Update on the emerging AI economy

This post compiles a few articles on AI’s emerging impact on the economy. 

If the chatter is to be believed, AI is eating the economy. One striking graphic on the importance of AI firms came from John Mauldin’s weekly newsletter.

Nowhere is AI’s role in economic growth more pronounced than in the US, contributing more than a third of US corporate investment.

Of course, the most salient manifestation of the outsized role of AI is in the stock markets (notwithstanding some pullback). 

The McKinsey Global Institute identified 18 arenas of future growth, in which they placed what they described as the AI foundation, consisting of three industries - semiconductors, cloud services, and AI software and services - at the top. 

They mapped the economic profit across the chip and compute value chain, with profits being concentrated in chip designers, foundries, cloud providers, and now memory. 

In what could have serious implications for competition, the report highlights the ever-tightening linkages downstream of the chip value chain. These linkages also raise entry barriers across the ecosystem. 

It identifies nine omniscalers - Alibaba, Alphabet, Amazon, Apple, Huawei, Meta, Microsoft, Samsung, and Tesla - as investing and dominating across the 18 arenas. The average omniscaler participated in six arenas and is rapidly expanding and deepening across and within arenas. 

These omniscalers generate more cash and invest more than others in the same arenas and those in other industries. 

Both in terms of market capitalisation and revenues, US firms dominate across the 18 arenas. 

In revenues, US firms generate over 50% of the revenues in these arenas, with those in the Greater China area (China and Taiwan) contributing 30%. Another threat to competition is the adjacency of arenas and the possibility of the omniscalers and other dominant incumbents in an arena also shifting to an adjacent arena. 

While AI dominates investments and equity markets, especially in the US, there is also growing interest in sectors that are by nature insulated from interest in sec AI development. 

Josh Brown, chief executive at Ritholtz Wealth Management in New York, has coined the term Halo, for Heavy Assets, Low Obsolescence, to signify sectors which have business models difficult for any AI model to replicate. 

These can be capital-intensive electric utilities, miners or oil companies, groups such as Italy’s Enel, or Rio Tinto and Shell in the UK… Halo businesses combine substantial physical capital with long-lived economic relevance. That means they should have sustainable business models in a world where “AI bots” could not just replace people but also any software programming and intellectual property previously thought to complement this new transitional technology… Commodities, for example, had already begun to attract interest from investors last year, helped by the weakness of the dollar.

Across the world, asset-heavy sectors like utilities, mining, and energy have been good investments, with their MSCI All Country World price indices outperforming the broad world index, even before the Iran war began in March.

On a historical scale, where does the ongoing AI capex stand, as of now?

The Economist has an article that points out that the ongoing AI capex boom may well be the largest investment surge in history. The biggest US technology companies that spent $450 bn on AI infrastructure in 2025 are expected to spend $900 bn this year, and another $1.4 trillion in 2027!

However, as the folks at a16z show in a blog post, while the AI-related arenas are dominating today and the capex surge (so far) is unprecedented, they are not in the same league as the railways or finance were in their respective periods of dominance in the nineteenth century.

One caveat here is that the 1900 economy was basically textiles, iron, coal, steel, and tobacco, the rails to transport them, and the banks to finance them, whereas today these sectors are a tiny fraction of the output.

In fact, even the $5 trillion estimated to be spent on AI and AI-related infrastructure by 2030 pales in comparison to that made on railways. 

By 1890 railway companies in the US alone had issued about $5bn worth of bonds. Adjusting for inflation that equates to about $180bn in today’s money. However, this understates the enormous scale of the undertaking, because the US economy was much smaller then. In 1890, $5bn was about one-third of America’s GDP, so the investment spree was arguably the equivalent of spending over $10tn today. It also resulted in an epic, generation-defining crash. In 1873, Jay Cooke & Co, the premier investment bank run by America’s dominant financier at the time, suddenly collapsed under the weight of unsold railway bonds. This caused a giant financial crisis and ushered in what was long known as the Great Depression, until the even larger one in the 1930s.

There are already questions being raised about the viability and sustainability of the AI capex surge. This is most salient in the valuations of the primary AI firms, which have pulled back sharply in recent months and lag many broader market indices this year. The Economist article discussed above has some very interesting revenue numbers about what would be required to justify such massive capex. 

A rough calculation finds that covering AI capex through identifiable AI income requires revenue on the order of $2.5trn per year, more than tech’s entire combined revenue today. Only a small minority of consumers seem willing to pay for personal AI subscriptions, so the real money will have to be made from selling to enterprises…For AI revenues to soar, more firms will have to use AI (what economists call an increase at the “extensive margin”) and use it more deeply (the “intensive margin”)… If, say, a third of firms across the OECD club of mostly rich countries adopt AI, then to generate $2.5trn of AI revenues the firms would have to spend about $100,000 a year on average. 

Against this requirement, the actual revenue estimates are in the range of just $150-200 bn!

Further, the signs of uptake and growth are not very promising, especially given the multiples of growth required.

In May, economists at the Census Bureau… found that about 33% of people now use AI at work, down from a peak of 46% in mid-2025… Researchers at the Bundesbank find that about half of German firms using AI do so for 5% of working hours or less. Ivan Yotzov of the Bank of England and his colleagues estimate that the average American executive uses AI for 1.7 hours a week—enough time to create a decent PowerPoint presentation, but not much more. For these dilettantes, free or ultra-cheap AI models are often good enough. Official data from Britain suggest that close to half of businesses using AI do not pay for it, presumably making do with the free tier of an American model or an open-source Chinese one… The median firm’s monthly spending per worker in June, however, was $10.66. Intuit, a software firm which tracks small and medium-sized businesses in America, Britain and Canada, reports that about one in ten has paid for a dedicated AI tool.

Furthermore, while the capex till date has been mostly driven by the large free cash flows of the omniscalers, it is now increasingly reliant on debt, drawn from risky segments like private credit markets. As a result, the four Big Tech firms are all expected to register their first deficit in years. 

And all this comes at a time when the richer world's citizens appear more wary of AI than those in the developing world.

And this

According to the Annenberg Public Policy Center, 61 per cent of Americans now oppose one being built in their backyard. The politics of the upcoming midterm elections reflect this, with candidates from both parties trying to portray politicians who favour the building of data centres as in hock to big corporations.

As a result, the regulatory free run that the AI industry has had, especially in the US, also thanks to the Trump administration, may be ending. It now stands on the substance of the technology to demonstrate its value proposition to drive the multiple-fold growth (and that too in a manner that accommodates the concerns of the political economy) required to justify the massive capex commitments. And interest rates are rising, and debt service costs are mounting. All this makes things even harder.

Some thoughts on Chinese manufacturing dominance and deglobalisation

Richard Baldwin has multiple blog posts that mine the data to add nuance to the Chinese manufacturing dominance, which, while undoubted and growing, includes some important subplots. 

For one, rapid productivity gains have widened China’s competitiveness gap but also reduced manufacturing employment, though it appears to be reversing slightly or stabilising since 2019, with Xi Jinping’s push toward high-technology industries and the Make in China 2025 campaign. But while manufacturing employment has been declining in the advanced economies, it has been rising in the emerging economies (excl China). 

The rise in emerging economies is especially pronounced since the pandemic. It has also been inching upwards in the advanced countries too.

Putting both together, Baldwin makes the case that since 2013, while China has been shedding manufacturing jobs, losing about an eighth of its base, emerging economies have been gaining and advanced economies have plateaued. However, with 134 million factory workers, China has twice more than the advanced countries combined.

Among the gainers in this giant manufacturing jobs reallocation, Vietnam, Nigeria, and Indonesia lead. Nigeria and Pakistan are surprising candidates. India has been a notable laggard. Together, the developing countries added 22 million. 

China’s export-supported factory employment is down 16% since 2013 and its share of the world total has fallen from 37% to 30%… China’s factory employment is now doing what the advanced economies’ did decades earlier: shrinking as productivity rises and as labour-intensive stages migrate to lower-wage locations.

This decline in Chinese manufacturing employment comes alongside a steeper decline in its manufacturing value added as a share of GDP. Its manufacturing GDP share has fallen over 15 years, from a peak near 32% in the mid-2000s to under 25% today. 

However, declines in manufacturing employment and share of value addition do not mean China’s dominance and vice-like grip on manufacturing is on the way down. Far from it. In fact, China’s manufacturing merchandise exports as a percentage of world GDP have been rising, even as they have declined for the rest of the world. 

In fact, the percentage has nearly doubled in the same period. It has also risen for Vietnam and India. The advanced countries have been the clear losers.

So, what explains this apparently contradictory trend of deindustrialisation in employment and value addition, co-existing with the doubling of global manufacturing export share?

The answer appears to lie in productivity. Over two decades, the price of a unit of China’s manufacturing value added fell by nearly half, which explains the whole of the decline in value addition by manufacturing. 

It reflects the most consequential fact about Chinese manufacturing: Chinese goods are getting cheaper at an astounding pace. Fast enough to distort the whole economy, maybe even the whole world economy. That price collapse is also why the Chinese are such fearsome competitors globally. It is why the rest of the world’s industry feels squeezed, and why politicians are responding with plans to hobble Chinese exports.

… there is a China-specific factor that is driving down manufacturing prices even faster. In China they call this competition “involution.” If you think the competition is tough outside of China, you should see it inside the country. China experts talk about the ruinous internal scramble in which too many domestic firms, egged on by rival local governments, drive down prices until nobody makes money.

This comes out clearly if we map the nominal and real price shares of manufacturing value addition. Since about 2014, the manufacturing sector has grown at the same real rate as the economy as a whole. 

The Chinese economy has made much more with a smaller set of resources. 

There is no doubt that the expansion of the Chinese manufacturing sector has been an era-defining phenomenon. The rise can be seen in the red line in the chart below. Real manufacturing output in 2025 is more than six times its 2005 level. That’s amazing. That’s the China shock 1.0 and 2.0 in a nutshell. Over the same twenty years, the whole economy (the grey line) grew only four-and-a-third times. So, manufacturing did outrun the economy. But only up to the mid-2010s.

Finally, he disputes the oft-cited deglobalisation argument. He points out that while the global manufacturing trade ratio fell by 2.6% in the 2008-22 period, it rose slightly by 0.3% if we exclude China. 

The main contributor was China’s GDP growing faster than its manufacturing exports.

China’s manufacturing exports boomed, but its GDP boomed more. China’s share of world manufacturing exports rose from 11.5% in 2008 to 19.3% in 2022. China’s share of world GDP went from 7.2% to 17.9%. Exports relative to China’s own GDP fell from 24.2% to 16.0% as it morphed into a normal mega-economy and became its own best customer.

Another driver of this was the reshoring of supply chains into China. However, this deglobalisation by China did not mean similar trends elsewhere. The rest of the world expanded the global value chains in their manufacturing. 

China’s domestic value-added share fell from 83% in 1995 to 72% in 2004, as the processing-trade boom stuffed Chinese exports with imported parts. It then climbed steadily to 81% by 2022. From the 2004 trough, that is nine percentage points of genuine input localisation… For the world excluding China, the domestic value-added share moved the other way: 68.7% in 2008 to 66.4% in 2022. Everyone else’s manufacturing exports became slightly more dependent on foreign inputs, not less. If you special-case China, there hasn’t been any localisation of manufacturing. The domestic content share has not risen and seems to be continuing to fall.

It would be useful to disaggregate the value addition destinations of the manufacturing exports of countries excluding China. It will not be surprising that a significant share of that value addition is coming from China. 

So, not only is China increasing the localisation of its own manufacturing, but it is also capturing a greater part of the value addition of its trade partners’ exports. 

Baldwin also points to how global export volumes continue to rise unabated, thereby contradicting talk of deglobalisation. He says that world trade never stopped growing, but only stopped outgrowing the world economy, pointing to a deceleration but not a reversal. 

None of these nuances and qualifications take away from the reality of the world economy’s China problem. Its manufacturing dominance and export dumping is among the biggest problem facing the world economy. 

China inverts the conventional wisdom of a large developing country in its development trajectory being a large importer of goods and services. Instead, even as its own market remains largely walled off, it has become a massive exporter of components and goods spanning the full spectrum of sophistication.

China also contradicts the orthodoxy of developing countries vacating lower-skilled industries and moving up the value chain as they develop. Apart from some limited vacation of space, as discussed above, China has uniquely retained its competitive advantage across the value chain of industries. Its competitive advantage spans across sectors, supply chains, and value chains.

This has also meant that even as the world has globalised supply chains, China has ended up increasing its localisation. As mentioned above, this means that China’s manufacturing dominance is boosted by both the localisation of its own manufacturing and by capturing an increasing share of the rest of the world’s supply chains. 

All this has meant that China neither offers its large consumption market nor vacates any significant part of the manufacturing landscape for its trading partners. Further, it wants to use the rest of the world as a dumping ground for its heavily distorted manufacturing industry. Furthermore, it is now showing an increased willingness to weaponise its manufacturing dominance not only for national security and strategic reasons, but to also prevent efforts to diversify supply chains away from the excessive dependence on China.