J.P. Morgan’s survey of 235 clients sums up the tension well: 73% expect the S&P 500 to rise by the end of 2027, while 57% now see US Treasury yields above 5%, compared with just 2% six months ago. Investors remain confident in corporate earnings while factoring in a durably higher cost of capital.
I. AI “Doom or Boom”: Roubini’s bet meets the test of profitability
a) The AI productivity shock as seen by Nouriel Roubini. Nouriel Roubini’s optimism stood out somewhat amid the concerns voiced this week. The economist, long associated with crisis scenarios, now believes that the convergence of artificial intelligence, robotics and other innovations could lift US potential growth from around 2% to 4% a year by the end of the decade. That would be a genuine regime change. Incomes, profits and tax revenues could rise durably, improving, all else being equal, the sustainability of public and private debt.
One question remains, perhaps the most important: where will value be created, and who will reap the benefits?
The discussions in Bangkok point to several geographies of AI: the United States enjoys a technological rent and could collect a double dividend: from the models, software and computing infrastructure, but also from their diffusion across the US economy. Taiwan and South Korea benefit more immediately from the shortages of semiconductors and advanced components. This rent, however, remains tied to the capacity cycle. China is taking a different path: building its own technology ecosystem and spreading AI throughout its industrial base. Europe could benefit above all from applying these technologies in industry and services.
The distinction is fundamental: the geography of innovation does not necessarily match that of productivity gains, nor that of profits. Economies that develop the technologies can benefit from selling them worldwide, while those that adopt them can quickly improve their competitiveness without bearing the full cost of development.
America’s advantage lies precisely in its ability to combine these two sources of growth, in line with the optimistic scenario put forward by Nouriel Roubini. But how the gains will be shared between component makers, model developers, user companies and investors remains uncertain. AI could thus deliver on its macroeconomic promise without guaranteeing comparable returns to everyone who financed its development.
b) The test of profitability. Joyce Chang and the J.P. Morgan strategists bring the debate back to the question of return on capital employed. AI-related capital expenditure could grow by 85% in 2026, then by 35-40% in 2027, before slowing to 11% in 2028. Amounts would keep rising, but not at the same pace: growth would slow nearly eightfold in two years.
At the same time, AI-related debt issuance could reach $500 billion in 2027, up from $350 billion in 2026, an increase of 43%.
There is no need to read this as the failure of AI, rather as the thorny transition towards a possible wall in 2028/29. For, after the rapid build-out of infrastructure would come the test of revenues, cash flows and, ultimately, valuations.
d) A market of rents and potential overcapacity? At the panel on Asian markets, Tobias Adrian, former director of the IMF’s FSR department, identifies four layers in the AI value chain: semiconductor and memory makers, the hyperscalers that supply computing capacity, model and application developers, and finally user companies. The same revolution, then. But neither the same interests nor, necessarily, the same returns.
For now, the first rents are accruing where supply is short. Hyperscaler investment could grow by 60% to 100%, compared with around 20% for semiconductor production capacity. This mismatch supports suppliers’ margins, particularly in South Korea and Taiwan.
A scarcity rent, however, does not last forever: it could erode as supply adjusts. Hyperscalers, for their part, will have to prove they can monetise the infrastructure they are financing. Yet open-source models, competition and efficiency gains could lower the cost of AI, to the benefit of users.
Unit revenues for some producers would suffer, even if lower prices stimulated usage, a possible illustration of the Jevons paradox.
An exceptional diffusion of AI would therefore not guarantee profits for those who develop it. On the contrary, part of the value could migrate to those who use it. Productivity gains could benefit banks, manufacturers or retailers without their having borne the initial technology investment. Part of these gains would be retained as margins, another passed on to consumers through competition.
The redistribution could also be geographic. Today’s component producers will not necessarily be the countries reaping the highest productivity gains tomorrow. On this point, the analyses of Tobias Adrian and Roubini converge.
Another question remains: the risk of overcapacity. If computing infrastructure expands faster than the revenues generated by its use, some investments could become less profitable even before the macroeconomic benefits of AI are fully visible (cf. the Neocloud companies, CoreWeave and Nebius).
On top of the industrial question comes the question of financing. Joyce Chang cautiously referred to the US mortgage market before 2008. The point is not to compare technology companies with subprime borrowers, but to recall the risks of layered debt, cross-commitments and illiquid structures.
II. Central banks facing less predictable monetary transmission
a) In his conversation with John Lipsky, former First Deputy Managing Director of the IMF, Bill Dudley, former President of the New York Fed, challenged Kevin Warsh’s intention to scale back the central bank’s forward guidance. Without a sufficiently readable reaction function, markets could misinterpret economic data, and in doing so complicate monetary transmission. In the United States, transmission runs mainly through financial conditions – long-term rates, equity valuations, credit spreads – rather than through the policy rate alone. Dudley favours setting out several scenarios, so that investors understand how the Fed might react if the economy were to deviate from its central path.
The dilemma is real. Despite several years of high rates, US unemployment is still cited at around 4.2%, while wage growth has reportedly slowed to 3-3.5%. Some Fed members see this as a chance for energy-driven inflationary pressures to recede without further major tightening. Dudley takes a more hawkish view: the resilience of activity and financial conditions suggests that monetary policy may not be as restrictive as assumed. Above all, he sees the risk of error as asymmetric: excessive tightening would be easier to correct than a lasting de-anchoring of inflation expectations.
b) The view of Megan Greene (ex-BoE) provides a counterpoint. She favours a more limited path, of around two rate hikes, and warns against an overly mechanical reading of inflation persistence.
c) In China, the logic is different. The People’s Bank of China reportedly now favours quantitative credit allocation tools, notably facilities such as Pledged Supplementary Lending, over broad changes in policy rates. A cut of 10 basis points has been mentioned. But most of the support would come through targeted financing. This choice reflects the limits of conventional monetary policy when private demand remains weak and real estate investment is falling by around 20%.
Cycles are diverging, but that is not all. The channels through which central banks act are also becoming less comparable from one economy to another.
III. Wealth concentration is changing how financial shocks are transmitted
a) Fed data cited in Bangkok show a rise in direct equity ownership among US individuals since 2019. This trend is taking place in an economy where the wealthiest 1% of households hold around 32% of national wealth, some $55 trillion, and where stock market capitalisation represents nearly 250% of GDP. These figures shed light on the role the wealth effect now plays in the consumption of affluent households. Shock transmission is becoming more asymmetric: market corrections can weigh on the spending of asset holders, while less well-off households remain mainly exposed to labour income and the cost of credit. The level of the indices therefore does not tell the whole story of domestic demand. Jahangir Aziz made this point regarding the difficulties facing some US consumers.
b) The bond market is changing too. Brad Setser observes that Asian external surpluses are no longer recycled as mechanically into official reserves and US Treasuries. International savings could increasingly flow into equities and private investments. In Treasuries, private investors now carry more weight than official holders. Long-term rates could therefore become more sensitive to portfolio shifts, term premia and US Treasury issuance. The shift in expectations is striking in the J.P. Morgan survey: 57% of investors see US yields above 5%, compared with just 2% six months ago. Yet allocations remain relatively stable: 43% keep a neutral duration, 25% are underweight and 20% overweight. This is not an exodus from bonds. Rather, a different reading of rates and a higher bar for risk compensation.
c) Bill Dudley also draws attention to non-bank structures. Their cross-exposures and funding conditions are often poorly understood. The Archegos affair is a reminder that a risk does not become worrying through its size alone: counterparties’ difficulty in gauging its concentration matters just as much. The growing complexity of AI-related financing, mentioned by Joyce Chang, could compound this difficulty.
One period comes up again and again in the discussions: 2028–2029. Part of the private debt taken on during the pandemic and since GPT will mature, while the US Treasury’s financing needs could increase and growth in AI investment would be slowing. The real test could then be liquidity and refinancing rather than the headline level of the indices.
IV. Geopolitical stability masking a build-up of imbalances
a) US-China relations are relatively stable. The economic imbalances, however, persist. At a panel on the rivalry between Washington and Beijing, speakers described a form of “stability without trust”: the two powers seek to avoid an uncontrolled confrontation while pursuing their technological and industrial competition. This lull does not rebalance trade. China is expected to maintain a trade surplus of $1.2 to $1.4 trillion, or around 6-7% of its GDP. Western restrictions could even reinforce China’s drive for productive self-sufficiency, at the risk of sustaining the very overcapacity they seek to combat. With the US elections approaching, Joyce Chang expects further trade tightening, notably through Section 301 investigations. The measures could extend beyond China to European and Asian partners, as part of a push against industrial overcapacity. In the end, the fragmentation of value chains could shift trade tensions more than resolve them.
b) The Iranian issue points to another risk: instability becoming entrenched. N. Kaneva foresees neither a swift resolution nor permanent escalation, but a prolonged confrontation in which the Revolutionary Guards would retain a decisive role. 54% of investors surveyed expect structurally high energy prices, compared with only 6% expecting a new major shock. Hence this line heard in Bangkok: “$90 could become the new $65” for oil. Even without a major supply disruption, geopolitics could therefore sustain a risk premium on energy.
VI. China and India: when growth and financial performance diverge
Investor mistrust of China is striking. In the survey presented by J.P. Morgan, 50% of respondents report having no allocation to China, compared with 42-44% six months earlier, while only 1% are overweight. This disaffection reflects real estate difficulties, weak domestic demand and trade tensions. Yet this mistrust creates an asymmetry: with expectations this low, even a modest improvement could trigger a marked return of capital. This is not, in itself, a buy signal. But positioning matters as much as the economic narrative in understanding future performance.
In India, Jahangir Aziz observes almost the opposite: growth close to 8%, a current account deficit limited to 1.5% of GDP, and yet capital flows below expectations. Good fundamentals do not necessarily make a good investment: the price paid and the return outlook remain decisive. Hence the interest in targeted opportunities (Brazil, Poland, Hungary or Greece) rather than in emerging markets taken as a single bloc.
Conclusion – The promises of growth put to the test of capital
I leave Bangkok with the image of a global economy that is holding up, but whose balance calls for closer attention. AI could durably accelerate productivity, as Nouriel Roubini argues, without necessarily guaranteeing returns on the capital invested. The potential gap between wealth creation and return on capital is not confined to technology. It can be seen in bond markets, in monetary transmission and even in geopolitical imbalances.
In my view: a) The point is not to exit AI, but to diversify the drivers of performance. US technological leadership remains an asset, but index concentration and uncertainty over durable profits argue for broadening geographic and sector exposures.
b) Quality could once again become a guiding thread. Favour companies that defend their margins, generate cash flow and fund their growth without relying too heavily on debt markets. European companies deserve a fresh look in light of this criterion.
c) Emerging markets offer another avenue for diversification. So far, a large share of their performance has come from Korea and Taiwan, both closely tied to the semiconductor cycle. ASEAN and, more selectively, China could offer other sources of return, less dependent on the US technology bet.
d) On currencies, the consensus for a durably weaker dollar looks less certain. The ability of the United States to attract capital, supported by its technological lead, calls for caution. As for bonds, markets may already be pricing in a good deal of the risk of rising rates. Short and intermediate maturities could regain appeal.
Looking ahead to 2028-2029, the promises of the new economy may well come up against the constraints of their financing. Growth and return on capital do not always go hand in hand over the course of economic cycles. Hence the importance of diversification, balance sheet quality and valuation discipline. The challenge remains to take part in major economic transformations without making the portfolio dependent on a single scenario.
