Palantir grew 93% and paid a 1.4% tax rate — only one of those is in the headlines

Palantir is growing faster at $2 billion a quarter than most software companies manage at a tenth the size — 93%, accelerating for a third straight quarter, with margins that widened rather than thinned. That part is not in dispute. What the headlines leave out is that the company paid 1.4% tax on its profits, which flatters the earnings figure everyone quotes by about a fifth. The business is exceptional; the question this report ends on is whether a price of roughly 114× normalised earnings has already spent the next five years of it.

Valuation — this is the entire bear case

At $163.24 the market pays ~$391B for a company guiding to $8.15B of 2026 revenue: ~48× forward sales, where large-cap software lives at 10–20×. Nvidia, the most direct AI beneficiary in the market, trades at ~20× trailing and ~12× forward.

There is no clean P/E to quote, because the reported one runs on a tax rate the company itself says is not the long-run rate. Normalising FY26 — the adjusted operating income guide of $4.89B, less ~$1.05B of stock compensation and payroll tax, plus ~$0.6B of interest and other income, taxed at Palantir's own 23% — lands near $3.4B, or about 114×.

What has to happen. At ~114×, reaching a 30× multiple — still a premium to almost anything — needs earnings to grow about 3.8×: four years at a 40% CAGR, a bit over five at 30%. Today's price embeds roughly half a decade of compounding at rates almost nothing sustains, just to arrive at expensive.

One concession the bears should make: the de-rating already partly happened. Palantir entered this print ~41% below its 52-week high and down ~32% year to date — the business compounded at 85% through the first half while the stock fell.

Risk — each isolated, do not blur

1. Valuation (dominant). A 30% de-rating requires nothing to go wrong operationally. That is not hypothetical; it is what H1 2026 did while revenue accelerated.
2. The tax line. When the valuation allowance releases, reported EPS growth decouples downward from operating growth, and that quarter will look ugly regardless of the business.
3. Concentration. US revenue is 81% of the total, and growth is existing customers spending more — 1,049 of them, up 24%, against revenue up 93%.
4. Bookings. If revenue keeps outrunning bookings, growth converges down to bookings. The only question is when.
5. Non-operating income. $91.8M of "other income" includes $66M of unrealised gains on public stakes the company mostly cannot yet sell — it reverses if those positions fall.
6. Competition (ranked low, deliberately). Every hyperscaler sells an AI application layer. This print made that case weaker, not stronger: 149% US commercial growth is not what share loss looks like.

Horizon and sizing (kept separate)

Horizon. Not a next-quarter setup — Q3 is guided to +83% and Q4 to roughly +72%, and Palantir has beaten its own guide eight quarters running, so the near term is a game of by-how-much. It is worth knowing the deceleration is a guidance choice rather than comp arithmetic: hold Q2's own +18.5% sequential pace and Q4 prints ~$2.71B, or +93%. The thesis resolves over 3–5 years on one question — does ~$8B of revenue become ~$30B+ before the multiple normalises?

Sizing considerations (not a recommendation). Palantir is the rare AI name that doesn't carry capex-cycle risk — $14.6M of quarterly capex means no depreciation wave, no debt, no negative-FCF window — so it diversifies against the hyperscalers rather than doubling them. Against that: at ~48× sales the return is dominated by the multiple, not the operations, and H1 2026 proved it. A position sized for the business will behave like a position sized for the multiple.