ChatGPT scooped up 90% of Congress’s AI cash last year, per new records

When it comes to spending actual money on AI, Capitol Hill has been sending a very clear signal: they’re an OpenAI shop. A new analysis of House disbursement records by CNBC shows that of at least $113,740 in total AI spending over the year ending March 31, nearly 90%—roughly $100,580 across 798 transactions—flowed directly to ChatGPT. Anthropic’s Claude was a very distant second, pulling in just $13,160 from 37 transactions. That’s not a gap; it’s a rout.

The partisan spending split is equally stark. Democratic offices shelled out $54,165 on AI tools, more than triple the $15,782 that came from Republican offices. Whether this reflects different attitudes toward the technology, different operational needs, or simply who controls the committees with bigger budgets is an open question the data alone can’t answer. These numbers also don’t capture any secret weapon: free accounts or AI capabilities bundled into broader software contracts, meaning the real usage footprint is almost certainly larger and harder to trace.

How are staffers actually using these tools? According to the report, it’s the unglamorous, high-stakes work that keeps the legislative branch running. They’re generating first drafts of memos, summarizing dense policy research, analyzing legislation, preparing hearing materials, sorting through constituent correspondence, and even drafting social media posts. It’s a classic efficiency play—offloading cognitive grunt work to a machine so humans can focus on strategy and judgment. For an institution often criticized for being technologically antiquated, this is a quiet but significant operational shift, with ChatGPT functioning as the default cognitive intern for hundreds of staffers.

What’s surprising isn’t just the dominance, but the lack of a competitive landscape. Claude’s reputation for safety and nuanced reasoning hasn’t translated into federal spending share, at least not yet. The report suggests congressional users have voted with their procurement cards for the most familiar and widely-hyped tool, but this degree of single-vendor dependency always raises eyebrows. In any other enterprise context, a 90% concentration on one supplier for a critical productivity function would trigger a risk assessment. For a branch of government dealing with sensitive policy and constituent data, that conversation seems overdue.