Palantir does not own any satellites. Their model is a software aggregation layer — they license data feeds from commercial satellite operators, fuse them through their AI platform (AIP / MetaConstellation), and sell the output to governments and defence agencies as a managed service. Their advantage is entirely in the software and AI layer, not the space hardware.
| Product | Function | What it does in the satellite context |
|---|---|---|
| MetaConstellation | Satellite orchestration layer | Tasks multiple satellite operators simultaneously, routes imagery, manages priority queues, delivers real-time alerts |
| AIP (AI Platform) | Core AI engine | Object detection, change detection, target classification, kill chain automation, natural language query on imagery |
| Ontology | Data model / knowledge graph | Links objects across sensor feeds — a tank spotted by SAR matched to same tank in EO imagery from 6 hours earlier |
| Maven Smart System | DoD military interface | Full ISR-to-strike workflow. Ingests satellite + drone + SIGINT. AI generates target packages. Used operationally in Ukraine and Middle East. |
| Foundry | Enterprise data platform | Data pipeline management, integration, analyst workbench, reporting |
| Apollo | Deployment / CI/CD | Manages software deployment across classified and unclassified environments, edge devices, and satellite payloads |
To replicate what Palantir delivers — without depending on Palantir's platform or its pricing — you need to build the same four layers independently: satellite data access, orchestration, AI analytics, and client delivery. You do not need to own satellites to start. The fastest path mirrors Palantir's own model: license the same satellite feeds, build your own AI layer on top.
| Layer | Palantir's Approach | Independent Approach | Recommendation |
|---|---|---|---|
| Satellites | License from 8 operators — no ownership | License same operators OR own constellation Phase 2+ | License first, own later |
| Orchestration | MetaConstellation (proprietary) | Build open-architecture tasking platform | Build — 12–18 months |
| AI Engine | AIP + Ontology (proprietary) | Build on open-source CV + custom fusion engine | Build — core differentiator |
| Client Interface | Maven / Foundry / Apollo | Build secure web dashboard + API + C2 integration | Build — 6–12 months |
| Edge AI (satellite) | Satellogic partnership (onboard inference) | Replicate via Satellogic API OR own onboard AI chips | License Satellogic first |
By going direct to the satellite operators, you bypass the Palantir markup entirely and get the same underlying data:
| Operator | Type | What You Get | Direct Annual Cost | Palantir Equivalent Value |
|---|---|---|---|---|
| Planet Labs | EO optical | Daily global basemap, archive access, tasking API, PlanetScope + SkySat feeds | $500K–$5M/yr | ~$50M of Palantir contract value |
| Maxar | High-res EO | On-demand tasking at 0.3m GSD, archive library, SecureWatch platform API | $1M–$10M/yr | ~$80M of Palantir contract value |
| BlackSky | Rapid-revisit EO | Sub-hourly revisit tasking, real-time alerts, site monitoring, dynamic collect | $2M–$8M/yr | ~$40M of Palantir contract value |
| ICEYE | SAR radar | All-weather tasking, archive SAR, ship detection, COSMO-SkyMed class resolution | $2M–$8M/yr | ~$60M of Palantir contract value |
| Satellogic | EO + Edge AI | Tasking API, onboard AI inference, multispectral imagery, hyperspectral option | $1M–$5M/yr | ~$30M of Palantir contract value |
| Tomorrow.io | Weather | Real-time atmospheric data, forecasting API, operational weather feeds | $100K–$500K/yr | ~$5M of Palantir contract value |
| Voyager / SIGINT | Space awareness | Space domain awareness data, orbital object tracking (government programme) | TBD — gov't contract | Defence programme value |
| Total direct licensing | $6.6M–$36.5M/yr | vs $480M+ charged by Palantir |
The core of Palantir's value is the AI Platform (AIP) and its Ontology — a knowledge graph that links objects, events, and entities across sensor feeds over time. This is buildable with open-source tools and commodity GPU infrastructure.
| Palantir AIP Capability | Independent Equivalent | Tools / Stack | Build Cost |
|---|---|---|---|
| Object detection in imagery | Custom CV model (PyTorch + TensorRT) | YOLOv9, SAM2, custom fine-tuned on satellite data | $3M–$8M |
| Change detection | Temporal diff + semantic segmentation | Siamese networks, ChangeFormer, custom pipeline | $2M–$5M |
| Multi-sensor fusion | EO + SAR + IR + Radar fusion engine | Custom transformer-based fusion, spatial-temporal indexing | $5M–$12M |
| Ontology / knowledge graph | Entity-relationship graph linking objects across time | Neo4j, Amazon Neptune, custom graph DB | $3M–$7M |
| Target identification | Military object classifier (tanks, missiles, ships, aircraft) | Fine-tuned ViT / EfficientDet on military imagery datasets | $4M–$10M |
| Trajectory prediction | Ballistic / vehicle path prediction model | Physics-informed neural networks + Kalman filter | $3M–$6M |
| Natural language query (LLM) | LLM interface over imagery + intelligence data | Claude API / GPT-4 API + RAG over structured intel data | $1M–$3M + API costs |
| Kill chain automation | Human-in-loop target package generation | Custom workflow engine + analyst review interface | $5M–$10M |
| Apollo (CI/CD deployment) | Kubernetes + GitOps deployment pipeline | K8s, ArgoCD, Helm — standard DevSecOps stack | $1M–$2M |
| Total AI engine build | $27M–$63M |
| Component | Specification | Purpose | Cost |
|---|---|---|---|
| Primary inference cluster | 64–128× NVIDIA H100 SXM5 | Real-time object detection on incoming imagery streams | $20M–$40M |
| Training cluster | 32–64× H100 (or cloud burst) | Model training, fine-tuning, quarterly retraining cycles | $10M–$20M or $2M/yr cloud |
| Storage (imagery archive) | 5–20 PB object storage | Raw imagery, processed tiles, vector detections, tracks | $2M–$8M |
| Network fabric | 400GbE InfiniBand interconnect | High-throughput data movement between GPUs and storage | $1M–$3M |
| Total GPU infrastructure | $33M–$71M |
Palantir's most technically advanced integration is with Satellogic — AI runs onboard the satellite before downlink. Two ways to replicate this:
| Option | Approach | Cost | Timeline |
|---|---|---|---|
| Option A (fast) | License Satellogic API directly — same partner Palantir uses. Deploy your AI models to their onboard processor via their edge AI SDK. | $1M–$5M/yr licence | 3–6 months |
| Option B (own) | Procure satellites with radiation-hardened NPUs (Ubotica/Spiral). Deploy your own models onboard your own fleet. | +$500K–$2M per satellite | 2–3 years |
The $300M figure assumed budget smallsats (1m–3m resolution). Here is the full picture across all tiers, including real-time video options:
| Tier | Satellite Class | Resolution | Video Capable | Unit Cost | 30-sat cost | 50-sat cost | Verdict |
|---|---|---|---|---|---|---|---|
| Budget | Smallsat — Planet Dove / ICEYE class | 1m–3m EO / 1m SAR | No (stills only) | $3M–$8M | $90M–$240M | $150M–$400M | Change detection, ship/vehicle tracking. Good enough for AI fusion. |
| Mid-Range ✓ Recommended | Planet SkySat / Satellogic / BlackSky Gen-2 class | 0.5m–1m EO | Yes — 30fps, 90s clips | $15M–$25M | $450M–$750M | $750M–$1.25B | Individual vehicles, aircraft ID, video stare. Palantir-equivalent tier. |
| High-Res | BlackSky Gen-3 / Maxar WorldView-class | 0.3m–0.5m EO | Yes — 30fps, limited duration | $35M–$80M | $1.05B–$2.4B | $1.75B–$4B | Licence plate-level detail. Full precision targeting. Premium tier. |
| Ultra-High (VLEO) | Albedo Clarity class — 320km orbit | 10cm EO + 2m thermal | Yes — continuous stare mode | $50M–$120M | $1.5B–$3.6B | $2.5B–$6B | Unprecedented commercial resolution. Individual person identification. Highest cost. |
| All-Weather SAR | Umbra / ICEYE XL class | 0.25m SAR + SAR video | Yes — SAR video mode | $15M–$30M | $450M–$900M | $750M–$1.5B | Works through cloud, night, smoke. Best for persistent all-weather surveillance. |
| Recommended mix: 15 mid-range EO + 10 SAR + 5 VLEO | ~$600M–$1.1B for 30 satellites | Balanced capability across all conditions | |||||
| Phase | Cost | Timeline | Capability Achieved |
|---|---|---|---|
| Phase 1 — MVP | $70M–$110M | 12 months | MetaConstellation + basic AIP equivalent. Real-time EO + SAR + AI object detection. First government demo-ready product. |
| Phase 2 — Full AIP | $120M–$200M | 30 months | Full Maven Smart System equivalent. Kill chain automation, ontology, all sensor types, C2 integration, classified deployment. |
| Phase 3 — Sovereign (budget tier) | $340M–$560M | 60 months | 30–50 smallsats at 1m–3m resolution + video mode. Full AI stack. Zero operator dependency. |
| Phase 3 — Sovereign (premium tier) | $700M–$1.5B | 72 months | 30–50 mid/high-res satellites at 0.3m–0.5m with real-time video feed. Albedo VLEO option included. |
| Total (full sovereign) | $440M–$760M | 5 years | Complete independent equivalent of Palantir's full stack + own constellation |
| Category | Annual Cost |
|---|---|
| Satellite data licensing (Phase 1–2: all 6 operators) | $15M–$40M |
| GPU compute + cloud infrastructure | $10M–$25M |
| Staff (200–350 engineers, AI, ops, BD) | $50M–$100M |
| Satellite operations (Phase 3 own fleet) | $20M–$40M |
| Cybersecurity + compliance | $5M–$15M |
| R&D + model retraining | $10M–$20M |
| Total annual OpEx | $110M–$240M/yr |
| Factor | Palantir (licence their platform) | Independent Build |
|---|---|---|
| Year 1 cost | $100M–$200M/yr (licence fee) | $70M–$110M (build) + $15M/yr data |
| Year 3 cost | $150M–$300M/yr ongoing licence | $110M–$240M/yr OpEx (own platform) |
| 5-year total | $750M–$1.5B (pure licence, no assets) | $440M–$760M (own platform + satellites) |
| Data sovereignty | All intelligence data flows through Palantir servers | Full control — air-gapped option available |
| Satellite access risk | Palantir can terminate — operators can withdraw | Own constellation = zero dependency |
| AI model access | Black box — you see outputs, not models | Full model ownership, auditable, customisable |
| Customisation | Limited to Palantir's product roadmap | Fully customisable to your specific threats/region |
| IP ownership | Palantir owns all IP | You own all IP — licensable to third parties |
| Time to first capability | 3–6 months (sign contract) | 12 months (MVP build) |
| Revenue potential | None — you are the customer | Sell to other nations — $100M–$500M/yr potential |
| Operator | Why | Integration | Priority |
|---|---|---|---|
| Planet Labs | Daily global coverage, large archive, mature API | Planet SDK (Python), STAC API | Day 1 |
| ICEYE | All-weather SAR — works at night and through cloud | ICEYE API, GeoTIFF delivery | Day 1 |
| Satellogic | Edge AI on satellite — sub-minute detection latency | Aleph platform API, edge AI SDK | Day 1 |
| Maxar SecureWatch | Highest resolution (0.3m) for precision analysis | SecureWatch WMS/WMTS/API | Month 3 |
| BlackSky | Sub-hourly revisit — fastest response to events | BlackSky API, webhook alerts | Month 6 |
| Tomorrow.io | Weather data improves SAR quality scoring + ops planning | REST API | Month 6 |
| Component | Technology | Notes |
|---|---|---|
| Object detection | YOLOv9 / RT-DETR fine-tuned on satellite imagery | Fine-tune on xView, DOTA, FAIR1M military datasets |
| Change detection | ChangeFormer / BIT / custom Siamese CNN | Pre-/post-event comparison at pixel level |
| SAR analysis | SARVIT / custom transformer for SAR data | Separate model pipeline — SAR ≠ optical preprocessing |
| Sensor fusion | Custom transformer with spatial-temporal attention | Fuse EO + SAR + IR + Radar into unified object track |
| Knowledge graph | Neo4j Enterprise or Amazon Neptune | Link detections across time, sensor, and location |
| LLM interface | Claude API (Anthropic) or GPT-4 Turbo + RAG | Natural language query over intelligence data |
| Inference serving | NVIDIA Triton Inference Server + TensorRT | Sub-100ms inference at scale |
| Training platform | PyTorch + Weights & Biases + MLflow | Experiment tracking, model registry, retraining pipelines |
| Data pipeline | Apache Kafka + Apache Flink + dbt | Real-time stream processing of incoming satellite data |
| Layer | Technology | Cost Model |
|---|---|---|
| Compute | On-premise NVIDIA H100 cluster (primary) + AWS/Azure GPU burst | $20M–$40M CapEx + $5M/yr cloud |
| Storage | Pure Storage / NetApp (on-prem) + S3-compatible object store | $3M–$8M for 10PB |
| Orchestration | Kubernetes (EKS/GKE) + Helm + ArgoCD | Infrastructure as code |
| Security | Zero-trust (BeyondCorp model), HSMs, AES-256, FIPS 140-3 | $3M–$8M setup + $2M/yr |
| Classified deployments | Air-gapped on-prem Kubernetes clusters, NSA Type 1 crypto | $5M–$15M per classified environment |
| APIs / delivery | REST + WebSocket + gRPC, API Gateway, CDN for imagery tiles | Included in engineering build |
To replicate Palantir's Phase 1 capability (MetaConstellation + basic AIP equivalent) within 12 months, the first 90 days are critical:
| Days | Action | Cost | Output |
|---|---|---|---|
| Days 1–30 |
Sign data agreements with Planet Labs, ICEYE, Satellogic Hire CTO + Chief AI Officer + 10 senior engineers Procure GPU cluster (32× H100 to start) Set up secure cloud infrastructure (AWS GovCloud or Azure Government) |
$12M–$20M | Data flowing. Team hired. Infra live. |
| Days 31–60 |
Ingest and index first satellite data feeds Deploy baseline object detection model (fine-tuned on xView dataset) Build orchestration API v0.1 — task satellites, receive imagery Set up analyst dashboard skeleton |
$8M–$12M | First AI detections on live satellite imagery. |
| Days 61–90 |
Add change detection pipeline Integrate SAR (ICEYE) into fusion layer Add Satellogic edge AI — onboard detections Internal demo — live detection of vehicles, ships, activity changes |
$5M–$8M | Demo-ready system. First government pilot conversation. |
| Day 90 total | Working prototype: 3 satellite feeds, 2 AI models, live dashboard | $25M–$40M | Functional MVP — government demo ready |
| Role | Background Required | Annual Cost |
|---|---|---|
| CTO | Space / defence AI systems, 15+ years. ex-Palantir, Planet, or national lab. | $400K–$700K |
| Chief AI Officer | Computer vision, satellite imagery AI. Published research or ex-DARPA/DIU. | $350K–$600K |
| Geospatial Engineering Lead | EO/SAR data pipelines, GDAL, STAC, cloud-native geospatial. | $200K–$350K |
| ML Platform Lead | PyTorch, TensorRT, Triton, GPU cluster management at scale. | $220K–$380K |
| Senior CV Engineers (×5) | Object detection, segmentation, satellite imagery domain. | $180K–$280K each |
| DevSecOps Lead | K8s, classified deployments, zero-trust, FedRAMP experience. | $200K–$320K |
True continuous real-time video from orbit does not yet exist commercially at global scale — but several systems deliver high-frame-rate video clips, persistent stare modes, and near-continuous coverage that enable real-time AI analysis. Here is every viable option as of 2026, from commercial to classified.
| System | Operator | Capability | Access |
|---|---|---|---|
| Starshield | SpaceX / NRO (US classified) | Continuous real-time EO + IR video stare. 183+ satellites launched by April 2025. $1.8B classified contract. Advanced IR sensors tracking ballistic + hypersonic missiles. Built with Northrop Grumman. First true persistent global video surveillance constellation. | US government only. Not commercially licensable. Partnership with US DoD required. |
| NRO EOCL | NRO + Planet + BlackSky + Maxar | Electro-Optical Commercial Layer — NRO contracts with Planet ($undisclosed), BlackSky ($1B/10yr), Maxar for priority tasking and rapid delivery of commercial imagery for classified use. | Allied nation intelligence agencies via Five Eyes / bilateral agreements. |
| French CSO / Optical | Airbus / DGA France | 0.35m resolution military EO. 3-satellite constellation. Priority tasking for NATO allies. Video mode available at classified tier. | French MoD + NATO partner nations. |
| Israeli Ofek / OPTSAT | IAI / Israeli MoD | Sub-0.5m military EO. Rapid revisit over Middle East. Real-time video mode for priority targets. | Israeli MoD + select partner nations. |
| Constraint | Why it exists | Workaround |
|---|---|---|
| No true continuous video yet | A LEO satellite passes over a point in 90–120 seconds. It physically moves too fast for a continuous stare from one satellite. | Multiple satellites in sequence — as one leaves, the next arrives. With 50+ sats over a region you approach pseudo-continuous coverage. |
| Downlink bandwidth | Raw 30fps 0.5m video generates terabytes per hour. Ground stations can't receive it all in real-time. | Onboard AI (Satellogic model) — only detections, not raw video, are downlinked. Reduces bandwidth 100–1000×. |
| Ground station contact windows | Satellite can only downlink when over a ground station — typically 8–12 min per pass. | ISL (inter-satellite links) relay data between satellites to the nearest ground station in real-time. |
| Cloud cover blocks optical video | Optical cameras see nothing through thick cloud. | SAR radar (ICEYE/Umbra) provides all-weather backup. Fuse both into AI pipeline. |
| VLEO orbital decay | At 320km (Albedo), atmospheric drag degrades orbit faster. Satellites need periodic re-boost. | Ion propulsion system on satellite. Adds cost but enables far better resolution than standard 500km orbit. |
| Satellite | Type | Qty | Role | Cost (own) |
|---|---|---|---|---|
| Albedo Clarity-class | VLEO 10cm EO + thermal video | 5–8 | Precision stare over highest-priority targets. 10cm resolution + thermal. Unprecedented detail. | $400M–$960M |
| SkySat-class | 0.5m EO video (30fps, 90s) | 10–15 | Broad area video tasking. 30fps clips. Palantir-equivalent imagery quality. | $150M–$375M |
| BlackSky Gen-3-class | 0.35m EO, rapid revisit | 8–10 | Sub-hourly revisit. Fills coverage gaps between video passes. Near-military resolution. | $280M–$600M |
| ICEYE / Umbra SAR class | 0.25m–0.5m SAR + video | 10–12 | All-weather backup. Night operations. SAR video for moving target tracking. | $150M–$360M |
| Satellogic-class + edge AI | 1m EO + onboard inference | 10–15 | High-volume coverage with AI detections transmitted in orbit. Lowest latency alerts. | $100M–$300M |
| Total recommended mix | 43–60 satellites | 43–60 | Persistent coverage, real-time video, all-weather, 10cm precision strikes capability | $1.08B–$2.59B |
Project MAX is built on a single core business model: long-term sovereign defence contracts with governments. The platform is contracted nation-by-nation, with each government signing 5–15 year agreements for persistent AI-ISR coverage, missile and drone early warning, and border surveillance. Phase 1 secures the first $10B–$20B in contracts. Phase 2 scales to a $500B multi-decade global contract portfolio across 20–30 governments worldwide.
| Region | Target Nations | Contract Type | Est. Contract Value | Term |
|---|---|---|---|---|
| NATO Core | UK, Germany, France, Poland, Netherlands | National ISR + Theatre Bloc | $1.5B–$4B each | 10 yrs |
| NATO Flank | Romania, Bulgaria, Baltic States, Norway | Theatre Bloc (shared) | $500M–$1.2B each | 10 yrs |
| GCC / Middle East | UAE, Saudi Arabia, Qatar, Israel, Jordan | National ISR + Dedicated Sat | $2B–$8B each | 10–15 yrs |
| Indo-Pacific | Japan, South Korea, Australia, India, Singapore | National ISR + Theatre Bloc | $1B–$3B each | 10 yrs |
| Southeast Asia | Philippines, Vietnam, Indonesia, Thailand | National ISR (standard) | $300M–$800M each | 5–7 yrs |
| Africa / Americas | Nigeria, Kenya, Brazil, Colombia, Morocco | National ISR (entry tier) | $150M–$500M each | 5 yrs |
| Total Portfolio | 20–30 governments | Mixed | $500B TCV | Avg 10 yrs |
| Year | Active Contracts | Annual Revenue | Cumulative TCV Locked | EBITDA Margin | Key Milestone |
|---|---|---|---|---|---|
| Year 1 | 1–2 gov | $500M–$1B | $3B–$6B | 20% | First founding contract signed, platform MVP live |
| Year 2 | 3–4 gov | $1.5B–$2.5B | $8B–$14B | 45% | Constellation Phase 1 live, 3 active ISR operations |
| Year 3 | 5–8 gov | $3B–$4.5B | $15B–$22B | 60% | Phase 1 portfolio complete ($10B–$20B TCV) |
| Year 5 | 10–14 gov | $6B–$9B | $50B–$80B | 68% | Full constellation live, Phase 1 renewals begin |
| Year 7 | 18–22 gov | $14B–$20B | $150B–$250B | 74% | Theatre blocs active, Tier 1 dedicated sat contracts |
| Year 10 | 25–30 gov | $28B–$40B | $400B–$500B | 78% | Phase 2 complete — $500B portfolio, exit event |
| Advantage | Why It's Hard to Copy |
|---|---|
| Sovereign satellite constellation | Capital-intensive. 4–6 year build time. Regulatory approvals per orbit. First-mover owns the orbital slots and government relationships. |
| 15-year contract lock-in | Once integrated into a government's defence infrastructure, switching cost is enormous — data history, cleared personnel, SOC integration, and political relationships all bind the client. |
| Non-US sovereign positioning | 20+ nations refuse or are restricted from using US platforms (Palantir, Maxar, NGA). Project MAX is the only credible non-US alternative at this capability level. |
| Classified AI training data | Models trained on real-world missile, drone, and CBRN signature data from live operations are not reproducible — every year of operation makes the AI advantage wider. |
| Network effect between contracts | Intelligence from 25 governments feeding the same AI layer means threat models trained on global data. Each new client makes the platform smarter for all clients. |