Report · Federal Spending · Aug. 29, 2026
Machine Age Money
Federal awards tagged to AI, compute, and semiconductors total $55 billion across 4,818 records, and their composition shifted sharply after 2020: grant obligations grew roughly tenfold while contracts held flat.
Executive summary
Private capital is mobilizing around the physical layer of AI. On Aug. 28, a16z announced the $1.1 billion Machine Age Fund, targeting chips, memory, networking, storage, data centers, and robotics. Its stated premise is that the buildout is outrunning the supply chain: rack power climbing from roughly 5-10 kW toward 1 MW, and a hardware industry accustomed to 20-30% annual growth facing demand that requires triple digits. Federal spending on the same layer has a much longer history, and it is public record. We pulled every award in USASpending.gov whose record matches AI, compute, or semiconductor keywords: 4,818 awards totaling $55.0 billion in current obligations. Every figure below links to its official record. We detail:
- Where it goes: the Department of Defense accounts for $18.0 billion (33%), and the five largest agencies for 72%.
- How it changed: grant obligations grew from about $0.3 billion in 2019 to $2.8 billion in 2025, while contract obligations stayed near $3 billion per year.
- How it fragments: 64 awards started in 2008 and 713 in 2024; the mean award fell from roughly $22 million to $7 million.
- Who receives it: the ten largest of 1,592 recipients hold 33% of the total.
Agency shares
The Department of Defense accounts for a third of obligations.
Each rectangle is a funding agency, sized by its share of the total. Defense holds $18.0 billion of the $55.0 billion (33%), and Health and Human Services holds $11.2 billion (20%). Clicking an agency shows its recipients; clicking a recipient lists the awards behind the total. Within an agency, recipients outside the top 40 are grouped into one hatched tile. In Defense, those 678 smaller recipients together account for 27% of the department's total, more than any single recipient.
Note: Within an agency, the top 40 recipients are shown and the rest grouped as “others.” Recipients appear once per funding agency. Source: USASpending.gov award records matching AI, compute and semiconductor keywords; pulled Aug. 2026.
Composition over time
Grant obligations grew roughly tenfold between 2019 and 2025.
Contract obligations, which cover data centers, IT services, and hardware for government use, have run near $3 billion per year for a decade. Grant obligations moved differently: about $0.3 billion in 2019, $1.1 billion in 2021, and $2.8 billion in 2025. The timing is consistent with the CHIPS Act and the post-2022 expansion of AI research funding, though award start dates alone cannot establish causes.
GrantsContracts
Note: Obligations grouped by each award's start year; 2026 is a partial year. Event markers are context, not causal claims. Source: USASpending.gov.
Awards became more numerous and smaller.
Each dot is one award, placed by start date and amount. The set contains 64 awards starting in 2008 and 713 starting in 2024, while the mean award fell from roughly $22 million to $7 million: a shift from a small number of large infrastructure contracts toward a broad research program. The largest awards, labeled below, mostly predate the current AI cycle.
Note: Log scale; awards under $100,000 omitted for legibility. Hover any dot for the award; annotated awards link from the search below. Source: USASpending.gov.
Agency–recipient flows
Five agencies account for 72% of obligations.
The same totals shown as flows from agencies (left) to recipients (right); ribbon width is proportional to dollars. The five largest agencies account for 72% of all obligations in the set. Hovering a ribbon isolates one agency–recipient pair; clicking a recipient opens its award file.
Note: Top eight agencies and twenty recipients by dollars; smaller flows omitted. Source: USASpending.gov.
Concentration
The ten largest recipients hold a third of the total.
The ten largest recipients, mostly defense and health-IT integrators plus one chip foundry, hold 33% of the $55 billion, out of 1,592 recipients in the set. Clicking a name opens its award file.
Note: Totals sum every award to the recipient in this dataset. Source: USASpending.gov.
Data
The underlying records.
All 4,818 awards, indexed and searched client-side; no query leaves the page. Each result links to its USASpending.gov entry.
Methodology
Methodology and limitations.
Scope and caveats
- A slice, not the whole federal budget. These are award records whose recipient, agency, or description matched keywords like
artificial intelligence,machine learning,semiconductor,supercomputing,quantum,GPU,microelectronics,data center,neural network. - Every figure is verifiable. The record link opens the official USASpending.gov page for that award. Nothing here is computed or invented; amounts and names are as the government reports them.
- Keyword matching is coarse. A contract for a building's data center and a grant for a machine-learning model both match. Read the description before drawing a conclusion.
- Award amounts are current total obligations at pull time, which can revise. Years reflect award start dates. All charts and search run client-side.