Palantir’s Sovereign AI: Profitable Growth and the Durability of Pricing Power

Palantir’s second-quarter results put concrete figures behind the commercial value of enterprise AI. Revenue reached $1.935 billion in the second quarter of 2026, up 93%, while GAAP operating income was $912 million, a margin of approximately 47%. Growth has reached the income statement.^[1]

Management describes its next opportunity as sovereign AI: enterprises controlling their data, business rules and use of models. The investment questions are what customers will pay for repeatedly, how much profit Palantir can retain, and whether a broader service brings higher hosting and technical staffing costs. The infrastructure implications depend on the contracts and costs connecting software adoption to computing demand.

U.S. commercial growth is rapid; government remains just as important

U.S. commercial revenue was $764 million, up 149%, and U.S. government revenue was $809 million, up 90%. Together, $1.573 billion represented approximately 81.3% of group revenue. Commercial growth was faster, but the government business still contributed slightly more sales. The whole quarter should not be reduced to a single enterprise-software narrative.^[1]

Reconstructing the prior periods from rounded growth rates gives approximately $307 million of prior-year U.S. commercial revenue and an increase of $457 million. The comparable government figures are approximately $426 million and a $383 million increase. Combined incremental revenue was about $840 million. Growth rates establish speed; absolute contributions establish which businesses are moving the group.

The company raised full-year revenue guidance to $8.150 billion–$8.158 billion. That remains management’s expectation, requiring delivery and recognition in the second half. For a business already growing rapidly, the next assessment should ask whether contracts make that path more reliable and whether expense growth supports sufficient delivery capability.

Government and commercial customers have different sales cycles, procurement practices and cancellation terms. Government programs can provide an enduring operational base; commercial customers can offer faster expansion. Both carry budget and purchasing risks. Keeping them distinct helps explain subsequent fluctuations and reduces dependence on a single industry narrative.

Sovereign AI sells control and execution

Management describes a system combining data integration, business objects and operating rules, security, audit, agent orchestration and model evaluation, with post-training inside the customer’s security boundary. Ontology can be understood as organizing business data and the actions permitted on it so that model output can become an operational process. Its value depends on tasks completed successfully.^[4]

Enterprise model selection goes beyond a general benchmark. A customer must decide which data can be accessed, who approves an action, how errors are traced and how the system connects to existing operations. Software that makes those activities faster and more reliable can command payment above the cost of raw model calls. That is the value Palantir seeks to capture.

Management also argues that customers should control proprietary weights and subsequent improvements, and said an open model outperformed frontier models on selected production tasks. Those are company claims, not universal performance conclusions supported by a disclosed test set and independent replication. Model substitution also does not eliminate the cost of migrating business processes.

The attractive shareholder scenario is one in which underlying models become cheaper and more interchangeable while operational integration remains scarce. Palantir could benefit from lower model costs while retaining the customer relationship and application pricing power. The risk is that cloud providers, model developers or internal teams deliver comparable integration, forcing lower prices or more customized work.

High margins require careful treatment of adjustments

Adjusted operating income was approximately $1.194 billion, or 62% of revenue, around 15 percentage points above the GAAP operating margin. The reconciliation attributes the difference principally to approximately $265.2 million of stock-based compensation and $17.3 million of related employer payroll taxes.^[1] Adjusted measures can illuminate operating trends, but common shareholders must still consider the economic cost and dilution associated with equity compensation.

Cash-flow labels matter as well. Operating cash flow of $1,216.167 million less $14.554 million of property and equipment purchases gives approximately $1,201.613 million. Adding back $18.746 million of cash payroll taxes associated with stock compensation produces the company’s adjusted free cash flow of $1,220.359 million.^[1] The $1.220 billion headline should therefore retain the word adjusted when compared with other companies.

This does not change the strength of quarterly cash generation. It matters for long-term valuation. Equity compensation is not a current cash payment, but offsetting dilution through repurchases can require cash. Without an offset, an increase in group cash does not necessarily translate into the same increase in value per share. Profit, cash and diluted shares belong in the same analysis.

Low capital expenditure also does not eliminate computing costs. Purchased cloud services can enter cost of revenue instead of appearing as owned GPUs on the balance sheet. The relevant measure is the total economic cost of delivery. Limited fixed-asset spending does not establish that increasing computing requirements will leave margins unaffected.

Contract value measures opportunity; cancellation terms determine certainty

Quarterly U.S. commercial total contract value, or TCV, was $2.132 billion, and U.S. commercial remaining deal value, or RDV, was $6.238 billion. The call also reported total RDV of $13.1 billion and remaining performance obligations, or RPO, of $4.9 billion.^[1][4] These describe new potential contract value, remaining deal value and unrecognized non-cancelable contracted revenue respectively. They cannot be added into a larger order book.

The company’s definitions assume that customers exercise all options and do not terminate contracts when calculating TCV and RDV, although most contracts have termination provisions. RPO covers non-cancelable unrecognized contracted revenue, but excludes remaining obligations for contracts with original terms of twelve months or less under an accounting practical expedient.^[1] A lower RPO figure does not mean the entire difference from RDV is unreliable; the coverage differs.

Dividing $2.132 billion of TCV by current U.S. commercial quarterly revenue does not establish a number of secured future quarters. Contracts can span years, include options and depend on delivery schedules. More useful questions concern successive conversion into recognized revenue, customer renewal and expansion, and whether longer contract duration has inflated the booking comparison.

Long commitments improve visibility without removing execution risk. Delayed customer projects, changed budgets and complex implementations can alter recognition. Estimates of future earnings should distinguish recognized revenue, optional contract value and non-cancelable obligations rather than give every contract the same conversion weight.

Who pays for computing determines the margin effect

Management said that assuming cloud-hosting responsibilities for a government customer increased second-quarter revenue costs. This is a more concrete financial signal than a general argument that AI requires more GPUs: a broader service can transfer costs previously borne by the customer into the software supplier’s income statement. Management expects faster implementation and additional workflows, but the change does not yet prove that incremental revenue will exceed incremental costs over time.^[4]

A simple sensitivity illustrates the effect. Hold quarterly revenue at approximately $1.935 billion and assume hosting or other delivery costs reduce gross margin by three percentage points, with operating expenses unchanged. Operating income would fall by about $58.05 million, approximately 6.4% of current GAAP operating income of $912 million. This is not a forecast; it isolates the earnings sensitivity to a modest gross-margin change.

Broader service can also support higher spending, longer contracts or better retention, offsetting the cost. Conversely, customers can transfer more delivery responsibility without accepting equivalent price increases. Revenue then grows while the quality of earnings deteriorates. The allocation of model calls, hosting, support and outcome responsibility is central to the contract economics.

Infrastructure analysis requires another step. Software revenue cannot directly identify GPUs, racks or megawatts. Customers may use existing cloud capacity or private equipment, and better model selection may reduce computing per task. Deployment location, procurement, cloud expense or persistent workload evidence is needed before estimating a specific data-center supply-chain effect.

Valuation depends on the duration of growth

The positive case combines revenue growth, operating earnings and cash generation already visible in the accounts. As customers place more business processes on a platform, expansion selling and migration costs can support durability. Greater model choice may increase the value of unified governance and operational integration. Those mechanisms still require support from retention, delivery efficiency and margins.

The strongest counterargument concerns how much application value the company can retain. Customers can develop critical functions internally, competitors can reduce integration costs, and technical staffing and hosting expense can rise with complex deployments. U.S. commercial and government concentration also leaves growth exposed to purchasing and budget changes. High growth does not eliminate those constraints.

Without a matched historical share price, diluted share count and comparable expectations, this article does not assign a valuation multiple or price target. What can be stated is that a growth-based valuation needs an explicit view of duration, normalized margins and cash generation per share. Strong company performance and an attractive stock at a particular price require separate analysis.

The next quarter should clarify commercial contract conversion, the gap between GAAP and adjusted earnings, hosting and technical investment, and whether cash generation per share keeps pace with group growth. If sovereign AI produces durable payments, those results should show it. Its implications for servers and data centers require a separate body of deployment evidence.