The Internal Revenue Service is quietly deploying artificial intelligence to transform how it identifies tax fraud, while across the Atlantic, European tech companies are racing to build the infrastructure needed to power this AI revolution. These parallel developments reveal a fundamental truth: as governments and businesses rush to adopt AI, they’re confronting both the promise of efficiency and the practical challenges of implementation.
The IRS’s AI Audit Experiment
According to documents obtained by WIRED, the IRS paid Palantir $1.8 million last year to develop a custom tool called the “Selection and Analytic Platform” (SNAP). This AI system is designed to help the tax agency identify the “highest-value” cases for audits, collection of unpaid taxes, and potential criminal investigations. For decades, the IRS has relied on more than 100 business systems and 700 methods to select cases, creating what the agency itself describes as a “fragmented landscape” leading to duplication of effort and suboptimal case selection.
The SNAP pilot program focuses on three specific areas: disaster zone claims, Residential Clean Energy Credits, and Form 709 Gift Tax Returns. What makes this approach particularly interesting is its potential to analyze “unstructured data from supporting documents” – everything from Venmo transaction logs to Etsy storefronts to detailed property appraisals. As Mitchell Gans, a professor at Hofstra University, explains, when someone gives away valuable property like a private business, the IRS requires “balance sheets and statements of net earnings, operating results, and dividends” to determine value.
The Infrastructure Behind the Intelligence
While the IRS experiments with AI for tax enforcement, European companies are building the physical infrastructure needed to make such AI applications possible. French AI startup Mistral has raised $830 million in debt financing to construct Nvidia-powered data centers across Europe, with plans to deploy 200 megawatts of AI computing capacity by 2027. “Scaling our infrastructure in Europe is critical to empower our customers and to ensure AI innovation and autonomy remain at the heart of Europe,” says CEO Arthur Mensch.
This European push for “sovereign AI” infrastructure comes as demand for computing power reaches unprecedented levels. Just over half of Mistral’s revenues come from Europe, where concern about U.S. foreign policy and the Trump administration’s threats to reduce support for European allies have brought new urgency to calls for tech decoupling from Silicon Valley giants.
The Hidden Costs of AI Efficiency
Behind the scenes of this AI boom lies a significant challenge: massive inefficiency in computing resource management. ScaleOps, a startup that builds software to automatically manage and reallocate computing resources in real-time, recently raised $130 million at an $800 million valuation. The company claims its software reduces cloud and AI infrastructure costs by as much as 80% by addressing a fundamental problem: GPUs sitting idle, workloads being over-provisioned, and cloud costs climbing uncontrollably.
“As part of my role at Run:ai, I met many customers, especially DevOps teams,” says ScaleOps CEO Yodar Shafrir. “While they really liked what Run:ai provided, they still struggled to manage their production workloads, especially as inference workloads became more common in the AI era.” The company serves enterprise clients including Adobe, Wiz, DocuSign, Salesforce, and Coupa, highlighting how widespread this efficiency problem has become.
A Broader Economic Context
These developments occur against a backdrop of significant market volatility in the AI hardware sector. Just this week, U.S. memory chip stocks lost nearly $100 billion in market value after a Google research paper suggested AI-driven hardware shortages might ease. The paper introduced TurboQuant, an algorithm that compresses AI models without losing accuracy, potentially reducing memory requirements. As Morgan Stanley analysts noted, “If models can run with materially lower memory requirements without losing performance, the cost of serving each query drops meaningfully, resulting in more profitable AI deployment.”
Meanwhile, billionaire venture capitalist Vinod Khosla, an early OpenAI investor, has proposed eliminating federal income tax for Americans earning less than $100,000 by raising capital gains taxes. He argues this tax overhaul is necessary to address voter fears about AI taking jobs, predicting AI job anxiety will be “the single biggest issue” in the 2028 U.S. presidential election.
The Balancing Act
The IRS’s move toward AI-powered audits represents both the potential and the challenges of government technology modernization. As accounting professor Erica Neuman notes, the IRS has “basically never had a full successful modernization since the 1960s,” plagued by technical difficulties and political unpopularity. Under the Trump administration, the agency saw more than 25,000 people resign or accept early retirement between February and July 2025.
What emerges from these parallel stories is a complex picture of AI adoption. Governments seek efficiency through tools like Palantir’s SNAP, while companies race to build the infrastructure to support such applications. Yet both face practical challenges – from computing inefficiencies to market volatility to political resistance. As AI continues to transform everything from tax enforcement to data center construction, the real question may not be whether the technology works, but whether organizations can manage the transition effectively.