WTO AI Trade Report: Jobs, GDP, and Global Impact

WTO just released yesterday, 17 Sep, a report about the impact of AI on trades, sectors, and jobs. They provide some statistics and future projections. Here is my take on it.

AI demand

The dependency chain of AI has 5 components, flowing in this order:

• Hardware, eg. GPUs, CPUs
• Cloud computing, eg. AWS, GCP
• Data, eg. internet, proprietary data
• AI models, eg. GPT, Gemini
• AI applications, eg. Microsoft Copilot, Perplexity

They point out 3 bottlenecks, semiconductor production, data, and energy, that could throttle the flow of the chain above. They estimate that by 2040, electricity output needs to grow by 19% and semiconductors and computers by 40% to meet demand.

First, given current investment and usage curves, I find the two numbers above conservative. It should be way larger than that. Second, I would not worry about long-term shortage because of two reasons:

• AI efficiency follows Moore’s law but at a much faster pace. Soon AI will consume much less computation and energy for the same tasks as now.
• More investment to R&D to find better and more efficient energy and chips to serve demand, as there are strong incentives to do so.

Country impact

They expect that by 2040, high-income countries will gain an additional 13% of GDP and a 35% increase in trades, with high confidence, as their tech and infra are already mature. For low and middle income ones, it is harder to predict. They put a range of 7%-15% for GDP, and 24%-36% for trades, depending on if they can catch up with high-income ones on the tech and infra side.

I would put a conservative bet on the lower end of the range, as low and middle income countries can get spillover of technology, but it is very hard to catch up on the infrastructure side, especially when all countries are competing for limited chips. Besides, there are barriers such as data and models are widely available in languages of rich countries, but not much in other languages. That could hamper tech adoption too.

The report suggests governments invest in digital infrastructure, such as broadband networks, data centers and reliable electricity, especially in low and middle-income countries. That I totally agree with, as catching up in tech can be done fast, but infra takes a long time, layer by layer. So better do it as soon as possible.

Labor impact

Workers in high-income countries are exposed a lot more to AI than low-income countries, 30% vs 5%. For high-skilled workers, the gap is much smaller, 68% vs 53%, which is promising. They also point out that high-skilled workers are exposed to automating their core ability like data analysis, programming, while low-skilled workers are more exposed to automating their side abilities like routine manual tasks. They imply that AI may replace knowledge workers more than routine labor. They also point out skilled labor is a major bottleneck, especially for low-income countries. They propose countries should prioritize technology, engineering and mathematics education, and equip workers with AI skills.

I agree with most of their view here, except I think labor-intensive work is safe from AI replacement for now, but may not be true in the next couple of years, when robotics advance enough to handle labor tasks efficiently.

They also discuss regulations, tariffs, etc., but I do not cover them here. If you are interested, please check out their report. I find that most numbers are reasonable, there is no hype like in the reports of consultancy firms, because of course, this is from WTO.

I’ve condensed this analysis into a slide deck (PDF) for quick reading and easy sharing.