Nobody in the AI race has much reason to be the first to slow down.
The companies keep spending because falling behind could be fatal. Investors do not want to miss the firms that come out on top. Washington increasingly talks about technological capacity as national power, especially in relation to China. Meanwhile, some of the people building the most advanced systems are warning that the pace itself is becoming dangerous. Reuters analysis ↗
AI is where this is easiest to see right now. It is not the only place. Similar patterns run through defence, energy, finance, crypto and federal IT: government increasingly relies on capabilities it does not always keep inside government.
At the same time, the federal workforce has been cut sharply. Across 22 major agencies, nearly 256,000 net jobs disappeared between December 2024 and January 2026 — more than 11 percent of the workforce across those agencies. Eighteen of the 22 lost more than a tenth of their staff. GAO ↗
That leaves an awkward question. Does a smaller administrative state mean less power, or simply more of the machinery of power sitting somewhere else?
Taken one by one, most of the lines on this map are ordinary enough. Governments buy software. Investors invest. Presidents nominate regulators. Families do business. The picture changes when the same people, firms and pools of capital keep turning up in different parts of the system.
Every line below has a source behind it. Some are contracts, some investments, some nominations, family ties or regulatory decisions. Most look ordinary on their own. The map is useful because it lets you stop looking at them one at a time.
{{ panelBody }}
Some functions really do disappear when government shrinks. Others turn up in enterprise platforms, contractors and public-private programmes. Technical talent is being recruited through schemes built to move people between industry and government. The Pentagon is using loans and direct investment as industrial policy. Elsewhere, government remains the buyer, regulator or strategic customer of systems it would struggle to rebuild for itself.
The money often arrives first. Private capital finances the data centres, software, energy systems, defence technology and other assets government may later decide it cannot do without. If the bets work, the owners keep the upside. Once the capacity becomes strategically important, however, the consequences of failure are harder to keep private. The Bank for International Settlements has already warned that the size and financing of the AI investment boom create new financial-stability risks. Reuters ↗
Money and politics are hard to prise apart here. Some of the people in this map are also donors, advisers, platform owners, presidential nominees, contractors or members of family and investment networks close to executive power. The rules themselves are being argued over by people with money at stake in the outcome. On September 15, the FTC chair publicly questioned whether incumbent AI firms asking for both new regulation and antitrust exemptions could end up insulating themselves from competitors. Reuters ↗
None of this began in 2025. Washington has always had defence contractors, political donors, revolving doors and private firms doing work the state cannot — or chooses not to — do itself. What is harder to ignore now is the overlap: a relatively small set of companies, investors and people appearing across more of those roles while the financial and technological stakes keep rising.
How much private capacity can the state come to depend on before that dependence starts shaping what government is willing — or even able — to do in the public interest?
The grouping is a simplification: several nodes belong to more than one network, and companies may compete even when the same investors or people recur. Each connection names the mechanism that links the nodes — who nominated, invested, regulated, awarded a contract or held a role — and “via” shows the next step in that documented chain. Names in the outer ring remain visible; select or hover over a node to follow its connections.
Every line is sourced. It records a relationship, not a shared purpose. The graph brings contracts, investments, nominations, family relationships, regulatory decisions and institutional links into one place so the wiring can be inspected rather than guessed at.