Empirical Evidence Base: NoxSoft/SVRN Chain Economic Framework

Document Version: 1.0 Date: 2026-02-23 Purpose: Hard empirical data supporting the thesis that a compute-backed, cooperative, middleman-eliminating economic framework is viable and timely.

Table of Contents

1. AI Productivity and Economic Impact
2. Small Team Economics
3. Universal Basic Income / Universal Basic Compute
4. Cooperative and Stakeholder Models
5. The Broken Creator Economy
6. Middleman Value Extraction
7. Compute Cost Trends and Viability
8. Network Effects and Platform Growth

1. AI Productivity and Economic Impact

1.1 Macro-Economic Projections

Caveat: McKinsey estimates represent theoretical potential, not guaranteed outcomes. Adoption rates, regulatory environments, and structural barriers will determine realized value.

1.1b Additional Studies: BCG/Harvard and Stanford/MIT

BCG + Harvard Business School RCT (Sept 2023): "Navigating the Jagged Technological Frontier" — 758 BCG consultants (7% of BCG's total workforce). Within AI's capability frontier: 12.2% more tasks, 25.1% faster, 40% higher quality. CRITICAL CAVEAT: For tasks OUTSIDE the AI frontier, consultants using AI were 19 percentage points LESS likely to produce correct solutions. Lower-performing consultants saw largest gains. Source: Harvard Business School Working Paper 24-013 (peer-reviewed).

Stanford/MIT Customer Service Study (2023): "Generative AI at Work" — 5,000+ customer support agents at a Fortune 500 company over one year. AI-assisted agents resolved 14% more issues/hour overall. Novice workers: 35% faster. Experienced workers: minimal to zero impact. Agents with 2 months experience + AI performed like agents with 6+ months without AI. Source: arXiv:2304.11771 (peer-reviewed).

1.1c Honest Multiplier Assessment

Current evidence supports a 1.2x to 2x productivity multiplier for most knowledge work, with specific narrow tasks reaching 2-3x. Claims of 10x are not supported by peer-reviewed evidence TODAY. However, the compounding argument is strong: if AI improves 5 different workflow steps by 1.5x each, end-to-end improvement is 1.5^5 = 7.6x. This is the mechanism by which the aggregate multiplier approaches 10x — but requires full-stack AI integration, not just copilot-style assistance.

1.2 Measured Productivity Gains: Software Development

Counterargument: GitClear's 2024 analysis of 153 million lines of code found that Copilot usage correlated with increased code churn and decreased code quality metrics, suggesting speed gains may come at the cost of maintainability. The productivity picture is nuanced: AI accelerates boilerplate and repetitive tasks but may introduce technical debt.

1.3 Measured Productivity Gains: Customer Service

1.4 Measured Productivity Gains: Legal

Caveat: Stanford HAI (2024) found legal LLMs hallucinated in approximately 1 in 6 queries. AI excels at routine, well-defined review tasks but struggles with nuanced legal reasoning and judgment calls.

1.5 Measured Productivity Gains: Medical Diagnosis

Caveat: A 2024 Nature Medicine study found chest X-ray models exhibited up to 20% performance drops when tested on external datasets (generalizability problem). AI excels in narrow, well-defined imaging tasks but cannot yet replace experienced clinicians across all scenarios.

1.6 Sector-by-Sector Automation Potential

1.7 The Automation Paradox: Historical Evidence

Key mechanism: Automation reduces unit cost, which reduces price, which increases demand, which can increase total employment even as per-unit labor decreases. Jobs transform rather than disappear.

2. Small Team Economics

2.1 Landmark Small-Team Exits

Source: Company filings, SEC disclosures, and press reporting at time of acquisition. Mojang notably never raised venture capital and earned nearly $1B in profit before selling.

2.2 Modern AI Companies: Extreme Efficiency

Sources: Sacra research, GetLatka, CB Insights, PitchBook, company announcements. Revenue figures for private companies are estimates based on industry reporting.

2.3 AI Startups vs. Traditional SaaS

Source: Analysis by Growth Mind (Substack), Pavilion, and Lucidity Insights aggregating data from AI startups reaching $100M+ ARR with fewer than 100 employees (2024-2025).

2.4 Big Tech Revenue per Employee (Comparison)

Key insight: AI-native companies are achieving revenue-per-employee ratios that rival or exceed the largest tech companies, with teams orders of magnitude smaller.

3. Universal Basic Income / Universal Basic Compute

3.1 Finland Basic Income Experiment (2017-2018)

Source: Kela (Social Insurance Institution of Finland), "Results of Finland's Basic Income Experiment 2017-2018," published 2020. Peer-reviewed publications in multiple journals.

Caveat: Sample limited to unemployed individuals; 2-year duration limits generalizability to permanent programs.

3.2 GiveDirectly Kenya UBI Study (2017-ongoing)

Source: Banerjee, Faye, Krueger, Niehaus, Suri, "Effects of a Universal Basic Income during the Pandemic," published via IPA and NBER (2023-2024). Peer-reviewed.

3.3 Stockton SEED Program (2019-2021)

Source: West and Castro Baker, "Preliminary Analysis: SEED's First Year," Stockton Economic Empowerment Demonstration, 2021. Subsequently peer-reviewed.

3.4 OpenResearch Study (Sam Altman-backed, 2021-2024)

Source: OpenResearch, three published research papers (2024). Largest US basic income study to date.

3.5 Alaska Permanent Fund Dividend (1982-present)

Source: Jones and Marinescu, "The Labor Market Impacts of Universal and Permanent Cash Transfers: Evidence from the Alaska Permanent Fund," American Economic Journal: Economic Policy, 2022. Peer-reviewed.

3.6 Negative Income Tax Experiments (1968-1982)

Source: Overview of the Final Report of the Seattle-Denver Income Maintenance Experiment, ASPE (HHS); Burtless, "Income: A Survey of Experimental Evidence," Boston Fed Conference Volume 30. Peer-reviewed.

Key finding across all experiments: Even in the worst-case interpretation, work reductions were modest (2-4 weeks per year), and much of the reduction was attributable to longer job searches resulting in better job matches, increased education enrollment, and increased caregiving time.

3.7 Summary: The "Laziness" Myth vs. Data

Conclusion: Across 6 major studies spanning 4 decades, 6 countries, and tens of thousands of participants, no study found that unconditional cash transfers caused significant workforce withdrawal. The "laziness" hypothesis is empirically unsupported.

4. Cooperative and Stakeholder Models

4.1 Mondragon Corporation

Source: Mondragon Corporation Annual Results, reported via TU Lankide and Co-operative News. Company-reported data.

4.2 Worker Cooperative Survival Rates

Additional findings from Perotin:

• During the 2008 crisis, French worker co-op employment INCREASED by 4.2% while employment in conventional firms DECREASED by 0.7%.
• Worker cooperatives have similar sensitivity to business cycles as conventional firms, but employment is more stable because workers can accept temporary wage cuts (and later restore them), while conventional firms must lay off.

Source: Virginie Perotin, "What Do We Really Know About Workers' Cooperatives?" University of Leeds, published by Co-operatives UK (2016). Peer-reviewed academic research.

4.3 ESOP (Employee Stock Ownership Plan) Performance

Source: National Center for Employee Ownership (NCEO) research reports (2022-2024); Rutgers University global ESOP study (2024).

4.4 Credit Unions vs. Commercial Banks

Source: Credit union industry data aggregated from NCUA reports and CreditUnions.com analysis (2024).

4.5 John Lewis Partnership (UK)

Source: John Lewis Partnership plc Annual Report and Accounts 2024/25. Company-reported, audited data.

Caveat: JLP has faced challenges in recent years, including zero partner bonus for 2023/24 and 2024/25. The cooperative model does not guarantee superior financial performance in all periods.

4.6 Platform Cooperatives

Source: Platform cooperative case studies from MIT Center for Civic Media, Start.coop, and direct reporting. Platform cooperatives remain small-scale relative to incumbent platforms.

5. The Broken Creator Economy

5.1 Platform Take Rates

5.2 Spotify: The Math of Poverty

5.3 Creator Income Distribution

5.4 The "1,000 True Fans" Theory vs. Reality

Kevin Kelly's 2008 theory: 1,000 fans x $100/year = $100,000 sustainable income.

5.5 The Middle-Class Creator Problem

The creator economy exhibits extreme power-law dynamics:

• A tiny fraction of creators earn substantial income
• The vast majority earn below poverty wages
• Platform algorithms amplify this inequality by favoring already-popular content
• Mid-tier creators are squeezed: too many followers for personal connection, too few for algorithmic amplification
• Platform take rates compound the problem by extracting 20-45% from already meager earnings

6. Middleman Value Extraction

6.1 Agriculture

Implication: When you buy food at a restaurant, the farmer who grew it receives 5.4 cents of every dollar. The remaining 94.6 cents goes to intermediaries: processors, transporters, wholesalers, retailers, and food service establishments.

Source: USDA ERS Food Dollar Series and Price Spreads from Farm to Consumer, official government data (2023, most recent available).

6.2 Real Estate

Source: Burnett v. National Association of Realtors settlement documents (2024); industry reporting via Fortune, CBS News.

6.3 Healthcare Administration (US)

Source: Health Affairs policy brief, "The Role of Administrative Waste in Excess US Health Spending" (2022); Trilliant Health market research (2023); CMS National Health Expenditure data.

Implication: Excess administrative costs in US healthcare amount to ~2-8% of total healthcare spending compared to peer nations, representing hundreds of billions in potentially eliminable intermediation.

6.4 Higher Education Administration

Source: American Council of Trustees and Alumni (ACTA); National Center for Education Statistics (NCES); Heritage Foundation analysis. Government and non-profit data.

6.5 Ride-Sharing

Source: National Employment Law Project, "Unpacking Uber & Lyft's Predatory Take Rates" (2025 update). Advocacy research based on driver earnings data.

6.6 Food Delivery

Source: Platform merchant agreements and ActiveMenus cost analysis (2024).

6.7 Financial Services

Source: Greenwood and Scharfstein, "The Growth of Finance," Journal of Economic Perspectives 27(2), 2013. Peer-reviewed. FRED data from Federal Reserve Bank of St. Louis.

6.8 Total Middleman Extraction Summary

Conservative estimate of addressable middleman extraction in the US alone: $500B-$1T+ annually.

6.9 Global Consolidated Rent-Seeking Table

Global total: $8-9 trillion annually in intermediation costs (avoiding double-count between financial intermediation and payments). A substantial fraction (30-60%) is rent-seeking amenable to technological disruption.

7. Compute Cost Trends and Viability

7.1 GPU Price-Performance (FLOP/s per Dollar)

Source: Epoch AI (primary academic source); NVIDIA corporate communications. Epoch AI analysis is based on public GPU specification data and is peer-reviewed within the AI safety research community.

Caveat: Huang's 1000x claim decomposes to: 16x from better number handling, 12.5x from reduced-precision arithmetic, 2x from sparsity, 2.5x from process node improvements. This is real but specific to AI inference workloads, not general-purpose compute.

7.2 Storage Costs

Source: Backblaze, "The Cost Per Gigabyte of Hard Drives Over Time" (continuously updated dataset). Company-reported based on their own purchasing data as a major storage buyer.

20-year perspective: Storage costs have fallen by >90% over the past two decades, with brief interruptions (2011 Thailand flood). The long-term trend is exceptionally consistent.

7.3 Bandwidth / Internet Transit Costs

Source: DrPeering.net (William B. Norton), "Internet Transit Prices: Historical and Projections"; TeleGeography IP Transit Pricing reports.

Key dynamic: Transit prices decline ~30% per year, but traffic grows ~50% per year, meaning total spend can still increase even as unit costs drop.

7.4 Cloud Compute (AWS)

Source: AWS corporate announcements; Souiri, "Forecasting Future Amazon Web Services Pricing," ICEAA Conference (2019).

Important caveat: The era of perpetual cloud price declines may be ending for specialized AI compute. General-purpose compute continues to decline, but AI-specific instances (with NVIDIA GPUs) have seen price increases in 2024-2025 due to demand exceeding supply.

7.5 Compute Cost Projection for UCU

Based on the empirical trends:

Projection: If 1 UCU is pegged to a meaningful unit of compute (e.g., 1 hour of a standard AI inference workload), its cost basis should decline by approximately 5-10x per decade for general compute, and potentially faster (100-1000x per decade) for AI-specific workloads. This means that compute-backed currency becomes progressively MORE valuable in purchasing-power terms as the underlying substrate becomes cheaper--analogous to how productivity-linked currencies strengthen over time.

Risk: Cloud provider pricing power and demand surges (as seen in 2024-2025 AI boom) can temporarily reverse hardware cost declines at the service layer.

8. Network Effects and Platform Growth

8.1 Growth Benchmarks

8.2 Discord Growth Case Study

Source: Discord company announcements; DemandSage, BusinessOfApps aggregations.

8.3 Network Effect Taxonomy for Two-Sided Marketplaces

8.4 Adoption Curves: Hard Data

M-Pesa (Kenya, 2007-2024):

Critical mass at ~25-35% of adult population (mid-2008). 78% of users cited social reasons as primary adoption driver. Source: NBER Working Paper 16721.

UPI (India, 2016-2025):

12,700x growth in 8 years. 500M+ active users. Triggered by demonetization (Nov 2016) + zero fees + interoperability mandate. UPI now processes 50% of world's real-time digital transactions. Source: PIB India, Wikipedia.

WeChat Pay (China, 2013-2023): From 30M to 100M users in ONE MONTH (Feb 2014) via Red Envelope campaign during Chinese New Year — 16M digital envelopes sent in the first 24 hours. Jack Ma called it "a Pearl Harbor moment." Users linked bank cards to send/receive, permanently activating payment capability. Reached 1.13B users by 2023. Source: Wikipedia, Fast Company.

Facebook (2004-2012): 1M → 1B in 9 years. Achieved 80%+ penetration at Harvard within weeks. Critical mass at ~15% of target community. The Sept 2006 opening (from 12M college users) triggered 22-month sprint to 100M. Source: a16z, Wikipedia.

Universal pattern: Critical mass is not reached through gradual linear growth but through a SPECIFIC TRIGGERING MECHANISM — agent density (M-Pesa), cultural tradition (WeChat), exogenous shock (UPI demonetization), or controlled saturation (Facebook). You engineer a trigger event that tips past critical mass.

8.5 Viral Coefficient

Methodological Notes

Data Quality Assessment

Key Counterarguments to Address

1. AI productivity gains may be overstated. The GitHub Copilot RCT is promising but narrow (JavaScript tasks). Real-world software development involves architecture decisions, debugging, and coordination that AI handles poorly. GitClear data suggests AI may decrease code quality.

2. Small team economics may not scale. Instagram and WhatsApp were acquired precisely BECAUSE they could not monetize independently. Post-acquisition, both required massive teams. The question is whether small teams can build sustainable businesses, not just acquisition targets.

3. UBI experiments are short-term. The longest study (GiveDirectly Kenya) is still running. Most experiments lasted 2-3 years. Long-term behavioral effects remain uncertain. The Alaska PFD ($1,000-3,000/year) provides the longest dataset but at amounts too small to constitute a living wage.

4. Cooperatives remain niche. Despite Mondragon's success, worker cooperatives represent a tiny fraction of global economic output. Scaling challenges (capital access, governance complexity) are real.

5. Compute costs may not keep declining. The 2024-2025 AI boom has caused GPU shortages and price increases at the cloud layer. Energy costs for AI training are growing rapidly. Environmental constraints may limit the cost curve.

6. Network effects are difficult to bootstrap. For every Facebook, there are thousands of failed social networks. Survivorship bias is extreme in this domain.

Citation Index

Academic / Peer-Reviewed

• Banerjee, Faye, Krueger, Niehaus, Suri. "Effects of a Universal Basic Income during the Pandemic." NBER/IPA, 2023-2024.
• Berman. "A rising tide that lifts all boats: Long-term effects of the Alaska Permanent Fund Dividend on poverty." Poverty & Public Policy, 2024.
• Bessen, James. "Toil and Technology." IMF Finance & Development, March 2015.
• Greenwood and Scharfstein. "The Growth of Finance." Journal of Economic Perspectives 27(2), 2013.
• Jones and Marinescu. "The Labor Market Impacts of Universal and Permanent Cash Transfers: Evidence from the Alaska Permanent Fund." American Economic Journal: Economic Policy, 2022.
• Peng et al. "The Impact of AI on Developer Productivity: Evidence from GitHub Copilot." arXiv:2302.06590, 2023.
• Perotin, Virginie. "What Do We Really Know About Workers' Cooperatives?" Co-operatives UK/University of Leeds, 2016.
• West and Castro Baker. "Preliminary Analysis: SEED's First Year." Stockton Economic Empowerment Demonstration, 2021.

Government / Official Data

• Bureau of Labor Statistics (BLS). Employment data.
• Centers for Medicare & Medicaid Services (CMS). National Health Expenditure Data, 2024.
• Federal Reserve Bank of St. Louis (FRED). Financial sector GDP data.
• Kela (Finnish Social Insurance Institution). Basic Income Experiment Results, 2020.
• National Center for Education Statistics (NCES). Higher education spending data.
• USDA Economic Research Service. Food Dollar Series and Price Spreads, 2023.

Industry Reports

• Epoch AI. "Trends in GPU Price-Performance." 2022-2024.
• McKinsey Global Institute. "Notes from the AI Frontier," 2018; "The Economic Potential of Generative AI," 2023; "Agents, Robots, and Us," 2025.
• National Center for Employee Ownership (NCEO). Multiple ESOP studies, 2022-2024.
• National Employment Law Project (NELP). "Unpacking Uber & Lyft's Predatory Take Rates," 2025.
• OpenResearch. Unconditional Cash Study results, 2024.

Company-Reported / Platform Data

• Backblaze. Hard drive cost-per-gigabyte analysis (ongoing).
• DrPeering.net. Internet transit pricing historical data.
• GitHub/Microsoft. Copilot productivity studies.
• Mondragon Corporation. Annual results 2024.
• NVIDIA/Jensen Huang. Compute cost projections.
• Spotify Loud & Clear. Creator payout data.
• TeleGeography. IP transit pricing reports.

This document was compiled on 2026-02-23 from publicly available sources. All figures should be independently verified before use in formal publications. Data points marked as "company-reported" may contain self-serving bias. Peer-reviewed sources are preferred where available.