Pentagon's AI Chip Billions Flow to Firms With Deepest Lobbying Roots
Federal records show DoD AI contract spending grew 1,507 percent in two years, and the empirical evidence says political connections — not merit — explain who wins.
The single most documented fact in the public record about American AI procurement is this: the Department of Defense increased its artificial intelligence contract spending from $269 million across 254 contracts in 2022 to $4.323 billion across 657 contracts in 2024, according to Brookings Institution analyses published in 2024 and 2026. That is a 1,507 percent increase in dollar value in 24 months. By 2026, contract counts had reached 1,319. The question federal records cannot yet fully answer — but begin to suggest — is whether that money followed merit or political access.
The empirical anchor for that question comes from an unlikely source: Bank for International Settlements Working Paper No. 1058, titled 'Evidence from Federal Contracts on the Value of Political Connections.' The paper's methodology examined federal contracting outcomes across industries and found that firms which engaged in lobbying received approximately one-third more in defense contract value than comparable firms that did not lobby. Similar premiums were documented for firms making campaign contributions and for firms with board-level revolving-door connections to government. Critically, the effect was most pronounced among firms with limited operational merit and contracts that received less scrutiny — sole-source awards, classified vehicles, and Other Transaction Authority agreements. The paper's authors conclude that results 'mainly, but not exclusively, support political connections over merit-based channels' as the explanatory mechanism. Applied arithmetically to a $4.323 billion AI contract portfolio, a one-third political connection premium represents a potential misallocation of over $1 billion in a single fiscal year.
The defense sector's lobbying investment is not incidental to this dynamic — it is structural. OpenSecrets documented that defense-sector firms spent $101 million on lobbying in the first three quarters of 2022 alone, a figure cited in Congressional Research Service Report R47751, which explicitly frames the defense-contractor-Congress relationship using the academic 'iron triangle' model drawn from Gordon Adams's 1981 study, 'The Politics of Defense Contracting.' CRS R47751 notes that this triangle — private industry, procurement agencies, and congressional committees — operates as a self-reinforcing system. Lobbying funds access; access shapes contracts; contracts fund further lobbying. By February 2026, OpenSecrets' own report, 'The AI-fication of K Street,' concluded that AI lobbying had become 'increasingly indistinguishable from defense lobbying' as AI transitioned from an emerging technology to a fixture of government contracting, and that civil society organizations advocating for safety and oversight maintained a 'significantly smaller' financial footprint than commercial sector lobbyists.
AnthropicCEO Dario Amodei provided the report's sharpest illustration of the tension between public positioning and financial entanglement. In January 2025, Amodei published a widely circulated essay lamenting tech's reluctance to challenge government and criticizing what he called 'extreme anti-regulatory policies' on AI, calling instead to 'put policy over politics,' according to Forbes reporting. Simultaneously, Anthropic was engaged in a $200 million dispute with the U.S. Department of Defense over an AI contract covering national security, warfighting, and enterprise uses, first reported by Axios. The public record does not resolve whether Amodei's policy advocacy influenced the contract dispute or vice versa. What the record does show is that a company whose CEO publicly criticized government entanglement was simultaneously negotiating a nine-figure government contract — a juxtaposition the available sourcing documents but does not explain.
On the electoral side, ABC News reported on April 2, 2025, that 'millions of dollars tied to artificial intelligence are pouring into the 2026 midterms,' with OpenAI specifically described as 'all in for 2026 midterms with government regulations looming.' The ABC News reporting noted that interest groups funded in part by AI industry leaders — including OpenAI-affiliated entities — were already shaping political advertising around the question of how government should oversee AI. This electoral spending is analytically consistent with a specific regulatory event: the Trump AI Action Plan of July 2025, which, according to Inside Government Contracts' reporting on the plan, directed the Department of Commerce and its CHIPS Program Office to 'remove extraneous policy requirements for CHIPS-funded semiconductor manufacturing projects,' to 'streamline regulations that slow semiconductor manufacturing,' and to 'review semiconductor grant and research programs to ensure that they accelerate the adoption of AI.' OpenAI, as one of the largest consumers of advanced AI chips, has a direct and quantifiable financial interest in regulatory frameworks that accelerate chip supply without imposing end-use conditions.
The CHIPS Act regulatory modifications deserve particular scrutiny. The original Biden-era CHIPS Act conditions — as documented in public legislative records — included domestic content requirements, workforce development provisions, and restrictions on the use of CHIPS funds by companies with significant Chinese operations. The Trump AI Action Plan's directive to remove 'extraneous policy requirements' could modify or eliminate those specific conditions, with direct financial consequences for the semiconductor firms that received CHIPS grants. The plan also references a 'new global chip export framework,' but the available text from the Inside Government Contracts report is truncated. Given that NVIDIA — which controls an estimated dominant share of the AI training chip market by multiple industry reports — has been the primary company affected by export controls on advanced AI chips to China under both the Biden and Trump administrations, any modification to that framework would carry quantifiable revenue implications. Those implications remain undisclosed in current public records.
What the assembled public record establishes is a documented feedback loop: DoD AI contract spending grew 1,507 percent in two years per Brookings; BIS Working Paper 1058 establishes empirically that lobbying firms capture one-third more defense contract value than non-lobbying equivalents; the defense sector spent $101 million on lobbying in three quarters per OpenSecrets; AI lobbying has merged structurally with defense lobbying per OpenSecrets' 2026 analysis; and AI-industry electoral spending is already shaping the 2026 midterms per ABC News. Both parties' elected officials sit on Armed Services and Appropriations Committees that authorize this spending, and both parties have received AI-industry contributions documented in FEC filings. The loop does not have a partisan color — it has a financial one.
What remains hidden is, in some ways, more important than what has been documented. The specific vendors receiving the 1,319 DoD AI contracts logged in 2026 Brookings data are not publicly identified in that analysis — a gap that a systematic FOIA request to USASpending.gov's contract database would begin to close. The specific FEC filing identifiers for OpenAI-affiliated PACs and Super PACs contributing to 2026 midterm vehicles are not resolved in available sourcing — a direct OpenSecrets FEC database query for executives including Sam Altman and associated entities would provide them. The exact 'extraneous policy requirements' being removed from CHIPS Act grants have not been specified in public regulatory filings — a Federal Register review and Commerce Department FOIA would establish the precise before-and-after conditions. And no published study has yet applied the BIS Working Paper 1058 methodology specifically to AI chip and AI software contracts — an academic gap that, if filled, would either confirm or refute the central analytical inference of this investigation. Until those instruments are deployed, the architecture of the money trail is visible. The load-bearing details remain in the dark.