Genesis Mission Phase II: DOE Bets $159M on AI-Driven Science

The U.S. Department of Energy has committed another $159 million to its Genesis Mission, the administration's flagship effort to fuse artificial intelligence with federal research infrastructure. The money funds 12 new Phase II projects across a deliberately wide range of fields. Read alongside a busy week for autonomous-lab announcements in both government and industry, the award looks like more than another grant cycle. It suggests AI-for-science is moving from proofs of concept toward working systems.

What DOE actually funded

The 12 Phase II awards bring the mission's Phase II portfolio to 14 projects in total. DOE also added six more Phase I awards to supplement its earlier round 1. The named projects include a Harvard-led effort on quantum computing error correction and an Oak Ridge-led project on designing quantum magnets. Other work targets fusion energy, biotechnology, geothermal energy and particle physics 1.

The breadth is the point. Catalysis and materials discovery often dominate talk about AI in science, but this round spreads money across energy production, fundamental physics and quantum hardware. Fusion and geothermal stand out because they tie AI work directly to DOE's energy mission rather than to pure research.

DOE Under Secretary for Science Darío Gil framed the awards as a step from promise to capability. He said the Phase II teams have already shown what happens when AI is integrated with the department's scientific resources 1. The phrasing matters. Phase II is meant to reward teams that cleared an initial bar, so the program now funds scaling as well as exploration.

The autonomy thread

The same week brought other Genesis-related news that points the same way. A roundup of agentic-AI research developments for October 5–13 noted that several Genesis selections fund robotics, embodied AI and what the program calls "Super Intelligence" projects. The goal of these projects is closed-loop, autonomous discovery testbeds 3. One of them, a SLAC-led Phase I project, will build a closed-loop agent for catalysis 3. Catalysis is still in the mix, then, but it is one piece of a larger portfolio.

The most concrete example came from Argonne. Its SYNAPS-I team demonstrated a conversational agent layer that takes goals in plain language and segments images using SAM3. The agent then reconstructs ptychography scans and decides on its own when to run higher-resolution scans after spotting regions of interest 3. The demonstration ran on DOE user-facility beamlines and Argonne's computing infrastructure 3. It is an early but real version of a "self-driving" experiment, in which software steers the instrument rather than just analyzing its output afterward.

Industry is moving in parallel

The private sector is pursuing similar ideas, though with narrower commercial targets. Danaher announced plans on October 7 for its first AI-powered autonomous lab, aimed at generating custom antibodies and other molecular tools 2. The company says the lab will combine AI, robotics and instruments from several of its operating companies in one connected workflow. It expects operation at scale in early 2027 2.

Danaher claims affinity-reagent discovery up to eight times faster and up to a tenfold rise in annual reagent output, from tens to hundreds per year 2. It also says each design will be verified and tested, with results fed back into the system 2. That feedback loop is the same closed-loop logic DOE is funding. Both are company and agency projections rather than independently confirmed results, and they should be treated that way.

On the software side, OpenAI's GPT-Rosalind upgrade, with pricing effective October 5, focused on agentic tool use and life-science benchmarks 3. This points to a market forming around AI agents built for laboratory work.

Why it matters

Taken together, these developments mark a shift in where the investment goes. The earlier wave of AI-for-science centered on models that predict structures or properties. The current wave connects those models to instruments, robots and feedback loops, so that systems can design, run and interpret experiments with less human handling. The weekly roundup called this a move from research prototypes to funded integration 3. That is a fair description.

The Genesis Phase II round is the clearest public signal of that shift. It spreads relatively modest sums across many domains instead of concentrating on one. That approach hedges bets and also tests whether a shared AI toolkit can work in fields as different as plasma physics and protein binding.

The open questions are practical ones. Can autonomous systems handle the messy edge cases that experienced experimentalists catch? Will speed claims hold up outside controlled demonstrations? And will $159 million go far enough at national-lab scale? The sources do not settle these. What they do show is that federal and commercial labs are now building toward the same target, with real hardware.