Palantir slid about 7% after Michael Burry blasted the company’s AI positioning, saying Anthropic is “eating Palantir’s lunch” and pointing to a surge in Anthropic’s annual recurring revenue from $9 billion to $30 billion in months. The post was deleted, but the damage stuck, with software broadly softer as investors confronted the risk that budgets are tilting toward direct relationships with AI model providers. The selloff sharpened a long-running split on the stock: Wall Street is torn between Palantir’s federal moat and its premium valuation, while retail dip buyers keep showing up on weakness.
Burry’s critique is straightforward: Palantir looks less like a high-growth software pure play and more like a consulting-heavy platform built on human deployment. He has argued for years that Palantir’s reliance on Forward Deployed Engineers to embed with customers caps scalability and compresses margins. Palantir’s 10-K buckets much of that work under professional services. For bears, that is code for revenue that rides headcount, not code reuse or software margin. It is not the narrative that gets you a durable AI multiple.
By contrast, Anthropic sells access to intelligence through a plug-and-play API. If you believe Burry’s cited ARR acceleration to $30 billion and accept that enterprises are rushing to “easier, cheaper, and more intuitive” tools, the spending curve skews toward the model layer. That sets a high bar for Palantir’s AIP suite to show self-serve adoption that does not require teams of engineers on site. CIOs will pick lower-friction integrations when time-to-value is under pressure. With procurement cycles compressing, the simplicity premium goes to model providers.
The market got a real-time stress test in March when the Trump administration moved to ban Anthropic from federal systems amid a dispute over safety guardrails. Reuters reported Palantir was told to remove Anthropic’s Claude AI from parts of its Maven Smart Systems and rebuild pieces of the platform. The episode underscored a key vulnerability: if Palantir leans on third-party models, policy shifts can hit delivery timelines, margin, and customer satisfaction. Burry’s point about dependency risk is not theoretical anymore.
That said, Palantir’s federal franchise remains a powerful moat. Wedbush’s Dan Ives, who kept an Outperform and a $230 target, called out a “golden path” in government AI where security, auditability, and data provenance win deals. Palantir pitches itself as model-agnostic infrastructure, the connective tissue that makes AI useable in high-stakes environments like defense and healthcare. But model-agnostic is a tougher sell if procurement leaders decide to buy “the brain” directly from Anthropic or its peers and build lightweight orchestration around it. The regulatory whiplash also raises the cost of integrating cutting-edge models into mission systems when a policy pen can force a rip-and-replace.
Even Palantir bulls concede the stock’s multiple is a liability. Morgan’s Sanjit Singh flagged the setup: 10 straight quarters of accelerating growth make Palantir a clear winner in early AI adoption, but at roughly 38 times 2027 sales, it takes near-flawless execution and upside to sustain the tape. If Anthropic’s growth story dominates budget narratives this year, every incremental dollar flowing to model inference is a dollar not flowing to data operating systems. That dynamic pressures the top line and the multiple at the same time.
Today’s move was amplified by a broader software chill as investors digested headlines around Anthropic’s latest model push, including chatter about a new release called Mythos. When the market prices a best-in-class premium, prints and pipelines must consistently surprise. If Palantir’s next updates show heavy professional services mix, slower AIP module adoption, or any sign of federal program delays tied to model restrictions, the valuation debate tilts Burry’s way. Conversely, evidence of product-led expansion and rising contribution margins would reset the bear case fast.
Burry’s core insight is about where value accrues in the AI stack. Foundation models are pulling spend because they feel like product, not project. That dynamic favors Anthropic, whose private valuation has reportedly hit about $380 billion. He also spotlights Palantir’s long road to scale, noting “it took $PLTR 20 years to get to $5 Billion,” and his disclosed long-dated put position in 2025 telegraphed a multi‑year skepticism that today looks validated by the tape.
Where Burry may be early is in dismissing “plumbing.” In regulated environments, product is the system, not just the model. The ability to secure, govern, and operationalize sensitive data across agencies and primes is the difference between a demo and a deployment. Palantir’s bulls argue you cannot stand up mission-grade AI without a hardened data layer and workflow engine. If that’s right, model providers capture hype, but platforms capture renewals and multi-year budgets. The question is not whether Palantir’s layer matters; it is whether it can prove leverage without armies of FDEs.
The order book shows dip buyers are not gone. Retail interest tends to build when headline-driven selloffs reset entries, and today fit the pattern. But institutional money is still in price‑discovery mode, weighing a stock that trades like a category leader against a category that is still defining where profits settle. If software peers remain soft on Anthropic-related fears and macro risk puts a ceiling on multiples, Palantir’s path of least resistance is choppy until new contracts or disclosures change the narrative.
Investors are trained now to ask for hard evidence. Watch for AIP customer adds, the ratio of software to professional services, and contribution margins that speak to product leverage. Commercial net revenue retention will matter more if federal programs wobble under model restrictions. Remaining performance obligations and deal sizes can indicate whether customers are committing to platform standardization or simply testing modules while shifting AI spend to model APIs.
Palantir has three clear ways to regain control of its story. First, lock down marquee federal and Fortune 100 renewals that explicitly standardize on AIP modules with minimal on‑site services. Second, show that model dependence is diversified across providers and that Palantir can swap models without rewriting systems when policy winds change. Third, consider inorganic moves that deepen the company’s AI stack, whether through model‑adjacent tooling or edge inference partners that tighten the product story without trying to become a foundation model shop outright.
Burry forced the market to reprice a familiar risk: a premium platform in a sector where the gravity of value may be shifting upward to the model layer. If Palantir demonstrates that the data operating system is where governance, compliance, and mission outcomes live—and does it with software margins—the multiple holds. If not, even blockbuster quarters can fail to move the needle. The next print and the next wave of contracts will tell investors which side has it right.
