Analysis by Daniel Mercer | Numbers & Judgment Editorial

Published September 19, 2026. Daniel Mercer is an editorial byline used by Numbers & Judgment for analysis produced by the publication’s editorial team.

Artificial intelligence can make an economy more productive without immediately making its consumers richer. That distinction may become one of the most important finance questions of the AI investment cycle.

Huang Yiping, a member of the People’s Bank of China’s monetary policy committee and dean of Peking University’s National School of Development, warned at the 2026 Tsinghua PBCSF Chief Economists Forum that broader AI deployment could deepen China’s existing imbalance between strong supply and weak demand. His argument is not that AI will fail. It is almost the opposite: if AI succeeds in raising total-factor productivity faster than household income and consumption rise, productive capacity can expand faster than the demand needed to absorb it.

Productivity is not purchasing power

Most corporate AI investment cases begin on the supply side. A company can process more transactions, write more code, automate more analysis, reduce unit labor requirements or increase output with the same resources. Those are legitimate productivity gains.

But an economy has another side. Somebody must ultimately have the income and willingness to buy the additional output.

Huang’s warning is especially relevant in China because the country is already working through weak domestic demand, a prolonged property adjustment and pressure on local-government balance sheets. AI-supported productivity can strengthen the supply side while doing relatively little, at least initially, to repair household purchasing power.

Who captures the AI dividend?

The distribution of productivity gains matters. If a large share of the economic benefit accrues to owners of capital while labor income grows more slowly, aggregate production can rise faster than aggregate purchasing power. Huang invoked the historical experience of early industrialization: large productivity gains did not immediately translate into comparable improvements in worker income.

That is not a forecast that history must repeat. It is a reminder that technological productivity and broadly distributed income can operate on different timelines.

The missing line in many AI ROI models

For CFOs, this creates a useful challenge to the standard AI business case. It is not enough to estimate what AI does to costs. The model should also ask what happens to demand.

• Productivity: How much additional output or capacity does the investment create?
• Distribution: Who captures the economic benefit—employees, customers, suppliers or capital owners?
• Demand: Does the investment expand the customer’s ability or willingness to spend?
• Pricing: If competitors receive similar productivity gains, how much of the benefit is competed away through lower prices?
• Utilization: Will the organization actually have enough demand to use the additional capacity?

The final question may prove particularly important. A productivity investment can look excellent at full utilization and disappointing when industrywide capacity expands faster than demand.

Balance sheets still matter

Huang also argued that repairing the balance sheets of local governments, financial institutions and companies is necessary for stimulus to generate new activity. That observation connects the AI discussion to a more conventional finance principle: productive capacity is only one ingredient in economic activity. Borrowing capacity, confidence, income and financial resilience affect whether households and institutions can convert opportunity into spending.

AI may therefore create a paradox for policymakers and businesses. The technology can improve the economy’s ability to produce at exactly the moment when weak balance sheets constrain its ability to consume.

The judgment

The strongest case for AI does not require assuming that every productivity gain immediately becomes revenue growth. Finance leaders should separate three questions that are often collapsed into one: Can we produce more? Can we produce more profitably? And will customers have enough demand to absorb what the economy can produce?

AI may ultimately improve all three. But they are not the same proposition.

For capital allocators, the demand-side question deserves a place beside every estimate of labor savings and throughput improvement. The most consequential AI ROI assumption may eventually be not how much more the technology lets us produce, but who will have the income to buy the additional output.

Sources

• Tsinghua University PBC School of Finance / CIFER, summary of the September 19, 2026 Chief Economists Forum.
• Sina Finance, transcript-style coverage of Huang Yiping’s September 19 remarks.
• Reuters, September 19, 2026, independent reporting and macroeconomic context.