Palantir Reports Tonight: The AI Skeptic Debate Turns on a $1.03 Billion Line Item, Not the $2.16 Billion Headline

The revenue guide is the softest part of the setup. The number that carries the argument is U.S. commercial revenue, and it has to run about 35% above the second quarter's $764 million for Palantir's own full-year guidance to hold. The company guided 2026 U.S. commercial revenue to more than $3.424 billion; the first half delivered $1.359 billion, so the second half must clear $2.065 billion — more than $1.03 billion a quarter.

Two of those lines do the real work. U.S. commercial is the only one that speaks to whether enterprises are paying for outcomes rather than for tokens — the charge [Alex Karp] has levelled at the model labs. Remaining performance obligations are the only one that says whether last year's bookings were real. Clear both and the skeptics of enterprise AI spending lose their best exhibit. Miss the U.S. commercial bar, landing between $950 million and $1.0 billion, and the full-year guide quietly becomes a fourth-quarter problem: Q4 would need roughly $1.09 billion, or 43% sequential growth from the second quarter's $764 million.

In May, Palantir guided second-quarter revenue to $1.797–$1.801 billion and delivered $1.935 billion, 7.6% above the midpoint. The third-quarter guide of $2.160–$2.164 billion sits about 0.8% below the $2.18 billion the Street has pencilled in. The pattern matters more than any single beat: this management guides below its own run-rate and then raises the year, so the guide midpoint is a floor, not a forecast.

The AI panic of the last two weeks was never about demand. It was about who has to fund the machines. In the September quarter, Amazon spent $77.3 billion of capital against $42.4 billion of operating cash flow, Alphabet spent $51.4 billion against $42.3 billion, and Meta spent $33.9 billion against $32.8 billion. Palantir spent $14.6 million against $1.216 billion. The revenue was not in dispute; the financing was.

That column is the debate in miniature. Palantir's trailing twelve months produced $3.36 billion of free cash flow on $42.0 million of capital spending — a 54.6% free cash flow margin and 0.7% capital intensity — and it holds $9.41 billion of cash and short-term investments against $211 million of lease obligations and no borrowings. CoreWeave carried $51.6 billion of debt and finance leases against $6.4 billion of cash at June 30 and paid $1.87 billion of interest over the trailing year. If the price of money for AI infrastructure keeps rising, the damage lands on the balance sheet that bought the machines, not on the one selling the software that runs on them.

The connection between the two debates runs through enterprise budgets, and it runs with a lag. Palantir's customers are the same chief information officers and defence buyers who decide how much of a $220 billion Amazon plan or a $170 billion Microsoft plan gets converted into software licences and services. Credit markets do not touch Palantir's revenue this quarter. They touch it in 2027 if a rising cost of capital stops the AI budget of a large enterprise from expanding — which is why tonight's print is a leading indicator of the funding debate rather than an answer to it.

Palantir's growth engine now has named partners on both sides, and the September announcements show where the demand is coming from. NVIDIA put its Nemotron open models inside Palantir's Foundry and AIP on the Sovereign AI Operating System reference architecture, with the first deployment on NVIDIA's own supply chain. Nebius was named Palantir's preferred sovereign AI infrastructure partner on Sept 8. Fujitsu signed for AIP and Foundry and became a Global Forward Deployed Engineering partner to carry the architecture into Japan.

The partner map also explains the revenue mix. The fastest-growing line is U.S. commercial, where the buyer runs its own operations on AI infrastructure; the federal line carries sovereign-AI urgency attached to it. And it explains the asymmetry in the whole trade: Palantir's upstream is a set of vendors selling into a budget the market is currently punishing, while its downstream writes software cheques out of operating budgets. When the funding-cost debate eventually cools, the layer that requires no new capital at all is the one still standing.

Two things could still break the print. The first is arithmetic: the enterprise value works out to roughly 92 times the company's own $4.5–$4.7 billion 2026 adjusted free cash flow guide, so the equity needs the annuity to be both real and permanent. The second is stock compensation, $836 million over the trailing year and 13.6% of revenue, which sits outside the adjusted headline. Strip it out and trailing free cash flow falls to $2.52 billion, and the same enterprise value works out to 167 times.

The customer base is also less diversified than a 68-times-sales multiple usually implies. Government revenue was $1.85 billion of first-half sales, so a federal budget disruption would hit the line the equity story depends on. Capital return is the other pressure point: nearly half of the more than 250 questions shareholders submitted before this quarter's results asked about dividends, splits and buybacks — the loudest signal from the register is that holders want the cash returned rather than reinvested.

The honest reading of tonight is narrow and still useful. No company can tell you whether $800 billion of capital spending is a bubble, and Palantir certainly cannot. What it can tell you is whether the layer charging for AI outcomes is still compounding: if U.S. commercial revenue clears $1.03 billion with net dollar retention at or above 150% and remaining performance obligations above $5.4 billion, the demand side of the AI trade has its best single piece of evidence yet. Palantir's numbers settle who is paying for AI while the credit market keeps pricing who is paying for the machines — and the gap between those two answers is where the next twelve months of this trade get made.