$242 Billion for AI Startups. Budget Cuts for AI Researchers. The Two-Track System That Is Remaking Science.

In the first three months of 2026, venture capitalists invested $242 billion in artificial intelligence companies. That figure — 80 percent of all global venture funding for the quarter — exceeds the total amount invested across every sector in any full year before 2018. OpenAI alone closed a $122 billion round. Anthropic raised $30 billion. xAI pulled in $20 billion. Four companies absorbed $188 billion, or roughly 65 percent of all venture capital deployed worldwide in a single quarter.

Now consider a parallel number: 17 percent. That is the current probability of winning a competitive NIH grant — the lowest in nearly three decades, down from 26 percent a year earlier. At the National Cancer Institute, the odds have dropped from one in 10 to one in 25. NSF's FY2027 budget request proposes cutting the agency to $4 billion, a 54.5 percent reduction that would eliminate roughly half its research capacity. The DOE's proposed cuts would slash $1 billion from the Office of Science.

These two realities exist simultaneously, in the same country, involving many of the same researchers. The result is the most dramatic funding divergence in the history of American science — and it is creating strategic decisions that every grant seeker needs to understand.

The Concentration Problem

The Q1 2026 numbers are staggering in absolute terms but even more revealing in their distribution. Of the $300 billion in global venture capital deployed, $250 billion — 83 percent — went to U.S.-based companies. Late-stage deals accounted for $246.6 billion across just 584 transactions, meaning the average late-stage round exceeded $400 million. Early-stage funding grew 41 percent year-over-year to $41.3 billion across 1,800 deals. Even seed funding rose 31 percent to $12 billion across 3,800 rounds.

But strip out the four frontier lab megarounds, and the picture shifts. The remaining $112 billion in Q1 funding was distributed across nearly 6,000 startups — still a record, but driven overwhelmingly by infrastructure, autonomous vehicles, robotics, and manufacturing applications. Basic research in AI safety, interpretability, fairness, and societal impact — the kind of work that typically lives in universities and is funded by NSF, NIH, or DARPA — received almost none of this capital.

This is not an accident. Venture capital funds products, not knowledge. A startup building a foundation model can project billions in revenue; a university lab studying the long-term cognitive effects of AI-generated educational content cannot. The market is performing exactly as designed. The problem is that the public institutions that historically funded the research venture capital will not touch are simultaneously being defunded.

The result is a two-track system: unlimited capital for AI applications and infrastructure, collapsing support for AI science.

What Federal AI Research Actually Looks Like Now

The federal government remains the largest funder of basic AI research in the United States, but its capacity is deteriorating on multiple fronts.

NSF's AI programs are caught in the broader budget crisis. The agency's $8.75 billion FY2026 appropriation preserved most AI-related programs, including the National AI Research Institutes, which fund multi-year collaborative projects across universities, and the new TechAccess: AI-Ready America initiative, which will establish up to 56 state-level coordination hubs with $1 million per year in funding. But 1,752 NSF grants worth $1.4 billion were terminated by DOGE in 2025, with the STEM Education directorate losing 839 grants ($888 million) and Social, Behavioral, and Economic Sciences losing 320 grants ($91 million). AI research that touched education, equity, or social science applications was disproportionately affected.

The merit review changes compound the problem. NSF reduced minimum external reviews from three to two, allowed one review to be conducted internally by NSF staff, and made panel discussions optional. A National Science Board survey found that only a bare majority of NSF staff believed they had sufficient expertise to evaluate every proposal. For AI researchers submitting cross-disciplinary proposals — the kind that blend computer science with social science, health, or education — the reduced review rigor increases the risk of proposals being evaluated by reviewers who do not fully understand the technical or domain-specific content.

NIH's AI investments are real but constrained. The agency has expanded its interest in AI/ML-driven clinical research, and PCORI recently opened $120 million in funding for patient-centered comparative effectiveness studies that explicitly welcome AI-augmented methodologies. But with NIH success rates at a 30-year low and the FY2027 request proposing a $5 billion cut, the competition for remaining slots is intense. Researchers whose AI work touches any area flagged by the administration — health disparities, environmental health, or DEI-related topics — face additional risk from the political screening mechanisms established by Executive Order 14332.

DOE's Office of Science is the quiet winner. Its FY2026 budget actually grew by 2 percent to $8.4 billion, and the DOE Genesis initiative is offering $293 million for AI-integrated science proposals with grants between $500,000 and $750,000 over nine months, serving as steppingstones to 3-year awards of $6 million to $15 million. For AI researchers working on materials science, particle physics, climate modeling, or energy systems, DOE remains the most stable federal funder.

DARPA continues to fund high-risk AI research through programs like CLARA (Compositional Learning-And-Reasoning for AI), which offers up to $2 million for high-assurance AI with mandatory open-source delivery, and broader BAA opportunities across its six technical offices. DARPA's funding is smaller in absolute terms but comes with fewer political constraints and faster award timelines.

The Industry Partnership Question

When the federal government retreats, the default assumption is that industry will fill the gap. The Trump administration has explicitly promoted this theory. But the data tells a more complex story.

Only 6.8 percent of U.S. research expenditures came from industry sources in 2023, according to AUTM. Industrial R&D spending is heavily concentrated in applied research and product development, not the curiosity-driven basic research that produces the foundational insights AI companies later commercialize. The transformer architecture that underpins GPT, Claude, and Gemini emerged from Google Brain's research division — but the attention mechanism it built upon drew on decades of academic work in neural networks, sequence modeling, and computational linguistics funded primarily by NSF and DARPA.

That said, the current environment is creating genuine opportunities for researchers willing to bridge the academic-industry divide. Sponsored research agreements, in which a company funds university research in exchange for licensing rights to resulting intellectual property, have grown steadily as federal funding has become less reliable. Joint ventures between university labs and AI startups are increasingly common, particularly in healthcare AI, materials discovery, and climate modeling.

The SBIR/STTR program, signed into law today after its six-month lapse, offers a formalized pathway. The reauthorization introduces Strategic Breakthrough Awards of up to $30 million for companies with prior Phase II awards and matching funds — a mechanism explicitly designed to bridge the gap between federal research and commercial deployment. DOD, NIH, and NSF are expected to publish new solicitations in late April through May 2026, with first Strategic Breakthrough Awards from DOD anticipated in Q4 FY2026 (July through September).

For academic researchers, the strategic question is not whether to engage with industry, but how to structure partnerships that preserve research independence while accessing capital that the federal government is no longer reliably providing.

Repositioning Your AI Research for This Market

The funding chasm creates specific, actionable implications for researchers writing proposals right now.

Lead with national security, energy, or health applications. The political environment rewards proposals that align with stated administration priorities. AI research framed as advancing American competitiveness, strengthening critical infrastructure, or accelerating drug discovery faces less scrutiny than work positioned around equity, education, or social impact — even when the underlying methods are identical.

Target DOE and DARPA before NSF. The Office of Science's 2 percent budget increase and DARPA's insulation from political volatility make these agencies more predictable funders for AI research than NSF in the current environment. The DOE Genesis initiative's $293 million specifically invites AI-integrated approaches, and DARPA's rolling BAAs provide continuous submission opportunities.

Build the SBIR bridge. If your research has any commercial potential, the reauthorized SBIR/STTR program should be a central element of your funding strategy. The new $30 million Strategic Breakthrough Awards and 90-day decision timelines represent the federal government's most aggressive attempt to connect research with deployment since ARPA-E's founding.

Use foundation funding to sustain what federal agencies will not touch. The MacArthur Foundation, Sloan Foundation, and the new wave of tech-philanthropy vehicles — including the GitLab Foundation's $4 million in AI-for-economic-opportunity grants — specifically fund work at the intersection of AI and social impact that federal agencies are deprioritizing. Foundation grants are typically smaller than federal awards but come with fewer restrictions and no political oversight requirements.

Do not ignore the private market entirely. The $242 billion in Q1 AI venture funding represents the largest pool of AI capital ever assembled. Academic researchers with applied AI expertise are in demand as advisors, consultants, and co-founders. A well-structured industry partnership can provide both funding and data access that no federal grant can match — provided the researcher negotiates publication rights and intellectual property terms that protect their academic mission.

Two Tracks, One Career

The AI funding chasm is not going to close. Venture capital will continue flowing to commercial AI at unprecedented scale. Federal research budgets will remain contested and politically volatile for the foreseeable future. The researchers who thrive in this environment will be the ones who learn to operate across both tracks — securing federal grants for foundational work, building industry partnerships for applied research, and using foundation funding to sustain the work that falls between.

The risk is that the two tracks diverge permanently: a commercial AI ecosystem that moves faster than regulation can follow, and a public research system too underfunded to study the consequences. That gap is where the most important AI research lives — and where the competition for remaining funds is fiercest. Tools like Granted can help researchers identify opportunities across federal, foundation, and SBIR programs to build the diversified portfolio this market demands.