly as it has before, or lose the fulcrum for reinvention because it relies too heavily on the external model ecosystem. Equating "having pivoted" directly with "being able to rise again even when disrupted" is the easiest over-elevation to commit on this question.

  How it treats mistakes and bad news — tough, a double-edged sword. Palantir's corporate character is distinct: facing controversy, it tends to confront head-on rather than quietly placate. The most typical case is the UK NHS data controversy — Palantir publicly accused the British Medical Association (BMA) of "choosing ideology over patient interest", and refused to back down even under pressure from the UK Parliament's Science, Innovation and Technology Committee recommending the government consider exercising a break clause in 2027.

  • Merit: this character of "not being swayed by public opinion, sticking to its own judgment" is, in long-doubted industries like to-G/defense, instead a survival trait — it will not abandon its strategic direction over a moment's public sentiment.
  • Hidden risk: combativeness also means it may underestimate the depreciation of "public credibility," a soft asset. A company that embeds critical processes into government and healthcare systems, once it continually opposes the public and regulators on "transparency/data governance," may over the long term erode the political foundation for renewals and expansion — exactly the public-opinion/regulatory disruption the report lists as a major risk.

  The Baillie Gifford-perspective conclusion: the self-reinvention gene "has empirical evidence but has not been tested by a crisis" (an above-average mark rather than full); treating bad news with a "candid, tough attitude" is a merit, but "whether it can truly rise again when disrupted" remains an open, untested proposition. It has proven in the past that it will evolve; it has not yet had the chance to prove that it can rise again from a desperate situation — and the high valuation precisely embeds the optimistic assumption that "it definitely can."

  11 juin 2026
• QDoes management (especially the founders) have a long-term vision, with interests deeply aligned with the company? Are they willing to sacrifice current profit for five to ten years out?7/10

  R

  Conclusion: Yes — management is a textbook "long-term vision + founders deeply aligned + willing to sacrifice the present for the long term" team — the trait Baillie Gifford values most, and Palantir should score high on this dimension. But an honest deduction is also required: its alignment method is "super voting rights + extremely heavy equity incentives + symbolic buybacks when overvalued," which is not friendly to minority shareholders, and governance must be viewed at a discount.

  Long-term vision and deep alignment — the positive evidence is solid. Baillie Gifford favors teams where "the founders are in place, interests are bound tight to the company, and they dare to invest for five to ten years out," and Palantir is almost a model:

  • The founding team is in place long-term with a stable direction: Alex Karp, Stephen Cohen, and Peter Thiel are still the core, and the company's strategy has been consistent for over a decade (from defense data to AIP), not the frequent swings of professional managers under short tenures.
  • Control is firmly in the founders' hands: the report discloses that the three founders maintain a total voting power cap of close to 49.999999% through the Founder Voting Agreement / Class F structure, with Karp himself holding large amounts of Class B/Class F and exercisable interests. The "binding" of interest alignment is real binding — the founders' net worth is highly aligned with the company's long-term value.
  • Willing to sacrifice current profit for the long term: the strongest evidence is its history of "not contracting even amid long-term losses" — years of continuously heavy investment in the R&D and sales engines is what bought today's realization. Even after turning profitable, FY2025 still invested $558 million in R&D expense, without cutting investment in the future to dress up short-term profit. This is exactly the "firing for years 3-10" capital-allocation orientation Baillie Gifford wants.

  But the "method" of alignment requires a deduction — this is the core of honesty. The same facts, viewed from another angle, are governance risk, and the report does not dodge it:

  • Super voting rights are unfriendly to minority shareholders: a voting-power cap of close to 50% means minority shareholders can scarcely influence major decisions, and governance is clearly "founder-supremacist." Interest alignment does not equal "governing in the interest of minority shareholders" — these two must be kept separate.
  • Extremely heavy equity incentives are both the cost of "investing for the long term" and a real cost to shareholders: SBC reached as high as $778 million in 2021, and FY2025 still had $684 million, with about $202 million again in 2026Q1 alone. Although SBC is non-cash, it is real dilution to shareholders — diluted weighted shares rose from 1.924 billion in 2021 to 2.565 billion in FY2025, about +33%. The good news is that dilution has clearly slowed in recent years, but "paying salaries by issuing stock" itself shifts part of the cost of "sacrificing the present for the long term" onto existing shareholders.
  • A flaw in capital allocation — symbolic buybacks when overvalued: the report documents it solidly, in January 2026 it repurchased only 8,452 shares at an average price of about $177.45 before immediately terminating the buyback program, while the current stock price is about $130.53. Doing symbolic buybacks at a high-valuation point and never paying dividends shows that within its merit of "daring to invest for the future" there is also the shortcoming of "not being good at rationally returning excess capital to shareholders."

  The Baillie Gifford-perspective conclusion: management's "long-term vision + interest alignment + daring to sacrifice the present" is real and strong (a high mark for the core of this dimension); but the alignment means (super voting rights + heavy SBC) are unfriendly to outside shareholders, and buyback discipline is mediocre, requiring a clear deduction on "governance friendliness to minority shareholders." In other words — this is a team you would be willing to let steer for you for ten years, but you have to accept that "once you board its ship, the steering wheel is never in your hands, and the ticket comes with continuous dilution costs."

  11 juin 2026
• QIf it disappeared tomorrow, how much would customers miss it? Is its growth method sustainable, not reliant on harming society and regulation?6/10

  R

  Conclusion: Two sides to consider — "indispensability" is strong: if Palantir disappeared tomorrow, the government and large-enterprise customers running critical processes on its platform would miss it greatly, and migration would be enormously painful; but the "social/regulatory sustainability of its growth method" is its real soft spot: in sensitive areas like defense, law enforcement, immigration, and healthcare, its expansion inherently comes with public-credibility and regulatory friction, and this side must be honestly deducted. Only when both dimensions pass does it count as a high mark, and Palantir is "high indispensability, questionable social sustainability."

  The first side — indispensability: quite high. The core of this Baillie Gifford question is "would customers miss it," which corroborates the switching costs in the moat. Palantir is not a lightweight tool that can be unplugged once installed, but embeds the data ontology, permission model, business processes, and frontline/factory/hospital/battlefield operations into the customer's organization. Hard evidence:

  • 2026Q1 net dollar retention of 150% — customers not only do not churn but are substantially adding on, a quantitative manifestation of "the more you use it, the more you can't do without it."
  • Deep binding on the government side: the U.S. Army consolidated 75 contracts into an enterprise agreement with a $10 billion ceiling and a term of up to ten years, and 2026Q1 U.S. government revenue grew 84% year over year. Once critical military/intelligence workflows run on Gotham, replacement is not a technical problem but a problem of operational continuity.

  So the answer to "how much it would be missed" is: for deep customers, Palantir disappearing would mean paralysis of core processes and a lengthy rebuild — a high mark for indispensability.

  The second side — social/regulatory sustainability: this is a real soft spot and must be honestly deducted. On the additional "not reliant on harming society and regulation" dimension that Baillie Gifford adds, Palantir does not answer well. Its most profitable, stickiest business (defense, law enforcement, immigration, healthcare data) is precisely the area the public and regulators are most sensitive to, and the controversy is structural, not incidental:

  • The UK NHS Federated Data Platform (FDP) is a living case: the Palantir-led consortium won the contract in 2023 (about £330 million in scale), after which the UK Parliament's Science, Innovation and Technology Committee recommended the government consider exercising a break clause in 2027, and called Palantir "the most concerning technology supplier".
  • The opposition comes from professional bodies and the legislature, not individual emotions: the British Medical Association (BMA) voted against the rollout of the FDP and lobbied at the national level; lawmakers from several European countries jointly urged the UK government to reconsider Palantir's NHS and defense data contracts, with the focus on patient-data access, transparency, and vendor lock-in.

  This shows that Palantir's growth has indeed touched the boundary of "social acceptability" in certain regions — even where it is legally compliant, large public-sector projects may be reassessed or even terminated due to shifts in the political climate. The report is right to list this as a major risk: on the criterion of "not harming society and regulation," its growth method cannot be given a clean bill of health.

  Putting the two sides together — a subtle tension. Palantir's indispensability is partly built precisely on its involvement in those "sensitive but essential" areas; and it is exactly these areas that discount its social/regulatory sustainability. It is not the kind of company whose "growth is built on harming users" (like addictive business models), but it is indeed one whose "growth is built on sensitive data power that triggers social controversy" — these two carry different risk natures, but both constitute disturbances to "clean long-term compounding."

  The Baillie Gifford-perspective conclusion: indispensability is high (a high mark), social/regulatory sustainability is questionable (a clear deduction). That customers can't do without it is a fact, but whether it can grow these sensitive businesses big and long without continually inflaming opposition from the public and regulators is an unsettled question that clearly drags on long-term certainty. Treating it as a "clean, frictionlessly compounding for ten years" target would underestimate this layer of structural political/public-opinion risk.

  11 juin 2026
• QWhat are the unit economics of this business (gross margin, incremental returns)? Do they get better or worse as it scales? Where is the money it earns spent?7/10

  R

  Conclusion: The unit economics are excellent, and clearly improve as it scales — high gross margin, light capex, strong operating leverage, high incremental returns. This is Palantir's most scrutiny-resistant dimension and the one that most deserves a high mark. The only place that needs an honest discount: it is not "zero-marginal-cost pure SaaS" (it carries delivery/on-site deployment costs), and a considerable portion of "the money earned" is distributed to employees as equity incentives, diluting the returns owners actually take home.

  Gross margin — high and rising, getting better with scale. FY2025 gross margin was 82%, and 2026Q1 gross margin about 87%. Gross margin rising with scale shows higher productization and falling unit delivery costs — a sign of "bigger is better" rather than "bigger is more tiring." But it must be honestly noted: the cost line includes third-party cloud hosting, subcontractors, and on-site service staff, so it carries an implementation/delivery-layer cost and is not Coca-Cola-style zero-marginal-cost. This does not affect the "excellent" verdict; it only marks down "top-tier pure software" by one notch.

  Operating leverage — this is the brightest part of the unit economics. The trajectory of operating margin is itself a report card of incremental returns: from -27% in 2021 → 32% in FY2025 → 46% in 2026Q1. The report is right — this was not squeezed out by layoffs or short-term cost suppression, but by "strong revenue growth layered with the release of operating leverage." In other words, for every additional dollar of revenue, a rising proportion falls to operating profit, and the incremental returns are excellent. Layered with 2026Q1 revenue +85% year over year, this is a textbook case of scale amplifying unit economics.

  Capex is extremely light — growth barely consumes money. FY2025 capex was only about $33.88 million, almost negligible relative to $4.475 billion in revenue. This means Palantir's growth is "asset-light" — it does not need to keep pouring heavy capital into expansion, a very valuable trait in unit economics, and the root cause of why its cash flow can scale rapidly. The direction of incremental capital returns is already very good: FY2025 operating profit of $1.414 billion, no interest-bearing debt on the books, extremely light capex (the report does not give a single-point ROIC due to definitional reasons, but the direction is clearly strong).

  Cash flow — the unit economics convert into real cash. FY2025 free cash flow was about $2.101 billion, and 2026Q1 adjusted free cash flow $925 million, a 57% margin. High gross margin + light capex + strong operating leverage, layered together, let revenue convert efficiently into cash — the ultimate validation that the unit economics are excellent.

  Where the money earned is spent — the rational part coexists with flaws:

  • The best spending: continuing to invest in the high-return product and sales engines (FY2025 R&D $558 million), the correct action of reinvesting high incremental returns to compound bigger.
  • The controversial spending (deduction needed): a large amount "distributed to employees" as equity incentives — FY2025 SBC of $684 million. This means a slice of the excellent unit economics does not truly land in existing shareholders' pockets, but is diluted to employees. The report's conservative framing is valuable: after deducting recurring SBC as a real cost, FY2025 "conservative owner earnings" were about $1.417 billion, and TTM about $1.958 billion — still strong, but not as dramatic as the "FCF stock star chart" looks.
  • The poor spending: symbolic buybacks when overvalued, and never paying dividends (see the management dimension), reflecting that the unit economics are excellent but capital-return discipline is mediocre.

  The Baillie Gifford-perspective conclusion: unit economics (gross margin, operating leverage, incremental returns, light capex) is the strongest of all Palantir's dimensions, clearly improving with scale, and deserves a high mark. The only discount is — it is not zero-marginal-cost pure SaaS, and a considerable portion of "the money earned" dilutes owners in the form of SBC. The business's own earning power is first-rate; the part that truly lands in shareholders' pockets must be discounted with a more conservative framing.

  11 juin 2026
• QFor it to rise five-fold in ten years, what conditions need to hold simultaneously? Are these conditions realistic? What does today's stock price imply?3/10

  R

  Conclusion: For Palantir to rise five-fold in ten years, three things need to hold simultaneously — "sustained ultra-high growth + maintaining extremely high margins long term + the market still willing to give a generous valuation ten years out" — and because today's starting valuation is already extremely high, all three must lean toward the optimistic end at once. The realism is "not nonexistent, but the requirements are very demanding." Today's stock price already implies the expectation that "Palantir must nearly do all of the above," leaving buyers a margin of safety close to zero. Under the Baillie Gifford framework, the price has overdrawn the imagination on this item, and a clear deduction is warranted.

  What conditions must hold simultaneously for five-fold in ten years. Break it into testable premises, rather than a single "keep being great":

  • Condition one · growth cannot collapse: revenue must maintain a growth rate far above large software companies for a very long time. The base is not bad — 2026Q1 revenue +85% year over year, full-year 2026 guidance about +71%. But five-fold requires this curve to decay slowly enough, not one year of high growth.
  • Condition two · margins cannot fall back: operating margin must stay high long term. The current 2026Q1 GAAP operating margin of 46% is already top-tier, and the five-fold narrative requires it not to be dragged down by competition and price wars.
  • Condition three · valuation cannot mean-revert: this is the most overlooked yet most lethal one. The market must still be willing to give it a multiple far above ordinary software stocks ten years out. Today's static P/E of about 146–154, P/S (TTM) of about 43.5, P/FCF of about 136 — if the multiple ten years out only reverts to the "mature high-growth software stock" range, then even if profit multiplies several times, the stock price could tread water or even decline.
  • Condition four · dilution and governance do not drag: SBC no longer runs out of control, and the super-voting-rights structure does not trigger major capital misallocation.

  Are these conditions realistic — demanding but not a fantasy. Item by item: growth (condition one) and margins (condition two) have a realistic basis in the short-to-medium term given current fundamentals, and Palantir's execution is indeed strong. The real difficulty is condition three: it requires "excellent operations" and "the market's continued generous pricing" not to slip for as long as ten years simultaneously. Historically, very few companies have, from such a high starting valuation, "digested the valuation through growth" via fundamentals while the stock still rose five-fold — this requires the company to be not just great, but "even greater than today's pricing already requires." The report's judgment is restrained: this is not impossible, but it belongs to the optimistic end of blue-sky imagination, and treating it as the base case is dangerous.

  What does today's stock price imply — this is the core of the question. Using the anchors from the report's three Owner Earnings discount scenarios:

  • Conservative-scenario intrinsic value about $18–30/share, neutral about $35–60/share, optimistic about $75–110/share (the report's self-built model, an assumption rather than fact).
  • While the current stock price is about $130.53 — higher than the report's "optimistic-scenario upper bound of $110."

  In other words, the expectation implied by today's price is even more aggressive than the most optimistic tier in the report's model: it requires a higher starting point, a longer period of ultra-high growth, and a lower discount rate or higher exit multiple, all layered together. From another angle, the current free cash flow yield is under 1%, while the 10-year Treasury is about 4.59% — buying it means forgoing current cash returns and placing the entire bet on "extreme success over the next ten years." The report's ten-year expected annualized return ranges are conservative -6%~-2%, neutral 0%~4%, optimistic 6%~10%, and only the optimistic end approaches the "five-fold" magnitude, while that end requires all conditions to land in the best cells at once.

  The Baillie Gifford-perspective conclusion: five-fold in ten years is not arithmetically impossible (conditions one and two have a realistic basis), but it requires "sustained ultra-high growth + extremely high margins + the market's long-term generous valuation" to hold simultaneously, and because the starting valuation already exceeds the report's optimistic-scenario upper bound, it leaves almost no room for error. This item must be honestly deducted — not on business quality, but on "the price having pulled forward and overdrawn the five-fold story." Baillie Gifford is willing to pay a premium for great growth, but when the price itself is built on "everything must be perfect," the upside is overdrawn and the downside is amplified — which is exactly the fundamental reason the report gives "Watch" rather than "Buy."

  11 juin 2026
• QWhy hasn't the market realized all this yet? Is it that they can't understand it, look down on it, or can't see far enough? What would become the "narrative inflection point"?3/10

  R

  Conclusion: For a stock like Palantir, that classic Baillie Gifford line of inquiry must be asked in reverse — the market is not "yet to realize how good it is," but "has already fully priced in (or perhaps over-priced) its goodness (and more)." So there is almost no cognitive-gap dividend here from "can't understand / look down on it"; it is closer to "seeing too far, discounting ten-years-out success into today." The narrative inflection point is therefore two-directional: a downward inflection (either growth or valuation loosening) is more realistic than an upward one. Under the Baillie Gifford framework, the cognitive gap on this item points in a negative direction, and a deduction is warranted.

  First, correct the premise of the question — this is not a stock that is "undervalued, market doesn't get it." This Baillie Gifford question presupposes that "the market may not understand, look down on, or see far enough," but applied to Palantir it must be honest: it is one of the most watched, most thoroughly researched, and most publicly debated companies in the current U.S. market. The evidence is the valuation itself — static P/E about 146–154, P/S (TTM) about 43.5, P/FCF about 136, current stock price about $130.53, market cap about $310–330 billion. The market is not "looking down on" it; on the contrary, it has given it the expectation-pricing of "one of the best large software platforms." The report's judgment is apt: you cannot call it cheap just because peers are also expensive; the more reasonable understanding is — its goodness has already been seen by the market and generously paid for.

  Examining the three possibilities of "can't understand / look down on / can't see far" item by item, almost none hold:

  • Can't understand? Does not hold. Palantir's realization is public and quantifiable: 2026Q1 revenue +85% year over year, U.S. commercial +133%, net dollar retention 150%. These are not complex clues hidden in footnotes, but headline numbers that the sell-side and retail investors see clearly.
  • Look down on it? Quite the opposite. The market gives a premium rather than a discount — this is a typical characteristic of "over-respecting," the opposite of "value being ignored."
  • Can't see far? If there is a bias, the direction is also "seeing too far, too optimistically": the current stock price is already higher than the upper bound of the most optimistic of the report's three scenarios (about $110/share), meaning the market has heavily pulled forward extreme ten-years-out success. This is not "not seeing far," but "treating the best long-term outcome as the base case."

  So where is the real "cognitive gap" — in the pricing of risk, not in growth. If there is anything on Palantir still systematically underestimated by the market, it is not its growth (that is fully or even over-priced), but its downside risk being underestimated. The counter-logic the report points out most deserves serious attention: if revenue growth falls back from 50%+ to 20%–25%, operating margin reverts to 25%–30%, and the multiple the market is willing to give compresses from over 100x to 30–50x Owner Earnings, then current buyers' ten-year returns could be only low single digits or even negative — and this "does not require the company to decline, only requires it to be excellent but not as excellent as today's stock price demands." In the market's current pricing, the probability weight given to "mean reversion," the most likely scenario, is too low — that is the side truly being overlooked.

  What would become the "narrative inflection point" — two-directional, but downward is more realistic:

  • The downward inflection (more realistic, any of several triggers suffices): first, U.S. commercial growth clearly decelerating for several consecutive quarters, breaking the "infinite AIP penetration" narrative; second, net dollar retention (currently 150%) weakening, shaking the core belief of "embedded so deep it can't be pulled out"; third, large government deals hit by budget/public-opinion shocks (such as the spread of controversies like the UK NHS contract being recommended by Parliament for consideration of termination in 2027); fourth, margins falling back with no productization explanation; fifth, SBC running out of control again, bringing accelerated dilution. Any of these materializing could lead the market to redirect the valuation from "perfectly priced" toward "normal high-growth software," which is exactly what the report calls "the biggest permanent capital-loss scenario is not necessarily the company declining, but a good company + too-high purchase price + mediocre ten-year returns."
  • The upward inflection (exists but harder): it would require AIP to make U.S. commercial even bigger than the already very high expectations the market holds, or a brand-new third growth curve to be validated, or margins to step up another level — the difficulty being that much of this "goodness" has already been digested by the price in advance, and what can truly drive the stock substantially higher again must be a surprise that "exceeds the already very optimistic consensus."

  The Baillie Gifford-perspective conclusion: the answer to this question for Palantir is "the market has not missed its goodness; instead, it may have missed its fragility" — the direction of the cognitive gap is negative, and there is no "buy what others discard" undervaluation dividend. The narrative inflection point exists in both directions, but at such a high starting valuation, the downward inflection has a lower threshold and is more realistic. This is consistent with the report's conclusion of "passes on the business, fails on the price" and "Watch rather than Buy": it is an excellent company, but today's price demands everything be perfect, and perfection is the easiest thing to break — which is exactly where a high-valuation target most deserves an honest deduction, rather than being elevated for the growth narrative.

  11 juin 2026