Skip to content

Reviewed edition · qasb report 10/2026 · Topic report · Economy

AI and the economy: Trade in AI goods grows, productivity unclear

October’s sources mainly measure trade, data centres and financing in the AI build-out, often in nominal values that include price effects. An economy-wide productivity effect has not been measured officially.

Published
October 27, 2026 · as of October 26
Reading time
13 min
Publisher
qasb Research
Method
qasb methodology v1

Key findings

  1. −1.1%in the level of global GDP by 2028, relative to baseline, in the IMF’s model scenario of an AI market correction.1Evidence: scenario
  2. 10%of global investment-grade bond issuance in 2026 (to July) came from big cloud operators, up from under 1% at end-2024.2Evidence: observed
  3. 67%more AI-enabling goods trade in the first half of 2026 than a year earlier, in dollars, partly from higher unit values.9Evidence: observed
  4. 41.7%higher than a year earlier: nominal private US construction spending on all data centres, January–August 2026.4Evidence: observed

1.0Summary

AI-enabling goods made up 47% of the growth in world goods trade value in the first half of 2026, partly price-driven. Large cloud and data centre operators (hyperscalers) accounted for a rising share of bond issuance. Most German venture capital went to AI start-ups in the third quarter.

  • The World Bank puts US import values for AI-enabling goods in early 2026 nearly 60% above a year earlier. The ECB says the euro area is less specialised in them than some rivals.
  • Private US construction spending fell 4.4% in 2026 (to August) but rose nominally for data centres. In Texas, ERCOT paused approvals to energise new data centres of 75 MW or more.
  • In the comparatively small sterling corporate bond market, AI hyperscalers accounted for 47% of issuance in 2026, according to the Bank of England.
  • On productivity there is a descriptive sector comparison from Ireland without causal evidence, a market-implied estimate and IMF scenarios, which are not forecasts.

2.0Starting point: AI goods trade is growing strongly in value

In value terms, the AI build-out is clearly measurable in merchandise trade, in AI-enabling goods such as semiconductors and servers. The WTO, World Bank and IMF measure it with different definitions and in values with price effects. Part of the increase reflects higher unit values. The ECB describes price pressures and the euro area’s position, and UNCTAD models how the value added of one AI server rack is split.

Figure 1

Growth in trade in AI-enabling goods

Change in value from a year earlier, in US dollars, according to the WTO

Nominal values, not volumes: according to the WTO, part of the increase reflects steeply rising unit values for accelerators and memory. The 2026 value compares the first half with a year earlier. Nothing is said about use or productivity.Evidence: observed · Sources 9

According to the WTO, AI-enabling goods such as semiconductors and servers accounted for 47% of the growth in the value of world merchandise trade in the first half of 2026. Their trade rose by 67% year on year, lifting their share of merchandise trade to 14.8% (2016 to 2023: 7 to 8%).9A1

According to the World Bank, US imports of AI-enabling goods rose by nearly 60% year on year in early 2026 while imports of other goods fell by nearly a quarter. In the twelve months to April 2026 the US imported US$739 billion of AI-related goods, one third of it from East Asia and Pacific.5A2

The IMF describes the AI economy as highly concentrated: China and the United States accounted for more than 75% of net imports of AI-related computing hardware in 2025 and 57% of data centres built since 2024, while Taiwan and Korea source 62% of net trade in this hardware. According to the IMF, some larger economies in sub-Saharan Africa are largely absent from the AI-driven technology upswing.1A3

In its September 2026 projections, the ECB expects euro area exports to grow more slowly than foreign demand in the shorter term and to lose further market share; according to the ECB this partly reflects strong global demand for AI-related goods such as computing, storage and networking equipment for data centres, in which the euro area is less specialised than some competitors.19A4

Further findings

  • In an UNCTAD model estimate for one AI server rack, 68% of traced value added goes to profits and less than 15% to workers.6A5
  • According to the ECB, AI-related products accounted for about two thirds of China’s export price inflation in May 2026.19A6

3.0Data centres: construction spending, costs, grid connections

All sources in this chapter come from the US: nominal construction spending on data centres is rising, contacts in three Fed Districts anecdotally report effects on the labour market, costs and materials, and in Texas ERCOT paused approvals to energise new data centres and crypto facilities of 75 MW or more. The US energy agency EIA does not show data centres separately in its annual electricity sales statistics.

Figure 2

Large-load connection requests in Texas

Conditional classification of large loads of 75 MW or more in ERCOT’s Batch Zero process, September 2026

Connection requests are not realised demand and may include speculative projects. Large loads include crypto mining and other consumers besides data centres. Conditional classification for planning years up to 2032; elsewhere the document gives 125.4 GW for studied load.Evidence: observed · Sources 16

The US Census Bureau estimates private construction spending on data centres at US$45.2 billion in January–August 2026, 41.7% more in nominal terms than a year earlier; total private construction spending fell by 4.4% over the same period.4A7

In the Texas grid operated by ERCOT, large loads of 75 MW or more totalling 66.4 GW (204 projects) were conditionally classified as base load and 127.9 GW (158 projects) as studied load in September 2026; following a directive by the governor on 3 August 2026, ERCOT paused approvals to energise new data centres and crypto facilities of 75 MW or more.16A8

In the Federal Reserve’s Beige Book of 14 October 2026, contacts in three Districts report effects of data centre construction: in the Kansas City District it has crowded out other production and put pressure on the labour market, in the Dallas District it is driving up labour costs and drawing workers away from other industries, and in the Richmond District builders cited challenges with materials tied to data centre demand.11A9

Further findings

  • In a small regional Cleveland Fed survey, 43% of firms reported direct or indirect involvement in data centre construction.10A10
  • According to the EIA, US commercial electricity sales rose in 2025, but the increase cannot be attributed to data centres or AI.15A11

4.0Financing: bonds, interconnections, concentration

The build-out is increasingly financed through bonds and through stakes among business partners. The IMF, BIS and Bank of England describe leverage and interconnection, not the probability or timing of a correction. In Germany most venture capital in the third quarter went to AI start-ups, a substantial part of it to two very large rounds of dual-use and defence start-ups.

Figure 3

Venture capital for German AI start-ups

Funds raised according to KfW Research, rounded

The 2026 value covers the year to date up to the third quarter. In the third quarter, AI start-ups accounted for 69% of German venture capital.Evidence: observed · Sources 12

According to the IMF's October 2026 Global Financial Stability Report, hyperscalers' share of global investment-grade bond issuance rose from less than 1% to 10% since 2024. Fourteen circular financing transactions disclosed since 2023 comprise about US$300 billion of equity and US$650 billion of debt in exchange for US$1.4 trillion of future revenue commitments.2A12

In a BIS analysis of 972 investment relationships among AI firms in 2021–2025, 46.4% of the disclosed value of these AI-to-AI deals (16.1% of deals) involved firm pairs that also had supplier or customer relationships; 55.2% of investment flowing into AI firms came from other AI firms.3A13

A reassessment of expected AI productivity gains could affect not only AI-related asset valuations but also sovereign debt markets, the Bank of England warns.

According to the record of the Bank of England’s Financial Policy Committee, AI hyperscalers accounted for 47% of sterling corporate bond issuance in 2026 up to the meeting on 25 September. The Committee warns that a reassessment of expected AI productivity gains could affect not only AI-related asset valuations but also sovereign debt markets.20A14

According to KfW Research, German AI start-ups raised about EUR 3.1 billion in 61 financing rounds in the third quarter of 2026, 69% of total German venture capital volume; since the start of the year the figure was about EUR 6.3 billion, roughly twice the previous record year 2025 (EUR 3.2 billion).12A15

Further findings

  • In an UNCTAD table that does not name the market explicitly, the ten largest stocks held 37% of market capitalisation in August 2026.6A16

5.0Use and competition: consumers, firms, markets

Card data in an NBER working paper without peer review show that only a small share of US consumers paid for major generative AI services in April 2026. Free or employer-paid use is not included. According to the World Bank, far fewer firms in India than in the US reported AI use, and comparability is not explained. Competition questions concern Google’s search and, in a model, money market funds.

Card data covering 174 million consumers show that in April 2026 about 2.9% of US consumers paid for at least one major generative AI service; annual customer retention at major providers ranged from 20% to 45%, and switching rose around major model releases.7A17

According to the World Bank, about 23% of firms in India reported using AI, compared with 43% in the United States, with a wider gap for more sophisticated uses; at the same time, supply links between South Asian suppliers and US customers grew faster in AI-exposed sectors after 2022.13A18

Annual customer retention at major AI providers ranged from 20% to 45%, and switching rose around major model releases.

Further findings

  • In September 2026 the UK CMA proposed allowing AI assistants in principle on the planned choice screen for Google search.17A19
  • If AI removed institutional investors’ inattention in an NBER model, money market fund fees would fall but liquidity shortfalls would rise.14A20

6.0Growth and productivity: AI effect not measured officially

In official statistics, AI’s contribution to growth and productivity is not yet shown separately. The findings in this chapter therefore approach the question indirectly, each with its own limits. For the US, the Joint Economic Forecast estimates growth contributions of AI-related investment, while for the EU no comparable analysis is possible because of data limitations.

Figure 4

IMF model scenarios for the AI build-out

Deviation of the GDP level from baseline; scenario A by 2029, scenario C by 2028

Model scenarios, not forecasts and not measured effects; the IMF gives no probabilities. Scenario A assumes global total factor productivity 0.8% higher over ten years. The world figure of −1.1% includes amplification through financial markets; the direct effect is up to 0.4%.Evidence: scenario · Sources 1

Ireland's statistics office CSO recorded 26.6% higher labour productivity in information and communication from 2023 to 2025 while firms' AI use rose (28.4% to 61.3%), but stresses that a causal link cannot be confirmed; in the manufacturing, energy and water supply sector group, productivity fell in 2024 despite higher AI use.8A21

The German institutes’ Joint Economic Forecast (autumn 2026) estimates that four AI-related types of investment (IT equipment, software, research and development, data centres) contributed a gross 0.7 percentage points to US growth of 2.1% in 2025 and one percentage point to 2.4% in the first half of 2026; for the EU the institutes see no increased momentum in ICT investment in 2024 and 2025.18A22

The IMF's October 2026 World Economic Outlook baseline assumes no additional acceleration of productivity growth from AI. In its model scenarios, the level of global GDP is 0.6% above baseline by 2029 with a stronger AI build-out and 1.1% below baseline by 2028 in an AI market correction with tighter financial conditions.1A23

Further findings

  • The UK statistics office ONS plans experimental estimates of AI compute services’ GDP contribution only after the first quarter of 2027.21A24
  • From stock price reactions, an NBER estimate infers that by December 2025 investors expected AI in software engineering to add 3.6% to GDP.22A25

7.0Evidence

Not every statement in this report is equally well supported. The table shows the key statements with their evidence type, AI relationship and sources.

StatementEvidenceAI linkSources
67% more AI-enabling goods trade in the first half of 2026 than a year earlier, in dollars, partly from higher unit valuesobservednot stated9
ECB: euro area exports grow more slowly than foreign demand, partly due to AI goodsmodel estimatenot stated19
41.7% higher than a year earlier: nominal private US construction spending on all data centres, January–August 2026observednot stated4
Texas: ERCOT paused approvals for new data centres of 75 MW or moreobservednot stated16
10% of global investment-grade bond issuance in 2026 (to July) came from big cloud operators, up from under 1% at end-2024observednot stated2
German AI start-ups received 69% of venture capital in the third quarter of 2026observednot stated12
About 2.9% of US consumers paid for a major AI service in April 2026observednot stated7
AI-related investment contributed a gross 0.7 percentage points to US growth in 2025model estimatenot stated18
−1.1% in the level of global GDP by 2028, relative to baseline, in the IMF’s model scenario of an AI market correctionscenarionot stated1

The 25 reviewed statements from 22 original sources fall into chapters 2.0 (6), 3.0 (5), 4.0 (5), 5.0 (4) and 6.0 (5), of which 18 observed, 4 model estimate, 2 scenario, 1 open. “Not stated”: the edition does not assess the AI relationship of these statements separately. The scope and limitations of every statement are listed in the appendix (A1 to A25).

Original sources checked on October 26, 2026.

8.0What organisations can do now

  1. Check dependencies on AI hardware: according to the IMF, Taiwan and Korea source 62% of net trade in AI computing hardware, and the ECB reports rising price pressures for technology-related goods. Monitor suppliers and prices in purchasing.
  2. Settle grid connection, staff and materials for data centres early: contacts in three Fed Districts anecdotally report effects on the labour market, costs and materials, and in Texas ERCOT paused approvals to energise new data centres of 75 MW or more.
  3. Check exposure to data centre construction: in a small regional Fed survey, 4% of involved firms earned at least half of their revenue from this business. The build-out is increasingly financed through bonds. Plan for order risks from a market correction.
  4. Measure the effect in your own organisation rather than assume it: in Ireland productivity in one sector rose as AI use grew, without a confirmed causal link, while in another sector group it fell in 2024 despite higher AI use.

9.0Method and limitations

Reporting period: October 1, 2026 up to the data cutoff of October 26, 2026 at 1:00 PM GMT+1 (Berlin time). This topic report is published before the end of the month; sources published after the cutoff follow in the next edition of the series. Compiled: October 27, 2026 at 12:23 AM GMT+1. This is the first baseline. The sources come from different years; their publication dates are not events in the reporting month. No change from an earlier edition was measured.

The report uses only reviewed public original sources. Customer conversations, private dossiers and personal customer assessments are not inputs. Observed, model estimate, scenario and open describe the evidence for each finding. Source values retain their original units and baselines; they are not combined into an overall score.

Editorial approval: qasb editorial team · Approved: October 27, 2026 at 12:25 AM GMT+1. The editorial presentation arranges the reviewed statements and contains no figure that is not in them. Presentation approved by the qasb editorial team · October 27, 2026 at 2:27 PM GMT+1.

Limits

  • This topic report covers one topic area: “Economy”. It is not an exhaustive survey of all studies or incidents.
  • A narrow study finding is not a forecast for a particular customer, occupation or security system.
  • No qualified finding does not prove no effect; source-access gaps are disclosed separately.
  • This report contains no illustrative measurement series or invented confidence rating.

10.0Sources

  1. 1World Economic Outlook, October 2026: Resilience Under Strain, Urgent Choices — Chapter 1: Global Prospects and Policies (incl. Box 1.2)International Monetary FundOctober 13, 2026www.imf.org/-/media/files/publications/weo/2026/october/english/ch1.pdf
  2. 2Global Financial Stability Report, October 2026 — Chapter 1: Beneath the Resilience: Rising Risks, Unresolved VulnerabilitiesInternational Monetary FundOctober 13, 2026www.imf.org/-/media/files/publications/gfsr/2026/october/english/ch1.pdf
  3. 3Circular relationships among AI firms (BIS Bulletin No 137)Jon Frost, Rudraksh Kansal, Kumar Rishabh, Vatsala Shreeti & Leanne Si Ying Zhang, Bank for International SettlementsOctober 1, 2026www.bis.org/publications/bulletin-137-circular-relationships-among-ai-firms
  4. 4Construction Spending, October 1, 2026: Value of Private Construction Put in Place, Not Seasonally AdjustedU.S. Census BureauOctober 1, 2026www.census.gov/construction/c30/pdf/priv.pdf
  5. 5East Asia and Pacific Economic Update, October 2026: Riding the AI WaveWorld BankOctober 6, 2026doi.org/10.1596/978-1-4648-2362-6
  6. 6Trade and Development Report 2026: The geoeconomics of developmentUnited Nations Conference on Trade and Development (UNCTAD)October 9, 2026unctad.org/system/files/official-document/tdr2026_en.pdf
  7. 7Competition in Generative AI: Evidence from Consumer Demand (NBER Working Paper 35953)Mark L. Egan, Dominic Russel, Claire Shi & Vincent Yao, National Bureau of Economic ResearchOctober 26, 2026www.nber.org/papers/w35953
  8. 8Productivity in Ireland 2024–2025: Artificial IntelligenceCentral Statistics Office IrelandSeptember 29, 2026www.cso.ie/en/releasesandpublications/ep/p-pii/productivityinireland2024-2025/artificialintelligence/
  9. 9Global Trade Outlook and Statistics, Update: October 2026World Trade OrganizationOctober 8, 2026www.wto.org/english/res_e/booksp_e/gtos1026_e.pdf
  10. 10SORCE Insights: An Update on AI and the Data Center Buildout (Cleveland Fed District Data Brief, doi:10.26509/frbc-ddb-20261014)Mitchell Isler & Brett Huettner, Federal Reserve Bank of ClevelandOctober 14, 2026www.clevelandfed.org/-/media/project/clevelandfedtenant/clevelandfedsite/publications/cleveland-fed-district-data-briefs/cfddb-20261014/cfddb-20261014-sorce-insights-update-on-ai-and-data-center-buildout.pdf
  11. 11The Beige Book: Summary of Commentary on Current Economic Conditions by Federal Reserve District, October 2026Board of Governors of the Federal Reserve System (erstellt bei der Federal Reserve Bank of Philadelphia)October 14, 2026www.federalreserve.gov/monetarypolicy/files/BeigeBook_20261014.pdf
  12. 12KfW-Venture-Capital-Dashboard Q3 2026KfW Research (Steffen Viete, Georg Metzger)October 8, 2026www.kfw.de/PDF/Download-Center/Konzernthemen/Research/PDF-Dokumente-Dashboard/KfW-VC-Dashboard-Q3-2026.pdf
  13. 13South Asia Economic Update, October 2026 (Adopting AI for Growth): Highlights from Chapter 2, Delivering the Promise of AI in South AsiaWorld Bank (Office of the Chief Economist for South Asia)October 6, 2026thedocs.worldbank.org/en/doc/8ad688ae83bb2966f2ae7f2316431df3-0310012026/related/SAEU-Oct-2026-Chapter-2-Highlights.pdf
  14. 14When Cash Wakes Up: Inertia in Institutional Money Market Funds and the Potential Effects of AI Disruption (NBER Working Paper 35954)Ali Hortaçsu & Younggeun Yoo, National Bureau of Economic ResearchOctober 26, 2026www.nber.org/papers/w35954
  15. 15Electric Power Annual 2025, Table 2.8: Sales of Electricity to Ultimate Customers by End-Use Sector, by State, 2025 and 2024U.S. Energy Information AdministrationOctober 21, 2026www.eia.gov/electricity/annual/table.php?t=epa_02_08.html
  16. 16Item 14: Batch Zero Update (ERCOT Board of Directors Meeting, September 14–15, 2026)Electric Reliability Council of Texas (ERCOT), Jeff BilloSeptember 11, 2026www.ercot.com/files/docs/2026/09/11/14-Batch-Zero-Update.pdf
  17. 17Consultation: User Choice Conduct Requirement (Google's general search services)Competition and Markets Authority (CMA)September 23, 2026assets.publishing.service.gov.uk/media/6ab29c76fe72ed1e2b02eebe/consultation_user_choice_conduct_requirement.pdf
  18. 18Gemeinschaftsdiagnose Herbst 2026: Aufschwung mit Strukturproblemen – Finanzpolitik auf Abwegen (Kasten 1.1: Der KI-Boom gibt der internationalen Konjunktur Rückenwind)Projektgruppe Gemeinschaftsdiagnose (DIW Berlin, ifo Institut in Kooperation mit WIFO, IWH, Kiel Institut, RWI) im Auftrag des Bundesministeriums für Wirtschaft und EnergieSeptember 24, 2026gemeinschaftsdiagnose.de/wp-content/uploads/2026/09/GD_2_2026_final.pdf
  19. 19Economic Bulletin Issue 6, 2026European Central BankSeptember 24, 2026www.ecb.europa.eu/press/economic-bulletin/html/eb202606.en.html
  20. 20Financial Policy Committee Record – September 2026Bank of England, Financial Policy CommitteeSeptember 30, 2026www.bankofengland.co.uk/financial-policy-committee-record/2026/september-2026
  21. 21Measuring artificial intelligence in the UK economy using a thematic account (Methodology)Office for National StatisticsSeptember 21, 2026www.ons.gov.uk/economy/nationalaccounts/satelliteaccounts/methodologies/measuringartificialintelligenceintheukeconomyusingathematicaccount
  22. 22The Macroeconomic Effect of AI: Sizing the Software Engineering Channel (NBER Working Paper 35793)Alex Blumenfeld, Jonathon Hazell, Chen Lian & Andreas Schaab, National Bureau of Economic ResearchSeptember 28, 2026www.nber.org/papers/w35793
Authors, reuse basis and rights records
  1. 1International Monetary Fund · Reuse basis: © IMF, no open licence; our own short factual paraphrase with full citation and link only, no text, tables or figures, no IMF seal and no implied endorsement; IMF terms: fair use is noncommercial only, up to 1,000 words or one quarter of the content, whichever is less; linking is permitted if not misleading; IMF content may not be used to train language models without explicit permission; no verbatim quotation in the published text · Rights and license record
  2. 2International Monetary Fund · Reuse basis: © IMF, no open licence; our own short factual paraphrase with full citation and link only, no text, tables or figures, no IMF seal and no implied endorsement; IMF terms: fair use is noncommercial only, up to 1,000 words or one quarter of the content, whichever is less; linking is permitted if not misleading; IMF content may not be used to train language models without explicit permission; no verbatim quotation in the published text · Rights and license record
  3. 3Jon Frost, Rudraksh Kansal, Kumar Rishabh, Vatsala Shreeti & Leanne Si Ying Zhang, Bank for International Settlements · Reuse basis: © BIS 2026, all rights reserved; brief excerpts permitted with attribution; our own German paraphrase only (not an official BIS translation), with attribution and link, no graphs; no verbatim quotation in the published text · Rights and license record
  4. 4U.S. Census Bureau · Reuse basis: Public-domain work of the US federal government (17 U.S.C. § 105); our own paraphrase and link · Rights and license record
  5. 5World Bank · Reuse basis: CC BY 3.0 IGO; our own paraphrase with attribution (World Bank) and link, no figures; not an official World Bank translation, and the World Bank is not responsible for this interpretation · Rights and license record
  6. 6United Nations Conference on Trade and Development (UNCTAD) · Reuse basis: © 2026 United Nations, all rights reserved; our own short factual paraphrase and link only, no figures or tables; no verbatim quotation in the published text · Rights and license record
  7. 7Mark L. Egan, Dominic Russel, Claire Shi & Vincent Yao, National Bureau of Economic Research · Reuse basis: All rights reserved (website notice © 2026 National Bureau of Economic Research); our own short factual paraphrase of the abstract and link only; no verbatim quotation in the published text · Rights and license record
  8. 8Central Statistics Office Ireland · Reuse basis: CC BY 4.0 (CSO statistical information); our own paraphrase with attribution and link, no graphics reproduced · Rights and license record
  9. 9World Trade Organization · Reuse basis: © World Trade Organization 2026; under the WTO terms, official publications may be reproduced for non-commercial purposes only, with attribution and after informing the WTO; therefore our own short factual paraphrase with attribution and link only, no text, tables or graphics; the WTO asks websites that link to it to inform it by e-mail (to be done before publication); no verbatim quotation in the published text · Rights and license record
  10. 10Mitchell Isler & Brett Huettner, Federal Reserve Bank of Cleveland · Reuse basis: CC BY-NC 4.0 (as stated in the document); our own factual paraphrase with attribution and link only, no text or graphics; no verbatim quotation in the published text · Rights and license record
  11. 11Board of Governors of the Federal Reserve System (erstellt bei der Federal Reserve Bank of Philadelphia) · Reuse basis: Public information of the Board of Governors of the Federal Reserve System according to the website notice (citation requested); our own paraphrase and link · Rights and license record
  12. 12KfW Research (Steffen Viete, Georg Metzger) · Reuse basis: Copyright KfW; according to the KfW imprint, items in the Research section may be passed on and reproduced for information purposes with the source stated; our own paraphrase with attribution and link, no graphics; underlying data from Dealroom.co; linking is permitted under the imprint if the KfW page fills the whole browser window; no verbatim quotation in the published text · Rights and license record
  13. 13World Bank (Office of the Chief Economist for South Asia) · Reuse basis: © World Bank Group; the website terms grant only limited non-commercial use without derivatives and with attribution; therefore our own short factual paraphrase and link only, no text, tables or graphics; not an official World Bank translation; no verbatim quotation in the published text · Rights and license record
  14. 14Ali Hortaçsu & Younggeun Yoo, National Bureau of Economic Research · Reuse basis: All rights reserved (website notice © 2026 National Bureau of Economic Research); our own short factual paraphrase of the abstract and link only; no verbatim quotation in the published text · Rights and license record
  15. 15U.S. Energy Information Administration · Reuse basis: Public-domain work of the US federal government according to the EIA notice on copyright and reuse (acknowledgment with publication date requested); our own paraphrase and link · Rights and license record
  16. 16Electric Reliability Council of Texas (ERCOT), Jeff Billo · Reuse basis: Under the ERCOT terms of use, public content may be redistributed unmodified with all notices, and raw data may be used in own analyses; here our own paraphrase with attribution and link only · Rights and license record
  17. 17Competition and Markets Authority (CMA) · Reuse basis: © Crown copyright 2026, Open Government Licence v3.0; our own paraphrase with attribution and link · Rights and license record
  18. 18Projektgruppe Gemeinschaftsdiagnose (DIW Berlin, ifo Institut in Kooperation mit WIFO, IWH, Kiel Institut, RWI) im Auftrag des Bundesministeriums für Wirtschaft und Energie · Reuse basis: All rights reserved (according to the report imprint); our own short factual paraphrase with attribution and link only, no text, tables or figures; no verbatim quotation in the published text · Rights and license record
  19. 19European Central Bank · Reuse basis: © European Central Bank; under the ECB terms the information may be used freely with the ECB cited as the source, and changes must be stated explicitly; here our own German paraphrase (not an official ECB translation) with attribution and link, no graphics; under the ECB terms, if the information is included in documents that are sold, buyers must be told it is available free of charge from the ECB; no verbatim quotation in the published text · Rights and license record
  20. 20Bank of England, Financial Policy Committee · Reuse basis: © Governor and Company of the Bank of England, all rights reserved; under the legal notice use is limited to personal or internal non-commercial purposes; therefore our own short factual paraphrase and link only, no text or graphics; linking is permitted under the legal notice if fair and without implying association or endorsement; no verbatim quotation in the published text · Rights and license record
  21. 21Office for National Statistics · Reuse basis: Open Government Licence v3.0 (according to the ONS website, unless otherwise stated); our own paraphrase with attribution and link · Rights and license record
  22. 22Alex Blumenfeld, Jonathon Hazell, Chen Lian & Andreas Schaab, National Bureau of Economic Research · Reuse basis: All rights reserved (website notice © 2026 National Bureau of Economic Research); our own short factual paraphrase of the abstract and link only; no verbatim quotation in the published text · Rights and license record

How to cite

qasb Research (2026). AI and the economy: Trade in AI goods grows, productivity unclear. Topic report 10/2026, data cutoff October 26, 2026. https://www.qasb.ai/en/reports/2026-10-wirtschaft

Appendix

For every statement: evidence type, change status, scope and limitations, and the observation period, source value, comparison and uncertainty where the source gives them. The texts are the approved entries of the edition, as is its summary.

Scope and limitations of all 25 statements (A1–A25)

Edition summary

October’s originals mainly measure the investment, trade and financing side of the AI boom: AI-enabling goods accounted for almost half of the growth in world merchandise trade value in the first half of 2026, data centres are driving construction spending and costs and straining grid connection processes, and financing increasingly relies on bonds, circular stakes and a few highly valued companies. An economy-wide productivity effect has still not been measured officially; there are descriptive sector data from Ireland, a market-implied model estimate and IMF scenarios, while according to the ECB the euro area is less specialised in AI goods and German AI start-ups are raising far more venture capital in 2026. Gaps remain in official data on electricity demand, prices and consumer spending on AI outside the US.

  1. A1Back to chapter 2.0

    Evidence: observed · baseline · Source 9

    Scope
    WTO Global Trade Outlook and Statistics, October 2026 update (8 October 2026): executive summary (pp. 2 f.) and analytical chapter on AI-enabling goods (pp. 30 ff.); AI-enabling goods defined as in the World Trade Report 2025 via HS6 codes, revised for this update; WTO Secretariat estimates.
    Limitations
    US dollar values, not volumes: according to the WTO, part of the increase reflects steeply rising unit values for accelerators and memory. The definition rests on an HS6 list revised for this update, which according to the WTO is narrower; the list is not reproduced in the report. Trade is highly concentrated (in 2025 the ten largest trading economies accounted for around 85% of exports); nothing is said about use or productivity.
    Observation period
    First half of 2026 versus a year earlier, compared with 2016 to 2025; published 8 October 2026
    Source value
    47% of the growth in the value of world merchandise trade in the first half of 2026 accounted for by AI-enabling goods
    Baseline
    Share of AI-enabling goods in merchandise trade 2016 to 2023: 7 to 8%
    Comparison
    First half of 2026: share 14.8%, growth 67% year on year (2024: 16%, 2025: 31%)
    Uncertainty
    No uncertainty stated in the source; nominal values, price effects not removed

  2. A2Back to chapter 2.0

    Evidence: observed · baseline · Source 5

    Scope
    World Bank, East Asia and Pacific Economic Update, October 2026, Chapter 1; AI-related goods defined by the WTO (2025) classification of AI-enabling products; trade data from sources including the US International Trade Commission and World Bank Trade Watch.
    Limitations
    The WTO definition also includes adjacent goods such as telecommunications equipment; nominal trade values include price effects. AI-related services are not covered; nothing is said about the use or productivity effects of the imported goods.
    Observation period
    Early 2026 year on year; twelve months to April 2026
    Source value
    739 US$ billion of US imports of AI-related goods in the twelve months to April 2026
    Baseline
    Definition based on the WTO (2025) classification of AI-enabling products
    Comparison
    One third from East Asia and Pacific; early 2026 year on year nearly +60% for AI-enabling goods and nearly −25% for other goods
    Uncertainty
    No uncertainty stated in the source

  3. A3Back to chapter 2.0

    Evidence: observed · baseline · Source 1

    Scope
    IMF World Economic Outlook, October 2026, Chapter 1 (pp. 5 f. with Figure 1.5; regional outlooks pp. 11 and 13); AI hardware basket based on HS codes, IMF staff calculations; Taiwan referred to by the IMF as Taiwan Province of China.
    Limitations
    Trade-based definition; according to the IMF, net trade in AI hardware grew mainly through prices rather than volumes. No measurement of use or returns in individual countries; the statement on sub-Saharan Africa is a qualitative IMF assessment without its own indicator.
    Observation period
    Data centres since 2024, net imports in 2025; published 13 October 2026
    Source value
    57% of data centres built worldwide since 2024 located in China and the United States
    Baseline
    All data centres built since 2024 (IMF definition)
    Comparison
    China and the US: more than 75% of net imports of AI-related computing hardware in 2025; Taiwan and Korea: 62% of net trade in this hardware
    Uncertainty
    No uncertainty stated in the source

  4. A4Back to chapter 2.0

    Evidence: model estimate · baseline · Source 19

    Scope
    ECB Economic Bulletin, Issue 6/2026 (24 September 2026), section 2 Economic activity (trade) and section 1 (China); September 2026 ECB staff macroeconomic projections.
    Limitations
    A projection and assessment, not a measurement of the AI share of market-share losses; the bulletin also cites broader competitiveness problems. At the same time, according to the ECB, AI-related activity supports digital services, business investment and some exports in the euro area.
    Observation period
    Data up to the second quarter of 2026, projection horizon to 2028; published 24 September 2026

  5. A5Back to chapter 2.0

    Evidence: model estimate · baseline · Source 6

    Scope
    UNCTAD Trade and Development Report 2026, Box I.3 (pp. 43–45): product-level value-added analysis based on an estimated bill of materials and financial statements of major suppliers; more than 87% of value added traced; excluding the parallel supply chain in China.
    Limitations
    A model estimate for a single hardware product; the result depends on bill-of-materials and allocation assumptions. Nothing is said about AI software, services or user industries; indirect benefits of AI diffusion are explicitly excluded.
    Observation period
    Analysis published 9 October 2026; reference year of the financial data not stated in the box text checked
    Source value
    68% of traced value added (post-tax operating profit)
    Baseline
    Traced value added of one GB200 NVL72 rack (more than 87% of total value added)
    Comparison
    Workers together below 15% (production labour 4.3%, non-production labour 9.9%); corporate income tax 12.4%
    Uncertainty
    No uncertainty stated in the source; model-dependent estimate

  6. A6Back to chapter 2.0

    Evidence: observed · baseline · Source 19

    Scope
    ECB Economic Bulletin, Issue 6/2026 (24 September 2026), section 1 External environment; assessments made with the September 2026 ECB staff macroeconomic projections.
    Limitations
    No separate price series for AI goods; the share of China’s export price inflation is given only in rounded form. For the euro area the bulletin gives no quantified AI effect on inflation; the 2026 US inflation projection was revised up partly because of higher prices of AI-related imports, without a breakdown of the effect.
    Observation period
    May to August 2026; published 24 September 2026

  7. A7Back to chapter 3.0

    Evidence: observed · baseline · Source 4

    Scope
    Monthly construction spending statistics (Value of Private Construction Put in Place), not seasonally adjusted, release of 1 October 2026; 'Data center' category within office construction; August value preliminary.
    Limitations
    The category covers all data centres, not only AI data centres; nominal values without price adjustment; monthly values are revised. No evidence on utilisation or economic returns of the facilities.
    Observation period
    January–August 2026 versus January–August 2025
    Source value
    41.7% change from the same period a year earlier, nominal, not seasonally adjusted
    Baseline
    US$31,920 million (January–August 2025)
    Comparison
    US$45,226 million (January–August 2026, August preliminary)
    Uncertainty
    No uncertainty stated in the cited table; preliminary and revised monthly values are flagged

  8. A8Back to chapter 3.0

    Evidence: observed · baseline · Source 16

    Scope
    ERCOT Board of Directors meeting of 14–15 September 2026, agenda item 14 (document of 11 September 2026); one-off Batch Zero study of grid connections for large loads of 75 MW or more, conditional classification of 3 September 2026; ERCOT is overseen by the Public Utility Commission of Texas (PUCT).
    Limitations
    Large loads include crypto mining and other consumers besides data centres; connection requests are not realised demand and may include speculative projects. The classification refers to planning years up to 2032 and was conditional; elsewhere the document gives 125.4 GW for studied load.
    Observation period
    Classification as of 3 September 2026, pause since 3 August 2026; document dated 11 September 2026
    Source value
    66.4 GW of large loads conditionally classified as base load (204 projects)
    Baseline
    Large loads of 75 MW or more in the Batch Zero process
    Comparison
    127.9 GW (158 projects) as studied load; excluded according to the chart: 302.2 GW (373 projects)
    Uncertainty
    No uncertainty stated in the source; conditional classification, verification pending

  9. A9Back to chapter 3.0

    Evidence: observed · baseline · Source 11

    Scope
    Beige Book, October 2026 (information collected on or before 5 October 2026), reports of the Federal Reserve Banks of Kansas City (p. 40), Dallas (p. 44) and Richmond (p. 23); qualitative feedback from business and industry contacts.
    Limitations
    Anecdotal, non-representative reports without quantities; according to the Beige Book not a commentary by Federal Reserve officials. Covers data centres in general, not only AI; the cost effect is not quantified.
    Observation period
    Reporting period up to 5 October 2026; published 14 October 2026

  10. A10Back to chapter 3.0

    Evidence: observed · baseline · Source 10

    Scope
    Survey of Regional Conditions and Expectations (SORCE), special questions fielded 17 to 24 September 2026, District Data Brief of 14 October 2026; Fourth Federal Reserve District (Ohio, western Pennsylvania, eastern Kentucky, northern panhandle of West Virginia); n = 185 on involvement, n = 79 on revenue share; involvement mainly through materials (32%), professional services (26%) and equipment (24%), each as shares of involved firms (n = 80).
    Limitations
    Small, regional, non-representative sample; involvement includes indirect supply and services. Covers data centres in general, not only AI; no information on amounts or employment. The views are the authors’, not necessarily the Federal Reserve’s.
    Observation period
    Survey 17 to 24 September 2026; published 14 October 2026
    Source value
    43% of responding firms directly or indirectly involved in data centre construction
    Baseline
    All responses to the question (n = 185)
    Comparison
    Involved firms (n = 79): majority below 10% of revenue, 19% between 10% and less than half, 4% at least half
    Uncertainty
    No uncertainty stated in the source; small sample

  11. A11Back to chapter 3.0

    Evidence: observed · baseline · Source 15

    Scope
    EIA Electric Power Annual with data for 2025, Table 2.8 (sales of electricity to ultimate customers by sector and state, 2025 and 2024, thousand MWh) with notes; Form EIA-861; released 21 October 2026.
    Limitations
    The increase cannot be attributed to data centres or AI, because they are not shown separately and reclassifications affect year-on-year comparisons; this is a measurement and definition issue, not evidence of AI-driven additional consumption.
    Observation period
    2025 versus 2024; released 21 October 2026
    Source value
    1,503,310 Thousand MWh of electricity sales to commercial customers in the US in 2025
    Baseline
    1,450,941 thousand MWh (2024)
    Comparison
    Data centres and cryptocurrency mining are not shown separately in the table; changes since 2023 partly reflect improved reporting
    Uncertainty
    No uncertainty stated in the table; values final according to the EIA, year-on-year comparisons affected by reclassifications

  12. A12Back to chapter 4.0

    Evidence: observed · baseline · Source 2

    Scope
    IMF Global Financial Stability Report, October 2026, Chapter 1 (section on an AI-driven market correction, Figure 1.2); market data from sources including Bloomberg, Dealogic and the BIS with IMF staff calculations; hyperscalers as defined by Bloomberg.
    Limitations
    Describes leverage and interconnection, not the probability or timing of a correction, and no realised losses. Only publicly disclosed transactions are covered; revenue commitments are contractual intentions, not realised revenues.
    Observation period
    End-2024 to July 2026 (bond share); disclosed transactions since 2023
    Source value
    10% share of hyperscalers in global investment-grade corporate bond issuance
    Baseline
    Less than 1% (end-2024 according to Figure 1.2)
    Comparison
    2026, data through July 2026
    Uncertainty
    No uncertainty stated in the source

  13. A13Back to chapter 4.0

    Evidence: observed · baseline · Source 3

    Scope
    BIS Bulletin No 137 of 1 October 2026; 1,246 AI firms across five supply-chain layers (compute, infrastructure, data tools, models, applications); deal data from PitchBook, supply relationships from FactSet, supplemented by manual checks.
    Limitations
    Describes interconnections, not evidence of mispricing or market abuse. Deal values are total financing-round volumes, not individual investors' shares; the authors' calculations were not reviewed by PitchBook. Views are the authors', not necessarily the BIS's.
    Observation period
    2021–2025
    Source value
    46.4% of the disclosed value of AI-to-AI investment deals where a supplier or customer relationship also existed
    Baseline
    All AI-to-AI investment deals, 2021–2025 (disclosed value)
    Comparison
    By count: 16.1% of deals
    Uncertainty
    No uncertainty stated in the source; depends on PitchBook and FactSet coverage

  14. A14Back to chapter 4.0

    Evidence: observed · baseline · Source 20

    Scope
    Financial Policy Committee Record, September 2026 (meeting of 25 September 2026, published 30 September 2026), paragraphs 3 and 8 to 13.
    Limitations
    A share of issuance in a comparatively small market; the record does not name the data source, and other estimates by banks cited there were not used. A risk assessment, not a forecast; according to the FPC, markets absorbed the fall in AI-related and semiconductor stocks in July 2026 without spillovers to core markets.
    Observation period
    2026 up to 25 September 2026; published 30 September 2026
    Source value
    47% of sterling corporate bond issuance in 2026 accounted for by AI hyperscalers
    Baseline
    All sterling corporate bond issuance in 2026 up to the meeting
    Comparison
    Hyperscaler issuance in sterling significantly smaller than in US dollars and euros according to the FPC
    Uncertainty
    No uncertainty stated in the source

  15. A15Back to chapter 4.0

    Evidence: observed · baseline · Source 12

    Scope
    KfW Venture Capital Dashboard Q3 2026 (p. 8, focus on artificial intelligence); Dealroom.co data as of 1 October 2026; AI as a cross-cutting category under Dealroom’s definition (including machine learning, generative AI, computer vision, natural language processing); financing rounds of German start-ups.
    Limitations
    Commercial database; figures may be revised retroactively and may differ from other sources. A substantial part comes from two very large rounds of dual-use and defence start-ups whose technology, according to KfW, relies largely on AI; the broad market developed more weakly (89 rounds of EUR 1 million or more in the quarter overall). Nothing is said about returns or employment.
    Observation period
    Third quarter of 2026 and January to September 2026 compared with full-year 2025; data as of 1 October 2026
    Source value
    69% of German venture capital deal volume in the third quarter of 2026 going to AI start-ups
    Baseline
    Total deal volume in the third quarter of 2026: about EUR 4.5 billion
    Comparison
    AI start-ups: about EUR 3.1 billion in 61 rounds; January to September 2026 about EUR 6.3 billion versus EUR 3.2 billion in full-year 2025
    Uncertainty
    No uncertainty stated in the source; cut-off-date data subject to revision

  16. A16Back to chapter 4.0

    Evidence: observed · baseline · Source 6

    Scope
    UNCTAD Trade and Development Report 2026, Chapter I, Table I.2 (p. 27); UNCTAD compilation based on LSEG data; comparison of the 1929, 2000 and 2008 episodes with August 2026, which UNCTAD labels a phase of artificial intelligence and automation.
    Limitations
    Descriptive indicators without causal analysis; the euphoria label is an UNCTAD assessment and says nothing about the timing or likelihood of a correction. The table does not name the market explicitly for every indicator; its notes refer to the S&P 500 and listed US companies.
    Observation period
    August 2026 compared with 1929, 2000 and 2008; published 9 October 2026
    Source value
    37% of market capitalisation accounted for by the ten largest stocks (August 2026)
    Baseline
    1929: 24 to 28%; 2000: 20 to 25%; 2008: 18%
    Comparison
    Largest single stock NVIDIA 8%; corporate profits 13% of GDP (1929: 8%, 2000: 9%, 2008: 10%)
    Uncertainty
    No uncertainty stated in the source

  17. A17Back to chapter 5.0

    Evidence: observed · baseline · Source 7

    Scope
    NBER Working Paper 35953 (Egan, Russel, Shi, Yao), 26 October 2026; credit and debit card transactions (174 million consumers, 114 billion transactions) and a demand-and-supply model estimated on them; nearly all paying users were on basic plans of US$30 a month or less.
    Limitations
    A working paper without peer review; only the publicly available abstract was checked because the full-text PDF could not be retrieved. Covers paid subscriptions, not free or employer-paid use. Interests: one author is a paid consultant to the payments firm Fiserv and two authors are unpaid members of its advisory council; according to the disclosure, Fiserv had no role in the study. The conclusion of strong price pressure rests on the model.
    Observation period
    Card data through April 2026; published October 2026
    Source value
    2.9% of US consumers paying for at least one major generative AI service
    Baseline
    174 million people in the card data
    Comparison
    April 2026; annual customer retention per major provider 20–45%
    Uncertainty
    No uncertainty stated in the abstract checked

  18. A18Back to chapter 5.0

    Evidence: observed · baseline · Source 13

    Scope
    World Bank, South Asia Economic Update, October 2026 (Adopting AI for Growth), highlights of Chapter 2, released with the report on 6 October 2026; sources according to the highlights include the World Bank Enterprise Surveys AI module 2025/26, FactSet and Lightcast.
    Limitations
    Only the official Chapter 2 highlights were checked, not the full report. Comparability of the surveys for India and the US is not explained there; the supply-link result shows an association with AI exposure, not causation, and gives no quantities or values.
    Observation period
    Surveys 2025 to 2026, supply links since 2022; published 6 October 2026
    Source value
    23% of firms in India using AI
    Baseline
    United States: 43% of firms
    Comparison
    Gap wider for more sophisticated uses according to the World Bank
    Uncertainty
    No uncertainty stated in the highlights

  19. A19Back to chapter 5.0

    Evidence: observed · baseline · Source 17

    Scope
    CMA consultation on a conduct requirement for Google’s general search services (23 September 2026, deadline 9 October 2026), section 4, paras 4.14 f. (p. 27) and para 1.3 (p. 4); based on Google’s strategic market status in search and search advertising designated in October 2025.
    Limitations
    A proposal, not a final decision; the document contains no market shares of AI assistants. According to the CMA, Google’s Gemini assistant was outside the scope of the designation; whether AI assistants measurably increase competition in search is open.
    Observation period
    Consultation from 23 September to 9 October 2026

  20. A20Back to chapter 5.0

    Evidence: scenario · baseline · Source 14

    Scope
    NBER Working Paper 35954 (Hortaçsu, Yoo), 26 October 2026; dataset of money market funds held by mutual funds 2010 to 2024, two-stage structural model of portfolio managers’ attentiveness and preferences; counterfactual without investor inattention.
    Limitations
    A counterfactual model calculation, not an observation: whether and how much AI actually reduces investor inattention is not measured. Working paper without peer review; we checked the abstract. The paper previously circulated under a different title; the NBER working paper version is new in October.
    Observation period
    Data 2010 to 2024, counterfactual scenario; published 26 October 2026
    Source value
    6 Factor by which fund-months with redemptions above weekly liquid assets would become more frequent (scenario without investor inattention)
    Baseline
    Estimated model with observed investor inattention (data 2010 to 2024)
    Comparison
    Fees roughly halved, yield dispersion near zero, market concentration about 20% lower, monthly fund flows about three times as large
    Uncertainty
    No uncertainty stated in the abstract checked; model-dependent

  21. A21Back to chapter 6.0

    Evidence: observed · baseline · Source 8

    Scope
    CSO statistical release 'Productivity in Ireland 2024–2025', Artificial Intelligence chapter of 29 September 2026; AI use from the enterprise survey on e-commerce and ICT (manufacturing, construction, selected services), labour productivity per hour by sector.
    Limitations
    A descriptive comparison without causal analysis; broad sector groups; Ireland's productivity figures are strongly shaped by foreign-owned firms. Not transferable to other countries; a September source included as context for the October findings.
    Observation period
    2023–2025; published 29 September 2026
    Source value
    26.6% increase in labour productivity (euro per hour worked) in information and communication, 2023–2025
    Baseline
    €328.4 per hour worked (2023)
    Comparison
    €415.9 per hour worked (2025); AI use by firms in the sector 28.4% (2023) and 61.3% (2025)
    Uncertainty
    No uncertainty stated in the source; descriptive comparison, no causal evidence

  22. A22Back to chapter 6.0

    Evidence: model estimate · baseline · Source 18

    Scope
    Gemeinschaftsdiagnose Herbst 2026 commissioned by the German Federal Ministry for Economic Affairs and Energy, Box 1.1 (pp. 20 to 22); update of a method of the Federal Reserve Bank of St. Louis using US data up to the second quarter of 2026; completed 17 September 2026, published 24 September 2026.
    Limitations
    Gross contributions of broad investment aggregates that also include non-AI spending; no import correction, and according to the institutes the net impulse to US output is considerably smaller. Official statistics have no separate category for AI investment; for the EU no comparable analysis is possible because of data limitations.
    Observation period
    2025 and first half of 2026; published 24 September 2026
    Source value
    1 Percentage points of gross contribution of four AI-related investment types to US GDP growth (first half of 2026 versus first half of 2025)
    Baseline
    2025: 0.7 percentage points with GDP growth of 2.1%
    Comparison
    First half of 2026: GDP growth 2.4%
    Uncertainty
    No uncertainty stated in the source; imported goods not deducted

  23. A23Back to chapter 6.0

    Evidence: scenario · baseline · Source 1

    Scope
    IMF World Economic Outlook, October 2026, Chapter 1 and Box 1.2; scenarios from the GIMF model (eight regions, including the US, China and the euro area): A strong AI build-out with early productivity gains, B capacity constraints in Asia, C market correction including US equity prices 20% lower in 2026.
    Limitations
    Model scenarios, not forecasts and not measured AI effects. Results depend on assumptions, such as global total factor productivity 0.8% higher over ten years in scenario A. The IMF gives no probabilities for the AI scenarios, and effects for individual countries such as Germany are not shown. The −1.1% includes amplification through asset prices, financial conditions and sovereign bond markets; according to the IMF, the direct effect of the unwinding of AI investment is up to 0.4%.
    Observation period
    Projections and scenarios for 2026–2031; published 13 October 2026
    Source value
    -1.1% deviation of the global GDP level from baseline by 2028 (scenario C, AI market correction)
    Baseline
    IMF October 2026 baseline projection without an additional AI productivity boost
    Comparison
    Scenario A (stronger AI build-out): +0.6% by 2029; scenario C (market correction): −1.1% by 2028, including −1.8% for the US and −1.7% for Asia
    Uncertainty
    The source gives no uncertainty range for the AI scenarios; deterministic model simulations

  24. A24Back to chapter 6.0

    Evidence: open · baseline · Source 21

    Scope
    ONS methodology article on the AI thematic account (21 September 2026), section 7 with roadmap; based on the supply and use tables of the national accounts; according to the ONS, the approach was agreed by the Advisory Panel of National Accounts (APONA) in September 2026.
    Limitations
    Describes the state of official measurement, not an effect of AI. Applies to the United Kingdom; comparable plans of other statistical offices were not checked. The adoption data and the experimental list of AI companies mentioned in the article are not an estimate of value added.
    Observation period
    Roadmap 2026 to 2028; published 21 September 2026

  25. A25Back to chapter 6.0

    Evidence: model estimate · baseline · Source 22

    Scope
    NBER Working Paper 35793 (Blumenfeld, Hazell, Lian, Schaab), 28 September 2026; sensitivity of stock returns to an AI stock index depending on firms’ payroll share in software engineering, mapped into productivity gains with a model.
    Limitations
    Market expectations, not measured productivity gains; the result depends on model and identification assumptions and on the reliability of stock price reactions. Only the software engineering channel; working paper without peer review, we checked the abstract.
    Observation period
    November 2022 to December 2025, updated to mid-2026; published 28 September 2026
    Source value
    3.6% higher level of GDP (baseline model, software engineering channel, market expectation)
    Baseline
    GDP level without the AI effect according to the model
    Comparison
    6.5% with an additional effect on research and development; equivalent to a permanent 32.6% productivity increase in software engineering; more than doubled by mid-2026
    Uncertainty
    No uncertainty stated in the abstract checked

And in your case? Check it with sources.

Start an analysisFor organizations
All reports