PLTR（Minutes）: US Corporate Revenue Seen Holding Its Q2 Torrid Pace Through End-2027

Dolphin Research Trans of PLTR FY26 Q2 Earnings Call

I. Key Takeaways

1) Full-year guidance raised sharply, the largest single upward revision in company history.

FY26 revenue guidance was lifted to $8.15–8.158bn, with a midpoint of $8.154bn (+82% YoY), up 11ppt vs. last quarter's guide.US commercial revenue is now guided to $3.424bn or higher, implying at least +134% YoY.Adj. OP is guided to $4.889–4.897bn; adj. FCF to $4.5–4.7bn; management still expects GAAP OP and GAAP net income to remain positive each quarter this year.

2) Q3 guidance

Revenue is guided to $2.16–2.164bn, with adj. OP of $1.292–1.296bn.

3) Q2 highlights

Totals: revenue of $1.935bn (+93% YoY; +19% QoQ), the fastest YoY growth on record; GAAP OP of $912mn (OPM 47%); GAAP net income of $1.062bn (margin 55%).Adj. OP was $1.194bn (margin 62%); both GAAP and adj. EPS were $0.41.GAAP net income, adj. FCF and adj. OP all topped $1bn for the first time.

By segment: US revenue was $1.573bn (+115% YoY; +23% QoQ), over 81% of total; US commercial $764mn (+149% YoY; +28% QoQ); US gov. $809mn (+90% YoY; +18% QoQ).Intl commercial was $182mn (+26% YoY; +2% QoQ); Intl gov. $181mn (+42% YoY; +5% QoQ).Total commercial revenue was $945mn (+110% YoY), and gov. revenue was $990mn (+79% YoY).

Cash flow and one-offs: CFFO was $1.216bn and adj. FCF $1.22bn (both 63% margin; +115% YoY), with $9.2bn in cash, cash equivalents and US T-bills at quarter-end.SBC was $265mn and stock-based payroll tax was $17mn.Unrealized gains on SpaceX holdings contributed $0.03 to GAAP EPS and $0.02 to adj. EPS. Rule of 40 scored 155%, up 10ppt QoQ, expanding for the 12th consecutive quarter.

4) Record TCV and backlog metrics

US commercial TCV orders were $2.132bn (+153% YoY; +81% QoQ), nearly $800mn above the prior best quarter.LTM US commercial TCV orders totaled $5.964bn (+117% YoY).

Total TCV orders were $3.4bn (+49% YoY), or +129% YoY on a dollar-weighted duration basis.Commercial TCV was $2.337bn (+118% YoY).

NDR was 157%, up 700bps QoQ; remaining deal value (RDV) was $13.1bn (+83% YoY; +11% QoQ); RPO was $4.9bn (+103% YoY; +10% QoQ).RPO is primarily commercial, excludes contracts with initial terms under 12 months, and excludes obligations beyond 'termination for convenience' clauses, both common in gov. work.

The company signed 220 deals above $1mn, including 98 above $5mn and 73 above $10mn, all record highs.US commercial customers rose to 653 (+35% YoY; +6% QoQ); US commercial TCV rose +124% YoY and +27% QoQ.For the top 20 customers, LTM rev. per customer reached $124mn (+67% YoY).

5) GPM methodology change and opex cadence

Adj. GPM (ex-SBC) was 86%, down QoQ due to higher costs from assuming cloud hosting for a gov. client.Management expects this to accelerate value delivery, improve efficiency and cost certainty for the client, and expand the scope of future workflows.

Q2 adj. opex was $741mn (+14% QoQ; +37% YoY), driven by continued AIP investment and technical hiring.Given seasonal onboarding and product/marketing programs, opex is expected to step up meaningfully in Q3.

Strategic commercial contracts (i.e., early-stage 'invest-for-revenue' arrangements) contributed roughly $0.4mn this quarter, or 0.02% of revenue.Each remaining quarter this year is expected to be below $0.5mn.

II. Call Details

2.1 Management Commentary

1) Demand narrative: enterprises are shifting to 'sovereign AI'.

Management sees a break in the LLM market: enterprises not on Palantir are letting token meters spin for low-quality outputs that do not map to business value, while turning their most sensitive secrets into training data for third-party models.This risks commoditizing their own businesses.On Palantir, demand centers on 'AI sovereignty'—enterprises retaining operational control over data, logic, actions and security—and management calls enterprise data the source of their 'alpha'.

Kirkland & Ellis was cited: with Palantir it built a new operating model for legal services, centralizing and compounding senior partners' expertise.Analyses, discussions and drafting that took days now take minutes, which management said would be impossible without Ontology.

2) Notable Q2 wins (all US commercial)

A multinational tech company that started with a single operating subsidiary in Q4 last year expanded to the entire group and signed a near $370mn, three-year contract.A global asset manager that began in Q1 signed a three-year TCV deal for $35mn, spanning asset management, automation and intelligence across the investment lifecycle.

A global software and services company joined an agent camp in May and signed an initial $15mn, five-month contract.A leading nonprofit health system that piloted in late 2025 converted to a three-year, $37mn TCV engagement.On a dollar-weighted duration basis, US commercial TCV reached $2.1bn, with YoY growth of 271%.

3) Product & tech: the AIP stack and post-training

AIP is positioned as the most ergonomic enterprise AI environment, integrating mixed mammal AI teams across heterogeneous, interdependent workflows.It converts tokens into real economic value in complex, high-stakes settings.

The stack includes data integration and transformation; Ontology and actions; security and audit; workflows; an agent SDK; agent orchestration with telemetry and observability; evaluation with customer-specific benchmarks; AIP Evolve; and new post-training capabilities (SFT and RL).This infrastructure captures operational telemetry to build an 'automated model factory' within the customer's security boundary, embedding intelligence into model weights the customer controls.

Management urges a pivot from 'benchmaxing' to 'benchmaking': customer-specific benchmarks are not just scorecards but steer operational workflows and post-training pipelines.AIP provides a control plane to continuously trade off cost, performance and latency by workflow, and to decide whether to run on proprietary or third-party weights.

Evidence: within 24 hours of integration, the standard, un-post-trained Nemotron Ultra beat frontier models on five production tasks.This suggests a few generic benchmarks can be gamed and that frontier models are not necessarily best in real-world performance.

4) Competition: head-to-head with frontier labs

Management believes the market has created far more intelligence than has been converted into value; stronger models alone will not fix this—AIP deployment speed is the constraint.A recent bake-off at a large Silicon Valley tech company pitted a frontier lab and its deployment team against AIP with Palantir FDEs.

The lab chose a ticket-automation problem and delivered no value.Palantir built agent swarms for each end-customer to proactively recommend marketing, packaging and pricing to drive revenue and usage, converting to a $10mn ACV deal while the lab was eliminated.Same customer, same timeline, same models—the difference was AIP and FDE methodology. Management added that only Palantir has FDEs; others have sales engineers who sell.

5) US gov. and Maven

For the first time, a 'program of record' selected Maven as its operating platform.It will use Palantir's open data standards, Ontology, developer tools, integrations and security primitives to deliver capabilities into the defense community's chosen C2 platforms.

Maven, as the joint force's developer and builder platform, now has 25k+ builders, including service members, civilians, contractors and industry.Palantir's LTM revenue from the Department of War remains under 25bps of the Pentagon's budget.

6) Talent & ecosystem: American Tech Fellowship

Last week in Washington, the company hosted the inaugural American Builder Summit focused on how AI is creating jobs and prosperity.The American Tech Fellowship (ATF) was created on the premise that the most transformative AI apps are often driven by front-line workers without traditional tech backgrounds. ATF graduates now exceed 1,000.

Example: Jonah, a blue-collar worker who joined a submarine parts maker 13 years ago and still works on the line, built an AI app that cut production scheduling from 30–40 days to under one day.Management's takeaway: 'models are commodities, American workers are not'.

7) CEO strategy and growth targets

Karp recapped PLTR's origins in US gov., where it developed the FDE model and early Ontology in the pre-AI, NLP era, rejecting Silicon Valley's parasitic software playbook of locking in clients without true value and instead fully aligning with customer interests.The company is extending the AIP stack into sovereign AI, requiring model orchestration and fine-tuning to deliver a complete sovereign stack.

It has partnered with NVIDIA and is expanding at the application layer.In classified domains, PLTR has entered model fine-tuning; Karp said models fine-tuned by Palantir on NVIDIA's stack outperform frontier models, with customers owning the weights and the alpha.Karp set a target to grow the business over the next 18 months at least at the current US commercial growth rate—an ambitious but achievable goal.

2.2 Q&A

Q: Any unexpected takeaways from the recent sovereign boot camp?

A: It was organized on the fly after this wave broke.Two years ago, it took four to five months to round up people for AIPCon; this time we thought we would just invite a few friends for lunch, but we were swamped with registrations—from invitees and non-invitees across corporate levels, including CEOs and operators. Operator interest is critical in enterprise work.

There was a large educational component.Clients already understand they need to control their alpha and that token maxing comes at their expense—transferring data, prompts, operating methods and expertise to third parties.What they need is education on what to do next: how to contract; how to work with open-weight and closed-weight models; how this runs in Ontology; whether Ontology is the protective layer they were told about and if it can create value; what it looks like in their business; and how to coordinate with the compute stack. We are willing to educate both customers and non-customers.

That is one reason NDR is unusually strong, and it will keep rising.Some dormant customers came, saying they now see why they need us—not just Foundry—and are moving across our stack: Foundry-only users now want Ontology and to enter the sovereign AI stack.Internally, hiring, retention and morale are also benefiting.

Our internal frame is to size the segment of the market that wants to create value and keep it.That segment has grown from a small part of what we did to a significant portion of US GDP.We therefore need partners and are scouting for extremely strong technical partners behind the scenes. Partners are not vassals; we do not need to agree on everything or on every customer, and we may compete at times—like in defense tech. Cooperation does not imply identical views; sometimes we compete, but we are aligned in direction, which lets us scale. I am pushing the company to look beyond year-end to next year's growth because it forces us to find capacity to absorb current demand.

Q: Three years into the AI revolution, enterprises knew proprietary data mattered. Why are they only now realizing they must own what models learn from their data? How did Palantir bet right?

A: There are two parts: effectiveness and scalability.From zero-to-one, focus on the application layer—turn the new thing into economic value. As value realization spreads and time passes, enterprises see some partners building things that compete with them. It takes time for this to permeate market psychology: I know this is valuable in the right hands and platform, but I also need to control the weights because the alpha comes not only from enterprise data but also from the metadata, reasoning traces and usage exhaust—which I currently cannot control. Those may be more valuable than the data itself. Over the last one to two quarters, this has become a market alarm.

As to why we bet right: we are truly aligned with customers, sometimes making decisions against our own economic interest.We support many European institutions where growth is weak, but without our products the consequences for terrorism and migration issues would be ten times worse. It is no longer in our economic interest, but we do it because we are believers.

From inception, this company has valued 'artistic intuition'—you cannot rely solely on scientific modeling; you need aesthetic judgment.We have made many big bets; fundamentally, the people at this table and hundreds at Palantir operate on artistic intuition.We call ourselves a 'colony of artists'. We prize insights seen far earlier than others and build major lines of business around them. That is hard for typical companies built around an execution playbook—today there is no playbook; the old 'build parasitic software and monetize' script no longer works. We joke that if those copying us paid, next year's guide would be in the bag.

We are also outsiders.Outsiders must produce truly good results because clients do not buy from us for our golf swing or steak dinners—they do not even invite us to steak dinners. Being outsiders caused pain in the first 18 years but will bring advantages in the next 18.

Q: Sovereignty has another risk—if you buy orchestration, consulting and harness from a lab, you get tied to its model. We have already seen frontier models pulled by governments or vendors, risking mission-critical outages. With Palantir, can clients switch models at will?

A (Karp): We already do this across the US gov.; our products support model switching.If you are locked into a product, that is monopoly capitalism at work—people want lock-in so they can later raise prices and degrade quality. We oppose that, because we stand with American workers, the American public and its institutions, and with Western institutions more broadly.

Leaders of large US enterprises are sophisticated and acutely aware of these risks.They do not like being maneuvered and treated like fools, so there is accumulated anger.I have spent time explaining that some involved are not cartoon villains; still, the commercial structure looks like 'heads I win, tails I win', and American business does not like that. Contract terms can support clients here.

A (Ryan Taylor): Our sole focus is converting tokens into value.Shyam's example is happening in every client conversation.Clients want to scale partnerships and redefine their industry position. This is not about lock-in to a single model but using the right model for the right task. Our contracts and structures are designed to support and compound the client's alpha.

A (Shyam Sankar): For too long, everyone has been captive to a few benchmarks.Models were designed around them and launched boasting scores that have little to do with your business. The real question is how to build benchmarks that represent your reality and your success criteria, then pick the model accordingly. The natural outcome—putting sovereignty arguments aside—is to ask how to climb this mountain and embed my business back into weights I control, which presupposes open models and sovereignty.

Moreover, you will not wait passively for a model to be pulled before switching.You will use automation to continuously judge whether a more suitable, affordable model appears at the next checkpoint.In the example I gave, I almost felt gaslit—Nemotron Ultra, without any post-training, outperformed frontier models in 24 hours. If you only looked at generic benchmarks, it seemed far behind; that should not have happened. The benchmark is valid for what it measures, but it is not my business or my customer's task. Grounding this in empiricism is how you accelerate converting tokens into real economic value.

Q: Any final words for the many retail investors listening online?

A (Karp): Your support has been critical to getting us here and will be critical to building a much larger company.We will help transform this country and its allies, across commercial and government.Our 'sovereign' framework as an organizing principle is open to everyone who wants a better world today and tomorrow, and we welcome participation in any form.

<End of text>

Risk Disclosure & Statement:Dolphin Research Disclaimer & General Disclosure