Everybody has a statistic about enterprise AI. Relying on the survey you come throughout, 70% to 90% of tasks by no means make it previous the pilot stage. For the final two years, “pilot purgatory” has been enterprise AI’s defining narrative.
Then got here Palantir. If Wall Road desires proof that enterprise synthetic intelligence has escaped the lab, few corporations make a stronger case. Palantir has posted development charges that might be extraordinary for a startup, not to mention an organization anticipated to high $7.6 billion in income this 12 months. CEO Alex Karp told shareholders that the corporate’s outcomes for the primary quarter of 2026 display energy “that dwarfs the efficiency of basically each software program firm in historical past at this scale,” whereas positioning its Synthetic Intelligence Platform because the working layer for a way enterprises reorganize work.
The tech big just lately joined Nvidia, Microsoft, Meta, and greater than 20 different corporations in urging U.S. policymakers to not prohibit open-weight AI fashions, which enterprises can run and tune on their very own infrastructure. It additionally revealed a corporate blog post claiming that corporations threat handing their aggressive edge to main AI suppliers. In accordance with Palantir, these suppliers may take up a corporation’s institutional data—its “alpha”—and ultimately monetize it. The answer, it suggests, is AI sovereignty.
Taken collectively, the strikes learn like a doctrine. One targets Washington, arguing in opposition to regulation of the open-weight fashions enterprises want. The opposite targets CIOs, urging them to not give up institutional data to another person’s infrastructure. Palantir is making an attempt to form the foundations round who controls enterprise AI.
Forward of its earnings report on August 3, the market is concentrated on whether or not Palantir can justify one in all software program’s richest valuations. However the extra revealing insights sit beneath the headline numbers.
Palantir’s Largest Progress Engine
Final quarter, Palantir delivered one of many strongest performances in enterprise software program historical past. Income climbed roughly 85% 12 months over 12 months to $1.63 billion, whereas U.S. business income surged 133% to $595 million. It raised its full-year steering by practically half a billion {dollars} in a single quarter and closed 206 offers price not less than $1 million.
However of its $377 million business income improve over the previous 12 months, $352 million got here from present clients. Common annual income from its 20 largest clients jumped from $64.6 million to $93.9 million in a 12 months, and by the March quarter had reached $108 million on a trailing-12-month foundation. Web greenback retention hit 150%, far past regular software program benchmarks.
“Palantir’s 20 largest clients grew about 45% final 12 months. Everybody else grew 65%. Many of the development comes from present clients shopping for extra. However it’s getting much less high heavy, no more,” says Mike Leone, VP and principal analyst at Moor Insights & Technique.
That helps clarify Karp’s most revealing remark from the last earnings call: “Solely seven of our salespeople really even actually promote.” As an alternative, Palantir depends on AI bootcamps and forward-deployed engineers, or FDEs, embedded with clients. Software program corporations sometimes scale by decreasing human involvement. Palantir scales by putting engineers inside buyer organizations till AI stops being an experiment and turns into operational infrastructure.
But when enterprise AI succeeds as a result of distributors embed deeply inside their clients, the place does software program finish and consulting start?
The consulting-in-disguise argument “has been the usual knock on Palantir for years, and the margins simply don’t help it,” Leone says. “True consulting corporations like Accenture or Cognizant run working margins within the mid-teens. Palantir ran 46% within the quarter ending in March.”
If something, the trade is drifting towards Palantir. Amazon Net Companies and Microsoft are constructing their very own FDE teams; Microsoft’s runs about 6,000 folks. “Palantir constructed the corporate round FDEs from the beginning, and everybody else is discovering out enterprise AI wants them to get off the bottom,” Leone says. Karp made the identical case in his shareholder letter with a single determine: annualized income per worker of $1.5 million.
However Liz Miller, vp and principal analyst at Constellation Analysis, is much less satisfied. “Proper now, many FDEs are nothing greater than better-trained and extra subtle inside gross sales reps—now rebranded,” she says. Her problem to Karp is to display why the shift issues.
Nowhere is the strain between software program and embedded companies extra seen than in Britain.
The Sovereignty Paradox
Whereas Palantir tells enterprises to scale back their dependence on frontier-model suppliers, one in all its largest clients is wrestling with the other query: How dependent ought to it grow to be on Palantir itself?
Its Federated Knowledge Platform, a contract awarded in late 2023 and reportedly price as much as £330 million ($444 million at present change charges), was meant to function the digital spine for information throughout England’s Nationwide Well being Service, one of many 4 nationwide well being companies within the U.Ok. Palantir positioned it as ruled AI in motion. As an alternative, it turned one of many world’s most politically scrutinized enterprise AI tasks.
Parliamentary committees have urged ministers to think about a break clause, greater than 100 NHS information and expertise professionals want the platform replaced, and the CEO of NHS England, Sir Jim Mackey, has questioned whether or not its claimed advantages have been objectively assessed, calling for an unbiased evaluation. (Palantir didn’t reply to Quick Firm’s request for remark.)
“Governance was by no means constructed to stop a belief disaster,” says Leone at Moor Insights. “You’re shopping for lineage and entry management. That’s forensics.” However “governance can’t show the system labored or get the folks affected to agree, and people are the 2 issues that blow up belief,” he says.
Likewise, Miller at Constellation Analysis believes “enterprises aren’t shopping for trusted AI. CEOs should not shopping for trusted AI. They’re shopping for indemnified AI.” CIOs need the governance kill change, she says, “however Palantir is proving repeatedly that promoting the promise to CEOs is the trail to revenue.”
The U.Ok. authorities hasn’t moved to terminate the contract. However the episode exposes an irony in Palantir’s doctrine. The corporate warns that hosted suppliers might lock clients in and accumulate data they’ll’t reclaim, whereas U.Ok. lawmakers are asking structurally related questions in regards to the NHS’s dependence on a single international provider woven into public infrastructure.
Sovereignty operates in a different way at every layer of the stack. An enterprise can swap one mannequin for an additional. Changing the workflows, ontologies, safety insurance policies, and operational logic constructed over years is one other matter. That’s why Palantir’s largest clients preserve spending extra. The corporate is not only increasing income however dependency—whereas warning clients in opposition to relying on anybody else.
To Miller, that dependency isn’t a facet impact. “What Palantir is doing is promoting to CEOs higher than anybody on the promise of AI in a approach that CEOs perceive—however fully befuddles each CIO,” she says. “They’re efficient, however the group shall be tremendous dependent. That is the enterprise mannequin. And thus far, it’s working, as confirmed by that income advantageous print.”
The sovereignty marketing campaign’s timing isn’t unintended. Traders fear that OpenAI, Anthropic, and Google will transfer up the stack and grow to be enterprise software program corporations. Palantir’s counterargument: Fashions have gotten commodities. In his Might letter, Karp described mannequin corporations locked in a race the place token prices have fallen a thousandfold and “winners and losers swap locations each six months,” whereas Palantir sits “in a class of our personal.”
If intelligence turns into plentiful, management turns into the scarce asset as a substitute.
The Pre-Earnings Query That Issues Most
The extra sturdy query is whether or not Palantir represents the way forward for enterprise AI, or an exception proving how exhausting it nonetheless is. Present numbers level in each instructions. The truth that Palantir has greater than 1,000 business clients—together with Airbus, Bain, GE Aerospace, and Stellantis—suggests industrial corporations are embracing AI sooner than skeptics predicted. But the expansion stays concentrated, high-touch, and, in Britain, politically contested.
“Firms this measurement virtually all the time decelerate,” Leone says—and Palantir sped up as a substitute. Past subsequent quarter, he says, the actual metric can be geography. “U.S. income greater than doubled within the March quarter and nothing exterior the U.S. grew half that quick. That hole isn’t a expertise downside. Regulated consumers exterior America are deciding whether or not they need a U.S. firm working their operations.”
Maybe enterprise AI has arrived, simply not within the kind many anticipated. “Enterprise AI isn’t mainstream,” Miller notes. Token counts measure cultural readiness, “but it surely merely doesn’t measure the success of what was constructed, generated or delivered with AI. The adults have to enter this chat and assign actual enterprise worth and metrics to AI software.”
That’s why Palantir’s upcoming earnings matter past Wall Road’s expectations. They might provide the clearest sign but of what enterprise AI is changing into: software program that corporations deploy or infrastructure they’ll’t simply substitute. That distinction may decide not solely who wins the enterprise AI race, however what profitable finally means.
