Palantir reported Q2 ended June 30, 2026. Revenue was $1.935 billion, up 93% year over year; U.S. commercial revenue was $764 million, up 149%; U.S. government revenue was $809 million, up 90%. U.S. commercial TCV was $2.132 billion. These are reported operating facts.
The call's added value is a control thesis: who owns data, business logic, model choice, model-switching rights and post-training weights. This review separates the reported results, management's view and the infrastructure read-through.
Growth is in revenue, contracts and cash flow
| Metric | Q2 2026 | Basis and meaning |
|---|---|---|
| Revenue | $1.935B / +93%year over year; +19% sequentially | Company revenue, not data-center capacity or compute procurement. |
| U.S. commercial | $764M / +149%year over year; +28% sequentially | TCV was $2.132B; TCV includes customer options and is not recognized revenue. |
| Cash and guide | $1.220B adjusted FCF63% adjusted FCF margin | FY revenue guide rose to $8.150–$8.158B; this is company guidance. |
U.S. commercial RDV was $6.238 billion, total RDV $13.1 billion and RPO $4.9 billion. Their cancellation and recognition conditions differ, so they should not be collapsed into revenue or immediately deliverable orders.
Sovereign AI is presented as a control plane, not a single model
Management argues that token consumption may not map to enterprise value and that external-model interaction can compromise control of proprietary knowledge. This is Palantir's view, not a verified conclusion about every model vendor or deployment.
CTO Shyam Sankar described data integration, Ontology and action layers, security and audit, agent orchestration, evaluation and observability, plus supervised fine-tuning and reinforcement learning inside a customer security boundary. The stated promise is workflow-level choice among models, cost, latency and performance.
“turn tokens into actual economic value”
Palantir's claim is that enterprise AI should be measured by operational outcomes, not token consumption.
“own the weights, you own the alpha”
Its thesis is that customers should retain control of weights and the operational advantage embedded in them.
Model-performance evidence
Management said that Nemotron Ultra beat frontier models on some production tasks; no full task set, methodology or independent replication was disclosed.
Preserve source access without republishing a licensed text
call_status: licensed-transcript. IR provides the official earnings materials and YouTube replay, but no official full text transcript was located. The accessible transcript is licensed from Quartr and displayed by StockAnalysis. This column uses attributed summaries and short quotations, not a reproduction or full translation.
The August 3, 5:00 PM ET call included Alex Karp, Shyam Sankar, David Glazer and Ryan Taylor. Q&A added two management claims: its product can switch models in U.S. government settings, and discussions of open versus closed weights, contract terms and customer-specific benchmarks are increasing. Neither statement came with disclosed deployment scale or customer names.
Official webcast replayQuartr-licensed transcript via StockAnalysis
The infrastructure effect depends on production workloads
- 01Enterprise tasks inside a security boundary
Keeping data, rules and post-training within a controlled environment can turn occasional API use into persistent production workflows.
- 02A multi-model control plane
Model substitution and customer-specific evaluation increase requirements for orchestration, observability, network isolation and data governance, not merely token volume.
- 03Compute, network and cooling still need proof
GPU supply, hosted-cloud cost, data-hall capacity or actual inference volume must be disclosed before a server, network, power or liquid-cooling read-through is confirmed.
The supported conclusion today: commercial contracts and the sovereign-AI proposition increase the possibility of production enterprise deployments, but they do not quantify IDC capacity or equipment orders.
Four falsifiable checks for next quarter
TCV conversion
Whether U.S. commercial TCV, RDV and RPO keep growing and convert into reported revenue.
Controlled-production evidence
Whether Palantir or customers disclose model switching, private inference, post-training or secure-boundary deployments at scale.
Physical constraints
Whether GPUs, hosted-cloud cost, networking, power or customer capacity become delivery constraints or quantified investments.
Margin and cash
Whether high growth holds gross margin and cash generation as cloud hosting and technical investment rise.
IDC ATLAS VIEWSovereign AI becomes an infrastructure theme only when proprietary workloads persist inside controlled production environments. Contract conversion, deployment scale and physical-resource disclosure are the next necessary evidence.