The Department of Energy is deliberately passing on adding Grok to its AI toolkit, signaling that even aggressive product launches and government backing may not guarantee adoption in the federal sector. While xAI's Grok 4.5 model arrived this week with new coding and domain-specific capabilities, the Energy Department's top technology official made clear that vendor agreements alone don't drive real-world use.

What's Driving Federal AI Adoption Decisions?

The Energy Department operates Joulix, a suite of AI tools available to employees. Rather than adding every frontier model on the market, the agency is taking a selective approach based on actual employee demand and job performance needs. "I could have an agreement with every vendor, every frontier model out there," said Dawn Zimmer, CIO at the Energy Department. "But which ones do people really want? Nobody's going to use all of them."

This pragmatic stance reflects a broader shift in how federal agencies evaluate AI investments. The Energy Department's decision to expand its AI portfolio grew partly out of lessons learned from the Department of Defense and Anthropic dispute, which exposed the risks of relying too heavily on a single vendor. That conflict prompted President Donald Trump to direct agencies to phase out Anthropic's technology, though a preliminary injunction has paused that process.

Why Isn't Grok Gaining Traction in Government?

Grok, made by Elon Musk's xAI company, has faced a longer and steeper climb to adoption compared to competitors like OpenAI and Anthropic. The model has drawn concern from civic and public interest groups for its reported tendency to produce biased, inaccurate, and concerning outputs. Despite these challenges, Grok has made some inroads; the Department of Agriculture is backing its pursuit of FedRAMP High authorization, a security certification required for federal use.

Yet at the Energy Department, demand simply hasn't materialized. "There hasn't been a huge demand for Perplexity," Zimmer stated. "We're not probably going to add Grok at this point; I haven't had a lot of demand for it." This reflects a critical gap between vendor ambition and actual user need in the federal AI market.

Perplexity, which was the first AI platform to sign a direct deal with the General Services Administration in November 2025, has similarly struggled to gain widespread adoption despite its government credentials. The company received FedRAMP Low authorization and is part of pilots at the Department of Justice and Labor, yet uptake remains limited.

How Are Federal Agencies Evaluating AI Models?

The Energy Department's selection criteria extend beyond vendor relationships and security certifications. The agency considers which models will "give them the best data to do the job they need to do here at Energy," according to Zimmer. This job-specific evaluation framework means that even well-funded competitors with government backing may not make the cut if employees don't request them.

Current adoption at the Energy Department reflects this demand-driven approach:

• Claude dominance: Anthropic's Claude model has overwhelming employee support, with the Idaho National Laboratory, Lawrence Livermore National Laboratory, and Pacific Northwest National Laboratory all piloting Claude for coding tasks. The agency also uses Claude to generate first drafts of documents and search for internal information.
• Grandfathered status: The Energy Department purchased Claude in 2024 and was grandfathered into continued use even after Trump's directive to phase out Anthropic technology, giving the agency time to evaluate alternatives.
• Expanding options: Joulix currently offers access to Google's Gemini, with an OpenAI integration actively in development.

"Claude, everybody wants it," Zimmer said. "We were grandfathered in because we did purchase Claude in 2024 so we have some time to keep working with it."

Dawn Zimmer, CIO at the Department of Energy

The Energy Department's cautious stance toward Grok comes even as xAI has accelerated its product roadmap. SpaceXAI released Grok 4.5 this week with technologies from its $60 billion acquisition of AI coding company Cursor, emphasizing improvements in coding, legal, and finance domains. The model also highlights cybersecurity capabilities, an area of growing focus for the U.S. government.

What Does This Mean for xAI's Government Strategy?

The Energy Department's lack of interest in Grok underscores a fundamental challenge for xAI in the federal market. While Elon Musk has positioned SpaceXAI as a major player in the AI race, with the company's June IPO raising $85.7 billion and ongoing investments in gigawatt-scale data centers, government adoption requires more than technical capability and funding. It requires employees to actually want to use the product.

Musk has recently signaled a shift in tone toward competitors. On Thursday, he posted on X that he was "clearly wrong about Anthropic," praising the company as "obviously currently the leader in AI." This acknowledgment came after Anthropic forged a deal with SpaceX to use all compute capacity at the company's Colossus 1 data center in Memphis, Tennessee, for $1.25 billion per month through May 2029.

For federal agencies navigating the AI landscape, the Energy Department's experience offers a clear lesson: vendor relationships, government certifications, and product launches matter less than solving real problems for real users. As agencies continue to expand their AI toolkits in the wake of the Anthropic dispute, adoption will likely follow demand, not the other way around.