Economic benefits of open foundation models
02.08.2024 - 02:55
This study estimates the core economic benefits of open foundation models for businesses in the US economy. This considers profit uplift, operational efficiencies and cost savings in rent reductions for US businesses having access to open foundation models versus closed foundation models.
Estimating the benefits of open foundation models in AI
To inform the US National Telecommunications and Information Administration consultation on “Dual Use Foundation Artificial Intelligence Models With Widely Available Model Weights”, this study estimates the benefits of a ‘diverse’ generative AI ecosystem, where both open and closed foundation models exist and are available to access, against the counterfactual of a ‘restricted’ ecosystem where only closed foundation models are available.
Open foundation models could support faster adoption of generative AI amongst US businesses and provide an additional $1.5 trillion in benefits for them in 2035
The transparency, competition and innovation supported by open foundation models could accelerate the adoption of generative AI by US businesses in 2035
Open models allow developers to access, deploy, and modify the model… which has played a key role in growing developer activity in generative AI
Open foundation models support transparency, competition and innovation, and are likely to drive higher adoption of generative AI across the US economy.
Open model weights allows researchers and authorities to inspect, test, modify and deploy models independently. US IT professionals recognize the security benefits of open source - 90% of US IT executives believe enterprise open source is as secure or more secure than proprietary software.
As seen with other technologies, open sourcing puts downward pressure on costs and gives businesses greater flexibility and choice. The top reason businesses use open source software is its cost competitiveness.
Publishing model weights enables developers to collaborate on fine tuning and crowdsource better, safer models. The number of third-party variant models publicly released to date is over 70,000.
We modeled likely adoption curves for generative AI across the two scenarios: diverse and restricted.
In a diverse future generative AI ecosystem, there are highly-capable open and closed foundation models (similar to the current state). In this scenario, 88% of US businesses adopt foundation models.
In a restricted future generative AI ecosystem, government regulation or other competitive dynamics results in predominantly closed foundation models. In this scenario, 55% of US businesses adopt Generative AI foundation models.
Read our latest posts
Rethinking the APR
Mandala's latest research, commissioned by Block, examines whether US credit disclosure requirements remain fit for purpose in a rapidly evolving lending market. Drawing on economic analysis, a survey of nearly 1,600 US consumers and small business decision-makers, and proprietary credit data, the report finds that Annual Percentage Rate (APR) disclosures can misrepresent the perceived cost of short-term borrowing and influence consumer choices. The report makes the case for modernising credit disclosure requirements, with a more flexible approach that better reflects different credit products while preserving consumer protection and access to credit.
15 Oct, 2026
The augmented attacker: the growing AI-driven cyber threat to Australian businesses
Mandala's latest research, commissioned by CyberCX (part of Accenture), shows that the AI-driven cyber threat to Australian businesses is growing. The report finds that AI could increase the frequency and severity of cyber incidents targeting Australian businesses. As AI-driven threats grow, Australian businesses could face losses reaching up to $37 billion a year. However, investing in foundational cyber defences such as patching known vulnerabilities could reduce annual losses from $37 billion to $13 billion, a $24 billion saving. Three-quarters of the savings are from preventative actions. The report highlights the actions required from businesses and government to improve cyber defence.
13 Oct, 2026
The Stagnant Nation: Lifting Australia’s Dynamism
Australians are going backwards on real wages and income. Real wage growth has fallen from 0.6 per cent in the 2000s to -0.7 per cent today, the slowest rate since the 1990s recession. Weak productivity growth is a key driver. Productivity growth has dropped from 2.2 per cent in the 1990s to just 0.1 per cent today. The Productivity Commission has pointed to declining economic dynamism as one cause for weak growth, with fewer businesses starting, growing and competing. We confirm declining economic dynamism across five key dimensions and estimate that reversing this decline seen since the mid-2000s could see a GDP uplift of $101 billion per year over the next decade. Our analysis finds that regulation could account for around 20 per cent of declining dynamism. Reducing regulatory burden to late-2000s levels could see GDP uplift of at least $20 billion per year over the next ten years.
7 Oct, 2026
Demonstrating the economic benefits of Airwallex in Australia
Mandala's latest research, commissioned by Airwallex, quantifies the direct and indirect economic benefits of Airwallex in Australia. Airwallex supported $4.6 billion in economic benefits in 2025, a figure roughly equivalent to 2 per cent of the Australian tech sector's contribution to GDP. Of this, $2.5 billion flows directly to platform businesses through productivity gains, new market access and cost savings, with the remainder reaching the broader economy as those benefits move through customers' supply chains. Airwallex supports 29,000 jobs across the Australian economy, and a quarter of economic benefits flow to regional, rural and remote areas. Small and medium businesses capture the largest share of any business segment on the platform, $2.1 billion annually, or $200,000 in economic benefits for the average SMB each year.
16 Sep, 2026