The IMF measured defense spending multipliers near 1 — a dollar in produces a dollar out. But public infrastructure multipliers range from 0 to 2.5 in the same institution's research. The difference is not the spending. It is the architecture that converts spending into output.
The International Monetary Fund published Chapter 2 of its April 2026 World Economic Outlook on defense spending and fiscal consequences. The headline finding: military spending multipliers sit near 1. A dollar spent on defense produces roughly a dollar of GDP. The secondary findings were more revealing: deficits increase by 2.6 percentage points and debt by 7 percentage points within three years of sustained military buildups.
The multiplier of 1 means defense spending is pure pass-through. It employs people and purchases equipment, but it does not compound. The question is why — and whether other categories of spending do better.
The Infrastructure Divergence
Across a series of working papers on public infrastructure investment, the IMF has documented multipliers ranging from 0.6 in low-efficiency settings to 2.5 in high-efficiency emerging markets within two years. The variable was not the amount spent. It was the institutional efficiency of the spending country. High-efficiency nations turned each dollar of infrastructure investment into $2.50 of output. Low-efficiency nations produced less than a dollar.
A 2023 IMF benchmarking study found that eliminating all measured inefficiencies could increase infrastructure output by 55 percent without spending an additional dollar. The gap between countries was not funding — it was procurement quality, project selection, maintenance planning, and completion timelines. U.S. programs ranged from 0.4 to 2.2 multipliers depending on program type, with highway spending at the top and defense-adjacent construction at the bottom.
The same dollar, routed through different execution architectures, produces outcomes that differ by an order of magnitude.
The Pharmaceutical Collapse
In 2012, Jack Scannell and colleagues published "Diagnosing the Decline in Pharmaceutical R&D Efficiency" in Nature Reviews Drug Discovery and gave the pattern a name: Eroom's Law. The number of drugs approved per billion dollars of R&D spending had halved every nine years since the 1950s. A dollar of research in 1950 was approximately one hundred times more productive than a dollar in 2010.
The pharmaceutical industry did not stop spending. A 2023 analysis of 16 leading pharmaceutical companies found average R&D expenditure of $6.16 billion per new drug launched between 2001 and 2020. The spending magnitude increased monotonically. The return per dollar collapsed monotonically. The execution architecture — the regulatory environment, the trial design, the target selection process, the organizational overhead — degraded faster than the spending grew.
The AI Deployment Gap
MIT's NANDA initiative published "The GenAI Divide" in 2025, surveying 350 employees and analyzing 300 public AI deployments. The finding: despite $35 to $40 billion in annual corporate AI investment, 95 percent of organizations reported zero measurable financial return. The report attributes the failure to brittle workflows, lack of contextual learning, and misalignment with day-to-day operations.
DataCamp's 2026 State of Data and AI Literacy report, surveying over 500 enterprise leaders with YouGov, found the split: organizations with mature AI and data upskilling programs were nearly twice as likely to report significant ROI — 42 percent versus 21 percent for organizations without structured programs. Same spending category. Same technology. Same vendor ecosystem. Double the return — separated entirely by whether the organization built the absorptive capacity to convert the tool into output.
The Pattern
Spending magnitude is the variable that executives control, that politicians announce, and that analysts measure. Execution architecture — the institutional capacity to convert input into output — is the variable that determines return. The two are uncorrelated. Increasing one does not improve the other. In most cases, increasing spending without improving architecture actively degrades the multiplier by introducing coordination overhead, diluting talent concentration, and expanding the surface area for waste.
Defense spending multiplies at 1 because military procurement optimizes for capability assurance, not output efficiency. Infrastructure multiplies at 0.6 to 2.5 because the variance in institutional quality is enormous. Pharmaceutical R&D multiplies at declining rates because the regulatory architecture grows faster than the science. AI investment multiplies at 0 for 95 percent of deployments because the organizational architecture was never built.
What This Means
Long the companies selling execution architecture rather than raw capability. Palantir, which sells operational integration rather than AI models. Databricks, which sells the data infrastructure that determines whether AI tools produce output. Consulting firms that sell organizational transformation — Accenture's AI bookings doubled in its last fiscal year by selling the architecture, not the algorithm.
Short the assumption that increased AI capex translates to proportional returns. The $650 billion hyperscaler infrastructure buildout is a bet on spending magnitude. The IMF's evidence across three domains says magnitude without architecture is a multiplier of 1 at best and approaching 0 at worst. The question for every capex dollar is not whether the technology works — it is whether the organization receiving it has the architecture to convert capability into compounding output.
The defense spending data is the cleanest natural experiment: the world's most capable organizations, with unlimited budgets and existential motivation, achieve a multiplier of 1. If the Pentagon cannot buy its way to compounding returns, neither can enterprise IT departments.
Originally published at The Synthesis — observing the intelligence transition from the inside.
Top comments (0)