Why DOE is an AI buyer in 2026
The Department of Energy is the largest federal funder of physical-sciences research and the steward of the U.S. national laboratory system. In 2026, DOE's AI budget is concentrated in three arcs: scientific machine learning that accelerates simulation and discovery, grid modernization and power AI that stabilizes a grid absorbing renewables and distributed resources, and national-security AI that supports nuclear nonproliferation, stockpile stewardship, and counter-WMD work. The Office of Science, NNSA, the Office of Electricity, ARPA-E, and EERE are all spending.
Precision Delivery Federal LLC is a Midwest-based AI small business aimed squarely at this work. We are SAM.gov active, NAICS 541512 primary, SBIR-registered, and we carry a production federal ML past performance anchor at a federal health agency. Geography matters in DOE work — we are a short drive from Ames National Lab and well-positioned for collaboration with Argonne, Fermilab, and Idaho National Laboratory.
DOE — Energy and Science AI Applications — Capability Fit Score
National labs we target
Each lab has a mission character and an AI-buying behavior worth knowing. The labs where our capability set fits:
- Los Alamos National Laboratory (LANL) — Nuclear weapons, nonproliferation, HPC-scale scientific ML, cyber. NNSA-funded work.
- Lawrence Livermore National Laboratory (LLNL) — Stockpile stewardship, HPC (Sierra, El Capitan), inertial confinement fusion, AI for simulation.
- Sandia National Laboratories (SNL) — Weapons engineering, cybersecurity, energy systems, autonomy.
- Oak Ridge National Laboratory (ORNL) — Frontier exascale system, AI for science, materials, bioenergy.
- Idaho National Laboratory (INL) — Nuclear reactor R&D, grid resilience, critical infrastructure cyber. Regional neighbor to Iowa.
- Argonne National Laboratory (ANL) — Aurora exascale, scientific ML, materials, energy storage.
- Pacific Northwest National Laboratory (PNNL) — Grid modernization, chem/bio, nuclear nonproliferation, national security.
- NREL — Renewable energy modeling, grid AI, power-systems optimization.
- Ames National Laboratory — Materials science, rare earths, our literal geographic neighbor.
DOE vehicles and opportunity paths
DOE SBIR / STTR
Roughly two annual solicitations, AI-heavy across program offices (Office of Science, EERE, NNSA, Fossil Energy, Nuclear Energy). Post-April-2026 reauthorization this is an active path. We are SBIR-registered and ready.
ARPA-E solicitations
Open topic and focused-program solicitations in energy breakthroughs. Small-team-friendly, outcome-oriented.
National lab subcontracts
Each lab runs its own procurement. For AI/ML work, a small business with FedRAMP-grade stack and production federal past performance is exactly the profile labs reach for.
DOE program office BAAs
Office of Electricity, Office of Science, EERE, NNSA all run BAAs relevant to AI/ML scope.
Consortium-led OTAs
Several OTA consortia intersect DOE grid and national-security work.
Governmentwide vehicles
OASIS+, Alliant 2, and GSA MAS occasionally used by DOE program offices for specialized professional services.
Where our AI/ML fits DOE mission
DOE AI scope is distinct from civilian-agency AI. The buyers are PhDs, the corpora are physical-sciences data, and the compute is HPC. Our fit categories:
Scientific machine learning
surrogate models that replace expensive simulation steps, physics-informed neural networks, operator learning (DeepONet, FNO) on PDE-governed systems, active-learning campaigns for materials and molecules.
Grid and power AI
short-horizon load forecasting, distributed energy resource orchestration, grid-edge anomaly detection, storm-impact prediction for utility operations, ML-assisted state estimation.
HPC-scale training and inference
distributed training on lab-scale GPU fleets, mixed-precision at scale, parameter-efficient fine-tuning of foundation models for science.
Materials and molecular discovery
graph neural networks for molecular property prediction, generative models for candidate structures, Bayesian optimization over synthesis campaigns.
Carbon capture and sequestration
ML for subsurface imaging, reservoir dynamics, sorbent screening, process optimization.
Nuclear nonproliferation analytics
ML over open-source, remote-sensing, and radiological signatures. NNSA-relevant scope.
Critical infrastructure cyber
ICS/SCADA anomaly detection, OT telemetry modeling, adversarial robustness testing of grid-control ML.
Reactor and fusion analytics
disruption prediction, plasma control ML, reactor diagnostics (relevant to INL, LLNL, PPPL-adjacent work).
Why a small business on DOE AI
DOE and the labs run aggressive small-business goals. Many lab procurements are set aside or strongly preference small business subs. A prime or lab-internal PI partnering with a SAM.gov-registered, NAICS 541512 small business with production federal ML past performance gets specialized AI capability and small-business credit in one move. We are also leaner than a large integrator, which means a higher fraction of SBIR and BAA budget flows to the actual science.
Beyond the economics, a small business is faster. Lab PIs working against a DOE SBIR Phase I deadline or an ARPA-E submission window do not have weeks to wait on a large prime's capture committee. We sign NDAs in a day and deliver technical narrative in under a week.
Past performance and what we bring
Our confirmed federal past performance is a production ML system at a federal health agency (HHS) — a live system, not a pilot, inside a federal ATO boundary, with NIST 800-53 controls and ongoing operations. That is directly relevant to DOE lab environments where ATO governance and audit logging are required. Pre-Precision, Bo delivered cloud migration and data platform engineering at federal consulting firms supporting HHS and IRS, and competed to Kaggle Top 200 globally — the competition-grade modeling skill that DOE science-ML work demands.
For DOE scope, we target and pursue rather than claiming delivered lab past performance. That honesty is how we earn a first task order.
Stack for DOE workloads
HPC frameworks
PyTorch FSDP, DeepSpeed, Megatron, JAX, distributed training on multi-node GPU clusters.
Scientific ML
physics-informed neural networks, neural operators (FNO, DeepONet), graph neural networks (PyG, DGL), active-learning pipelines.
Foundation models
Claude, GPT-4, Llama, Mistral via appropriate federal paths; parameter-efficient fine-tuning on domain corpora.
Data platforms
lakehouse architectures, governed science-data stores, experiment tracking (MLflow, W&B).
Cloud + on-prem
AWS GovCloud, Azure Government, on-premise lab HPC environments, hybrid bursting patterns.
Security
NIST 800-53 controls by default, audit logging, provenance on every generation, adversarial robustness testing.
Geography and why Iowa matters for DOE
Precision Federal is 50010. Ames is home to Ames National Laboratory (operated by Iowa State University under a DOE contract) — a DOE Office of Science lab focused on materials science and rare-earth processing. Iowa State's computational and data science ecosystem runs through the city. We are within a day's drive of Argonne and Fermilab and well-positioned to support INL work. For DOE program managers who value geographic diversity in their small-business partners, Iowa places us outside the D.C./Bay Area clustering that dominates most federal AI procurement.
How to engage
If you are a lab PI, a DOE program manager, a prime on a DOE IDIQ, or a commercialization partner on an SBIR or ARPA-E opportunity, email bo@precisionfederal.com. Include the lab or program office, the vehicle or topic, and the scope. We respond within 24 hours with a fit assessment, rough level of effort, and a teaming construct.