Palantir Q2: The Costly Engine Behind Carrier AI Platforms

Palantir booked $1.94 billion of revenue in the second quarter of 2026, up 93% from roughly $1 billion a year earlier, on net income of $1.07 billion (CNBC, August 2026). US commercial revenue grew 149% to $764 million, and the company raised full-year guidance to between $8.15 billion and $8.16 billion. Somebody is paying for that growth, and among the payers are the insurance carriers that have rebuilt their underwriting stacks on Palantir's Artificial Intelligence Platform.

The Quarter in Numbers: Commercial Demand Outruns Government

Palantir built its reputation on government work, and the government business is still expanding at a pace most software companies would headline: US government revenue rose 90% year over year to $809 million (CNBC, August 2026). The story of the quarter sits on the other side of the ledger. US commercial revenue of $764 million grew 149% from a year ago, and on a compounded basis has risen 380% since 2024. Management now expects US commercial revenue in excess of $3.42 billion for 2026, raised from prior guidance of $3.22 billion. Adjusted earnings per share came in at 41 cents against the 35 cents analysts expected, and the year-ago comparison, roughly $329 million of net income at 13 cents per share, makes the $1.07 billion print a more than threefold increase in profit in four quarters.

Chief executive Alex Karp was not modest about it. "To my knowledge, no businesses at our scale has even grown half this much," he told CNBC on the earnings day, adding that the trajectory "looks like this is going to go on for at least another 18 months" (CNBC, August 2026). The market agreed with the direction: shares rose 12% on the release, recovering part of a 29% year-to-date decline driven by doubts about whether enterprise AI spending would hold up.

One figure in the release deserves more actuarial attention than it will get from equity analysts. Remaining US commercial deal value, the contracted work Palantir has signed but not yet delivered or recognized, more than doubled from a year ago to $6.24 billion (CNBC, August 2026). That number is not a forecast. It is a stack of signed commitments from enterprises, insurers among them, that have agreed to keep paying.

Where Insurance Sits Inside the Commercial Book

Palantir does not disclose an insurance vertical in its quarterly headline figures, so the carrier exposure has to be read from the deployments that are public. The clearest case is AIG. The carrier's chief executive Peter Zaffino has described a multi-agentic underwriting build on Palantir infrastructure, telling investors that AI is advancing faster than the company expected (Reinsurance News, 2026). That build now spans at least three layers: Foundry deployed into AIG's Lloyd's syndicate operations to run LLM agents against submission data, the AIG Assist underwriting assistant that the carrier credits with a 40% binding lift across eight lines in the first quarter, and an orchestration layer coordinating agent workflows across the underwriting organization (AI News, 2026).

actuary.info has tracked each layer of that stack as it appeared in filings and earnings commentary, and the pattern across them is consistent: the platform dependency deepens with each deployment, and each deployment is justified by the operating results of the previous one. AIG closed the first quarter of 2026 with $774 million of General Insurance underwriting income and an 87.3% combined ratio, improved from 95.8% a year earlier (AIG Form 10-Q, May 2026). Whatever share of that improvement management attributes to the AI program, the attribution itself becomes the argument for the next expansion, and the next contract.

AIG is the most visible carrier on the platform, not the only one. Palantir's bootcamp-to-contract sales motion has pulled in commercial customers across banking, health systems, and specialty insurance, and the doubling of remaining deal value to $6.24 billion says the cohort signed multi-year commitments at an accelerating rate. For the carriers inside that backlog, the relevant question is no longer whether the platform works. It is what the platform will cost at renewal.

The Renewal Math: What 82% Growth Guidance Implies About Pricing

Palantir's raised full-year guidance implies roughly 82% revenue growth for 2026. Growth at that rate comes from three places: new customers, expanded scope at existing customers, and price. The first two dominate today. But a vendor that has become operationally load-bearing for its customers, and whose chief executive is publicly forecasting at least 18 more months of comparable growth, holds pricing leverage that it has not yet fully exercised.

Carriers have seen this movie in other parts of the stack. Core policy administration vendors, catastrophe model licensors, and data providers all followed the same arc: an adoption phase priced to win the market, followed by a harvest phase priced to the switching cost. By the harvest phase, the switching cost is not the license fee. It is the migration risk of moving underwriting workflows, model pipelines, and years of configured logic off a platform that the front line now depends on daily. An underwriting organization that books a 40% binding lift on a platform assistant does not decommission that assistant over a rate card dispute.

The actuarial expense implication is direct. Technology expense allocated to underwriting sits inside the expense ratio, and a platform bill that steps up at renewal lands in the combined ratio just as surely as a reinsurance rate increase. The difference is that reinsurance pricing is negotiated annually by people who benchmark it obsessively, while enterprise platform renewals are negotiated by procurement teams against a vendor holding the operational keys. Expense-ratio planning for 2027 and 2028 should treat AI platform costs as a line item with real escalation risk, not a fixed technology overhead.

The arithmetic is worth doing explicitly, even in illustrative form. A carrier writing $10 billion of net earned premium at a 26% expense ratio carries $2.6 billion of expense; a $60 million platform relationship is about 0.6 points of that ratio, and a 30% renewal escalation moves the combined ratio by roughly two tenths of a point. On its own that is survivable. What changes the shape of the exposure is the consumption model. Agentic architectures meter by usage, not by seat: every submission triaged, every document parsed, every agent workflow invoked draws compute billed through the platform. A carrier that grows its agent-run volume 3x because the tools work well has tripled the metered base before any rate escalation applies. Reinsurance buyers would recognize the structure immediately: it is a variable-rated treaty where the exposure base grows with your own success, negotiated without a broker, benchmarked against nothing.

The Attribution Problem: Whose Combined Ratio Improvement Is It

There is a second, quieter actuarial issue inside the AIG showcase numbers, and it applies to every carrier that will point at operating results to justify platform spend. A 40% lift in binding across eight lines, the figure AIG attached to AIG Assist in the first quarter, is a production statistic, not a profitability statistic. Binding more of what underwriters see changes the mix of the book. If the assistant lets underwriters clear more submissions per day, the marginal risks bound are ones that previously fell out of the queue, and whether those marginal risks price adequately will not be visible in a combined ratio for several development years on the casualty side of the book.

The clean test is cohort separation: track loss emergence on business bound through the assisted workflow against business bound conventionally, controlling for line, segment, and vintage, and let the cohorts season before declaring the platform accretive. Few carriers run that discipline on their own technology programs, because the program's sponsors own the measurement. Pricing and reserving actuaries are the natural counterweight. An actuary who signs a rate filing or an opinion supported by AI-era loss picks has standing to ask whether the improvement attributed to the platform survives a mix-adjusted view, and the answer feeds directly back into the renewal negotiation: a vendor whose measured contribution is two points of loss ratio commands different pricing than one whose contribution is a faster queue.

Vendor Concentration Is Now a Model-Governance Question

The NAIC's model bulletin on insurer use of artificial intelligence, adopted by more than 20 jurisdictions as of mid-2026 (NAIC, 2026), reaches third-party AI systems explicitly: insurers are expected to maintain governance over models they buy, not just models they build, including vendor due diligence, validation access, and contractual audit rights. A carrier whose underwriting decisions flow through agent workflows on a single external platform has concentrated three exposures at once: operational continuity, model risk, and pricing power, all with the same counterparty.

That concentration is measurable, and regulators are starting to ask for the measurement. A reasonable internal exercise for any carrier with a material AIP footprint: quantify the revenue-weighted share of underwriting decisions that touch the platform, the recovery time if the platform were unavailable for a week, and the expense-ratio impact of a hypothetical 30% renewal increase. If any of those three answers is uncomfortable, the governance documentation should say what the mitigation is. Regulators reviewing AI programs under the bulletin's framework will eventually ask versions of these questions, and carriers that have quantified them first will have better answers and better negotiating positions.

What the Build-vs-Buy Ledger Looks Like After This Quarter

Palantir's quarter strengthens both sides of the build-vs-buy argument, which is what makes the decision genuinely hard. On the buy side, the results validate the platform: customers are expanding because deployments are working, and no internal actuarial technology team can match the engineering throughput that $1.94 billion of quarterly revenue funds. On the build side, the same results quantify the dependency: a vendor growing 93% with a $6.24 billion committed backlog and a public 18-month growth forecast is a vendor whose future pricing reflects its strength, not its customers'.

The carriers best positioned for the harvest phase will be the ones that kept optionality: internal teams that understand the models running on the platform well enough to migrate them, contract terms negotiated while the vendor was still in adoption mode, and expense projections that assumed the platform bill would grow faster than premium. The ones worst positioned will be those that treated the platform as infrastructure too foundational to price. Nothing in Palantir's second quarter suggests the bill gets smaller from here.

Three markers over the next two quarters will show which way the economics are tilting. First, AIG reports second-quarter results on August 6, and any quantified update on the AI program, binding lift, expense-ratio contribution, or expanded scope, doubles as a data point on how deep the platform now reaches into the book. Second, Palantir's third-quarter print will show whether the $6.24 billion backlog keeps compounding; a backlog that doubles again while US commercial growth holds near 149% would mean the lock-in is accelerating faster than any carrier's internal build could catch up. Third, watch for the first carrier to disclose AI platform spend as a discrete expense item in a 10-K or rate filing. The first such disclosure will hand every other carrier's actuaries the benchmark that renewal negotiations currently lack, and it will hand regulators reviewing programs under the NAIC bulletin a number to anchor their third-party governance questions.