Palantir Just Grew 93% While Wall Street Kept Calling AI a Bubble

Palantir Just Grew 93% While Wall Street Kept Calling AI a Bubble

Let me tell you something that's been bothering me all week: the market spent months telling us AI was a bubble, and then Palantir reported a quarter that grew 93 percent year over year and the same market acted surprised. I've been running hosting infrastructure for over a decade, and I've learned to trust what the bills say more than what the analysts say. The bills at Palantir say enterprise AI is not a fantasy. It's invoiced, it's recurring, and it's growing faster than almost anything I've ever seen in software.

Bloomberg Tech ran the story Tuesday morning — "Palantir Surges as AI Demand Accelerates" — and it's the kind of earnings report that should make every independent hosting provider sit up and pay attention. Not because you're going to buy Palantir stock. Because this quarter is a signal about where real AI demand lives, and it is not where the doom loop on social media thinks it lives.

The Quarter That Should Have Ended the Debate

Let's put the numbers on the table. Palantir reported Q2 revenue of $1.94 billion, up 93 percent year over year, crushing the $1.8 billion Wall Street expected. GAAP earnings per share came in at $0.41 against $0.35 expected. US commercial revenue grew 149 percent to $764 million. US government revenue grew 90 percent to $809 million. CEO Alex Karp called it "otherworldly," and for once the hyperbole matched the spreadsheet.

Here's the part that matters more than the headline number: the company raised its full-year guidance from $7.65 billion to $8.15 billion to $8.16 billion — that's 82 percent annual growth — and guided Q3 to $2.16 billion, above the $1.997 billion consensus. Adjusted free cash flow margin is 63 percent. The Rule of 40 score hit 155, which for a software company at this scale is almost unheard of. US commercial remaining deal value jumped 124 percent to $6.238 billion. That is not a pilot program. That is customers signing long-duration contracts and putting money down.

The stock jumped 10 percent right after the print and another 17 percent in premarket trading Tuesday. And here's the kicker: this is the same stock that was down 30.8 percent year to date entering the quarter. Same company. Same products. Same market. The only thing that changed was the actual numbers.

The Sovereign AI Line That Should Scare the LLM Labs

Karp said something on CNBC that deserves more attention than the revenue beat. He said sovereign AI is winning, and he described it as an approach where customers control their data, their technology, and the value created from their AI deployments. Then he took a direct shot at the LLM labs — OpenAI, Anthropic, the token sellers — arguing that their model captures customers' intellectual property instead of helping them retain it.

I've been saying versions of this for months, and it's the single most important business dynamic in AI right now. Every enterprise I know that actually runs production workloads is terrified of handing its proprietary data to a token-metered API run by a lab that trains on everything it touches. They want the model, sure. But they want to own the pipeline, the fine-tuning, the deployment, the data. That's exactly what Palantir sells, and it's why customers are choosing it even when the political noise around the company is loud — the US government revenue grew 90 percent to $809 million in a quarter where the company was at the center of immigration-policy controversy. Businesses vote with contracts, not with op-eds.

For hosting providers, this is the most important takeaway in the whole report: the winning AI architecture is not "rent tokens from a lab." It's "deploy models where your data lives, on infrastructure you control." That trend is your tailwind.

The Market Signal Problem — Again

Here's what frustrates me as someone who runs real infrastructure: the market sent contradictory signals all year, and Palantir's quarter exposed exactly how broken the signal is. AI stocks sold off on "jitters" while actual AI demand accelerated. Palantir was down 30 percent for the year while its revenue was compounding at 93 percent. That's not a bubble correcting. That's the market guessing and the invoices proving it wrong.

I wrote before about the signal confusion in this market — when Google stock drops on a $205 billion capex plan while Supermicro surges on a $60 billion backlog, every supplier in the chain gets distorted pricing signals. Palantir is the same phenomenon on the software side. The market spent 2026 pricing AI as a speculation, and the companies actually selling AI to enterprises are reporting numbers that look like a different economy entirely.

Don't misread me: I'm not saying the correction narrative was pure fiction. There are real overbuild signs in the data center world — I've covered the cancellations and the lease pullbacks. But Palantir's quarter proves the demand side of the ledger is real. The problem is the market can't decide whether AI is a bubble or a boom, and it's swinging between both stories every earnings cycle.

The Counter-Argument — It's One Company, Not a Trend

The skeptics will say Palantir is one company with government contracts, political tailwinds, and a cult CEO — that its numbers don't prove enterprise AI scales. Jacob Bourne of eMarketer actually addressed this directly: Palantir has been "the clearest counter-example to the claim that enterprise AI doesn't scale past pilots, and accelerating growth makes that harder to argue with." Even Bourne flags the real open question: whether AI eventually collapses the software layer Palantir occupies. Fair point. Models get cheaper every quarter, and a lot of what Palantir does could get commoditized.

But here's the thing about the counter-argument: it's the same argument they made about AWS a decade ago, and about Salesforce before that. "It's one company." "It's a fad." "It'll get commoditized." Meanwhile the contracts keep getting signed. The 124 percent jump in remaining deal value is not a one-quarter artifact. Customers are committing to multi-year engagements. That's not a bubble behavior — that's a buildout behavior.

What This Actually Means for Independent Hosting Providers

First — read the demand signal correctly. Enterprise AI is not slowing down; it's concentrating into platforms that control the full stack — data, models, deployment. The "AI is dead" narrative is wrong for the same reason the "AI is everything" narrative was wrong: both ignore the invoices.

Second — position as the sovereign option. Karp just mainstreamed the phrase "sovereign AI" — customers keeping control of their data and their AI value. For an independent hosting provider, that's your exact value proposition against the hyperscalers. You're not a token meter. You're the place where a customer's data and models live under their control. Lean into that hard.

Third — don't build your capacity plan on the stock market's mood. The market whipsaws between bubble and boom every quarter. Your power contracts, your colo deals, your hardware orders — those are multi-year commitments. Plan against the 93 percent growth numbers, not against the 30 percent drawdowns.

Fourth — watch the LLM labs' response. If OpenAI and Anthropic start pivoting toward enterprise data-control products — and they will — that's a competitive shift in the market you serve. Palantir just made "we don't capture your IP" a selling point, and the labs are going to have to answer it.

The Structural Reality — Demand Is Real, the Signal Is Just Noisy

Here's the bottom line on the whole thing, ent? The AI infrastructure buildout has real problems — power, water, community consent, debt, overbuild in specific pockets. I've written about all of them, and none of that is fiction. But Palantir's quarter is a reminder that underneath all that noise, there is a genuine, invoiced, recurring revenue wave of enterprises deploying AI on infrastructure they control. The market narrative and the economic reality have decoupled, and that decoupling is exactly where independent providers can make money.

The hyperscalers are fighting over who gets to rent tokens to the world. Palantir just proved that a growing chunk of the market wants to own the whole stack instead. That's your opening. That's the wedge.

The Bottom Line

I don't own Palantir stock, and this isn't investment advice — I run servers, not portfolios. But I've seen enough boom-and-bust cycles in this industry to know what a real demand signal looks like, and 93 percent revenue growth with 63 percent free cash flow margin and raised guidance is a real demand signal. The market can call AI a bubble all it wants. The invoices say otherwise.

Your move: build the sovereign-AI-friendly infrastructure now — the private clouds, the data-control tools, the local model hosting — because the customers Palantir just proved exist are going to need somewhere to run that isn't a hyperscaler token meter. Be ready when they come knocking.

— Allan Ali, Founder

This article was produced with AI-assisted research and editorial support. Reporting is based on sources cited in the article.