Investment in computing equipment tied to the AI boom has risen to about $400 billion per year in the United States—roughly three times the 2023 level—but the measured contribution of that investment to US GDP growth has been limited. A new report argues that part of the reason is a statistical blind spot: the value created by US-based fabless chip designers, most prominently Nvidia, is frequently produced abroad and therefore not captured in current GDP accounting.
How large is the understatement?
The report estimates that US GDP growth was underestimated by roughly 0.3 percentage points over the past year. If Nvidia’s growth continues at its present pace, the gap could grow to nearly two percentage points of annual growth by 2028.
Why does this happen?
The core issue is how GDP statistics treat factoryless (fabless) chipmakers. Companies that design chips in the United States but have them manufactured, assembled and sold abroad can generate substantial US value without recording conventional exports. Specifically:
- No goods export is recorded because no physical product crosses US borders;
- No IP export is recorded because foreign buyers do not make an explicit, separable payment for US-held intellectual property.
As a result, much of the value added by US design and R&D does not appear in the conventional GDP measures.
How did the authors verify the missing value?
The authors examined every statistical category where Nvidia’s value-add could plausibly appear—goods exports, IP exports, service exports and merchanting—and found no evidence that this value is being recorded in any of them. They confirmed this assessment with the Bureau of Economic Analysis (BEA).
Why it matters now
This statistical blind spot is not new, and it applies to other factoryless manufacturers besides Nvidia. Historically the unrecorded value was small, so it did not materially distort GDP. Nvidia’s very rapid expansion has made the unrecorded share large enough to meaningfully understate measured GDP growth.
Potential fixes
The report identifies two approaches to address the undercount:
- Follow the 2025 update to international guidelines, which calls for recording overseas sales by factoryless manufacturers as goods exports. Implementing this would bring the overseas sales into US export statistics and GDP.
- Alternatively, record the firms’ markups as IP exports. This would diverge from current international standards, but could avoid political resistance to classifying overseas production as US manufacturing.
Either approach would require methodological changes and international coordination and could take years to implement. Until changes are made, measured US GDP growth is likely to remain understated.
Scope and implications
The issue affects any US company that concentrates design, R&D and intellectual capital domestically while outsourcing physical production. Correcting the measurement would alter macroeconomic aggregates, affect policy assessments and influence international comparisons of GDP performance.
Report and acknowledgements
The full report is available on Epoch’s website. The authors thank JS Denain, Lucio Melito, Mike Waugh, Benny Kleinman, Greg Burnham, Josh You, and Elliot Stewart for their feedback and support.
