House appropriators stalled on a $12 billion artificial intelligence research package Tuesday, splitting along party lines over whether half the money should flow to Pentagon applications rather than civilian science agencies. The impasse threatens a September 30 funding deadline and could delay grants that university labs rely on for model safety and alignment studies.
The Republican draft allocates $6 billion to the Defense Advanced Research Projects Agency and service-branch AI offices, citing competition with China. Democrats countered with a bill preserving National Science Foundation and National Institutes of Health shares at 2025 levels while adding $2.4 billion for NIST testing infrastructure.
Where the Money Goes
NSF currently funds roughly 38 percent of U.S. academic work on interpretability and bias measurement in large language models. NIH supports clinical AI trials and biomedical foundation models. DARPA focuses on autonomous systems and battlefield decision support. Each agency operates on different timelines; labs cannot redirect DARPA awards to open scientific publication without clearance.
Appropriations staff said the Republican plan would not cut total AI spending but would reorient it. Democratic staff argued reorientation equals cut for civilian researchers who lack classified channels to publish findings.
Shutdown Politics
Congress must pass twelve spending bills or a continuing resolution before fiscal year end. AI funding sits inside the Commerce-Justice-Science subcommittee bill, which also funds the FBI and federal courts. Leaders rarely pass single-issue AI measures without broader packages.
Moderate Republicans from semiconductor-heavy districts — Arizona, Oregon, New York — pressed leadership to decouple NSF from defense fights. Representative Mike Garcia said fabs and universities in his district share talent pipelines; starving NSF while boosting DARPA breaks that link.
House Freedom Caucus members conditioned their votes on riders barring any agency from requiring diversity disclosures in AI grant applications. That rider collapsed talks with the Congressional Black Caucus and progressive Democrats who control enough seats to block floor passage.
Research Community Reaction
Stanford and Carnegie Mellon administrators warned that three multi-year safety grants expire in October without renewal language. A letter signed by 140 university presidents said uncertainty alone is causing faculty to decline new PhD students for federally funded AI ethics projects.
Defense contractors submitted separate letters supporting DARPA increases. Lockheed Martin and Palantir cited operational testing needs; they did not oppose civilian funding explicitly but ranked Pentagon priorities first.
What Happens Next
Subcommittee chairs scheduled a markup for July 8. Senate appropriators released a bipartisan draft maintaining the 2025 split with a modest NSF increase, increasing pressure on the House to converge. White House budget officials said they would accept either formula if overall non-defense research does not fall below inflation.
Investors watching government AI spending should monitor continuing-resolution length. A three-month CR freezes new grant awards; a full-year CR at 2025 levels avoids shutdown but delays the expanded safety research both parties claim to want.
