Every time I decide that Nvidia is the greatest stock on the planet, I remember Palantir. The Guardian is horrified by Palantir and its fire-breathing CEO, Alex Karp. My reaction is that if the Guardian doesn’t like Palantir, they must be doing something right. Read this analysis.
If youy don’t fancy reading it, it is quite long, the key message is that Palantir is uniquely placed to benefit from the strongest following winds in software history.
Digital-twin platforms are evolving into the twenty-first-century equivalent of enterprise-resource-planning suites, and Palantir (NASDAQ:PLTR) is positioned to own the orchestration layer of that stack. The market backdrop is unusually forceful: analysts expect the twin segment to compound at roughly 34% and reach about USD 156 billion by 2030, making it one of the largest incremental software opportunities of the decade. Hyperscalers have already removed the historical six-month integration slog; Amazon’s (AMZN) AWS IoT TwinMaker and Microsoft’s (MSFT) new Fabric Digital Twin Builder let customers spin up fully mapped replicas of factories or supply chains in hours, not quarters. Meanwhile, vendors such as NVIDIA (NVDA) and Schneider Electric are embedding cross-platform ontologies inside Omniverse so that twins can interoperate rather than fragment, unlocking the same network effects that propelled SAP and Oracle in the 1990s.
Into that opening steps Palantir’s ontology-driven operating system, which can sit natively on Azure Government, AWS, or Databricks, wrapping real-time telemetry, low-code agent builders, and sovereign-grade security into a single pane of glass. The result is a product set that turns falling AI-compute costs into a structural advantage: Stanford’s 2025 AI Index pegs GPT-3.5-class inference down 280-fold in just 18 months, while Andreessen Horowitz tracks a 1,000-fold slide across the past three years. As those costs keep compressing (driven by custom silicon like AWS Trainium 2, whose design brief is explicit to “meaningfully” lower inference pricing) the economic argument for deploying Palantir-powered twins becomes overwhelming. Boards rarely walk away from compounders of that magnitude, and the company’s latest quarter shows that they are not: revenue rose 39% year over year to USD 884 million, smashing consensus and prompting management to lift full-year guidance to 36% growth.
AI Cost Curve Accelerates Adoption
The secular decline in inference pricing is not linear but exponential, and Palantir is architected to harvest every new yard of efficiency. AWS’s Trainium 2 datasheets cite double-digit gains in price-performance versus last year’s silicon, and early benchmarks show Trainium instances renting for less than half the hourly rate of comparable NVIDIA H100 GPUs. Because Palantir’s Apollo layer is cloud-agnostic, customers can arbitrage whichever chip fleet offers the best economics on any given day, letting them recompile models without rewriting pipelines. That agility matters when the marginal cost of running a GPT-grade agent keeps shrinking by an order of magnitude every twelve months. Stanford’s data reveals that tasks once priced at USD 60 per million tokens in late 2022 now clear for six cents. In practical terms, a supply-chain planner can simulate hundreds of demand shocks overnight for the same budget that bought a single scenario two years ago, pushing the envelope of what “real-time” optimisation means.
Cheaper compute then feeds back into the twin thesis. IDC expects 35% of the Global 2000 to operate supply-chain twins by 2027, migrating from pilots to board-mandated systems of record. Palantir’s Foundry ontology already stitches together ERP, MES, and IoT signals; falling model costs simply widen the surface area that can be modelled in memory. The platform’s agent-builder lets non-technical analysts drag and drop large-language-model calls into workflows, eliminating one of the last friction points for mass adoption. MarketsandMarkets sees the AI-agent platform category itself exploding from about USD 8 billion next year to more than USD 52 billion by 2030 (a 46% CAGR) because enterprises do not have the developer capacity to code bespoke logic for every scenario. Palantir bundles that agent capability natively, turning every twin into an active decision-maker rather than a passive dashboard.
Finally, the architectural decoupling insulates Palantir from the GPU supply chain. Whereas vendors that rely on a single cloud SKU risk throughput shortfalls, Palantir can shift between AWS, Azure, and private clusters. Expect that resilience to matter as next-generation chips like NVIDIA Blackwell and AMD MI 300 crowd the market; the cheapest provider on any given quarter will win the incremental workload, and Palantir’s abstraction layer lets users follow price rather than rewrite code.
Edge and Sovereign Data Catalysts
Digital twins thrive on data density, and the edge is about to flood them with it. Gartner projects that 75% of enterprise-generated data will be created and processed outside traditional data centres by 2025 versus just 10% in 2018. IoT Analytics counts 18.8 billion connected devices today and expects that number to reach roughly 40 billion by 2030, a 14% CAGR. Each sensor upgrade widens Palantir’s addressable market because Foundry’s ontology can ingest streaming telemetry without schema surgery, letting operators see turbine vibrations, warehouse temperature, or fuel-burn anomalies in one semantic layer.
Low-latency connectivity is catching up. GSMA Intelligence reports that 5G links already top two billion and will blanket more than 40% of the planet’s population next year. That footprint slashes transport lag between edge nodes and Palantir’s cloud microservices, enabling closed-loop control of physical assets. On a factory floor, for instance, a Foundry-based twin can detect sub-millisecond anomalies, trigger an AI agent to adjust a robot’s torque and feed the new data back into the model, all without human intervention.
Data sovereignty, once a barrier, is turning into a tailwind. Confidential-computing revenue is forecast to rocket from roughly USD 24 billion in 2025 to more than USD 350 billion by 2032, clocking a 46% CAGR as regulators demand in-use encryption. AWS Nitro Enclaves already isolate sensitive workloads for healthcare and defense customers, and Palantir’s software is certified to run inside those trusted execution environments. The upshot is simple: enterprises no longer have to choose between regulatory compliance and real-time analytics. By abstracting enclave management, Palantir makes sovereign computing an invisible feature rather than an integration project, opening new verticals (banking, biotech, classified networks) that historically balked at sharing raw data.
Hyperscaler Competition Flywheel
Contrary to bear arguments that cloud giants will commoditize ontology layers, recent deals suggest the opposite. In the past twelve months, Palantir has inked or deepened formal alliances with all three major US clouds. Microsoft will deploy Foundry, Gotham, Apollo, and AIP inside Azure Government at Impact Level 6 and above, giving defense agencies a turnkey path to classified AI analytics. AWS markets Palantir as an Advanced Technology Partner and highlights joint wins such as the UK’s NHS and the US Army’s TITAN program in public keynotes, a signal that Seattle views the relationship as incremental, not cannibalistic.
The Databricks pact, announced in March 2025, may be the most strategic. Under the agreement, Unity Catalog and Palantir Virtual Tables deliver zero-copy access between lakehouse data and Palantir’s ontology, eliminating the ETL tax that has historically discouraged cross-platform analytics. Customers running heavy Spark batch jobs can surface those results instantly in AIP agents, while Palantir’s semantic layer feeds back real-time operational metrics into Databricks’ training pipelines. Every incremental query spins additional compute on whichever cloud hosts the lakehouse, so hyperscalers win even as Palantir captures orchestration economics.
Why do the clouds tolerate a seemingly competing layer? Because their bottleneck is GPU utilisation. Each new twin, agent, or simulation burns GPU hours (and whoever hosts the job collects the rent). As long as Palantir drives net-new demand, the large clouds have every incentive to cooperate. That dynamic mirrors the early SaaS-on-AWS era: Salesforce did not undermine EC2, it filled it.
Agentic Workflows Take the Helm
The human interface for enterprise software is shifting from dropdown menus to conversational agents. Gartner says 70% of new corporate apps will be built with low- or no-code tools by next year, up from less than 25% two years ago. MarketsandMarkets pegs AI-agent platform revenue at roughly USD 7.8 billion in 2025, surging past USD 50 billion by 2030. Palantir anticipated that pivot by embedding agent builders directly into AIP. Business users can now ask natural-language questions of the twin and receive an answer grounded in live ERP, IoT, and weather feeds.
Key to that capability is the ontology. Unlike generic chatbots that hallucinate, Foundry agents operate on a schema that has already been approved by domain experts, enforcing typed relationships between entities. That structure dramatically reduces the chance of generating non-compliant or nonsensical recommendations. Early adopters report cycle-time reductions from weeks to hours when rolling out new workflows because they bypass ticket queues and IT resources.
Falling compute costs bring the marginal cost of an additional agent toward zero, so organisations can spawn hundreds of specialised assistants rather than a single generic helper. When each line manager or field engineer can commission a personal agent without budget approval, usage skyrockets. Palantir, which charges on a platform-plus-cloud-usage basis, captures that volume through both subscription and infrastructure rev-share.
Financials
Palantir’s first-quarter 2025 print underlines the operating leverage beneath the narrative. Revenue rose 39% to USD 884 million, well ahead of the Street’s 31% forecast, driven by a 55% surge in US sales to USD 628 million. Commercial revenue from US customers grew 71% to USD 255 million, eclipsing government growth for the first time in company history. Gross margin held firm at 80%, defying concerns that rapid AI adoption would dilute the mix. Operating income hit USD 214 million, a 24% margin, and more than double the prior-year level, thanks to disciplined opex and falling unit compute costs.
Balance sheet strength remains a moat. Palantir exited the quarter with USD 4.3 billion in cash and equivalents against zero debt, preserving optionality for selective M&A or incremental share repurchases. Free cash flow was USD 301 million, converting at 34% of revenue (a level typically associated with mature software vendors, not 30%-plus growers). Management therefore raised full-year revenue guidance to a range of USD 3.89-3.90 billion, implying 36% growth at the midpoint, and lifted US commercial revenue guidance to 68% growth, signalling confidence that the AIP flywheel is only beginning.
Stock-based compensation, once the main bear talking point, fell to 18% of revenue versus 29% two years ago, a trajectory that suggests GAAP profitability will no longer depend on pro forma adjustments by 2026. With headcount growth decelerating relative to top-line momentum, that dilution trend should continue to improve.
Seeking Alpha, 26 June, LL Insights
Before I write more, here is a simple explanation of digital twin platforms.
Digital twin platforms are software systems that create virtual replicas of physical objects, systems, or processes. These platforms leverage real-time data and simulations to monitor, analyze, and optimize the performance of their physical counterparts. They serve as digital counterparts, enabling users to simulate, integrate, test, monitor, and maintain the physical twin.
AI Overview
Not everyone shares my and LL Insights’ enthusiasm.
What do analysts say about Palantir?
While chasing Palantir’s momentum is tempting right now, valuation analysis suggests the stock is too pricey. Palantir stock could be headed for a considerable reversion sooner than later — making this a questionable time to buy the stock.
Yahoo Finance, 9 July
What hardly anybody points out is that, according to the analysts, it has always been a questionable time to buy the stock. Analysts are a conventional bunch, finding it hard to think outside the box. This makes them similar to their institutional clients and explains why the latter are reluctant to buy Palantir, which is a retail trade.
If you want a share to rise a lot, you need plenty of bears who will eventually be forced to change their minds. It is also the case that many analysts and fund managers find Alex Karp’s gung-ho John Wayne approach to defending America hard to take. I expect they would appreciate it more if they found themselves on the front line.
Karp is cool for me, and I find the shares phenomenally exciting.
Share Recommendations
Palantir. PLTR
