BIS Paper Warns AI Boom Could Stall on Weak Consumer Demand, Testing Bitcoin Bull Case

A working paper from the Bank for International Settlements models an artificial intelligence boom that lifts output but gradually starves itself of consumer demand. The scenario challenges a popular argument among bitcoin investors that nearly every economic outcome, including an AI-driven productivity surge, ends with the cryptocurrency trading higher.

The contrast is sharpest when the BIS research is set beside a memo from Matt Hougan, chief investment officer at Bitwise, dated 14 October. Hougan argued that bitcoin stands to benefit across a wide range of macroeconomic paths. The BIS paper, published two days earlier, describes a fourth path his framework does not explicitly address, according to CryptoSlate.

Three routes to a higher bitcoin price

Hougan's memo sets out three broad outcomes. In the first, AI generates abundance, making intelligence-based services cheaper and pushing value towards assets that remain scarce. Bitcoin, with its fixed supply, fits that description.

In the second, an AI investment bubble bursts and tips the economy into recession. Policymakers respond with intervention and monetary expansion, which in turn eventually lifts demand for bitcoin as a hedge against currency debasement.

The third is a more measured scenario: modest productivity gains that support gradual adoption, continued inflows into exchange-traded funds and larger allocations from institutional investors.

Hougan acknowledges that drawdowns may occur along the way and that risks exist beyond the macroeconomic environment. But each of his paths rests on a common assumption, namely that abundance, instability or monetary expansion ultimately translates into additional demand for bitcoin.

The demand bottleneck

BIS Working Paper No. 1381, titled "Transformative AI and demand bottlenecks" and written by Phurichai Rungcharoenkitkul, takes a different angle. It models an economy in which AI automates a growing range of tasks while investment absorbs an ever larger share of output.

As that happens, consumption shrinks as a proportion of gross domestic product. The consumer market that ultimately pays for new products and services becomes thinner relative to the economy, and further AI innovation becomes less profitable. The innovation engine that drives the boom begins to lose fuel.

Under the paper's main calibration, the portion of output going to consumption slides from roughly 69% to 62.1% by the time additional innovation no longer pays. Long-run sustainable growth declines from a pre-AI baseline of 2% to roughly 1.6%.

The headline figures, however, would not immediately reveal the problem. Actual output growth peaks at just under 2.4% about 15 years into the AI era and remains above 2% for around 35 years. Businesses keep accumulating machinery and computing capacity even as the incentive to innovate weakens, so the economy looks robust while its underlying foundations narrow.

The paper also models a far more optimistic case in which output grows about 5.5% a year two decades in and continues to accelerate, a reminder that the bottleneck is one outcome among several rather than a forecast.

Why shares matter more than totals

The mechanism hinges on consumption's share of the economy rather than its absolute level. Total consumption can keep rising in this model. The issue is that it grows more slowly than output as a whole.

CryptoSlate illustrated the arithmetic. If output rises by 10% while consumption's share drops from 69% to 62.1%, an index of the consumer market moves from 69 to 68.31. That amounts to a 1% contraction in the consumer base inside an economy that is producing 10% more. An investment-led expansion can therefore present as strength while the spending base beneath it erodes.

Labour sits at the centre of this dynamic. The model estimates that workers' share of tasks falls from 60% to somewhere between 15% and 50% over three decades. In the demand-bottleneck calibration, real wages end up about 15% below the path they would otherwise have followed without that degree of automation.

Those results depend on significant assumptions: a fixed labour supply, a static set of tasks available to workers and wages that track marginal productivity. If new kinds of work emerge, or if wage-setting behaves differently, the outcome could shift materially. The paper's value lies less in its precise numbers than in the feedback loop it identifies.

Interest rates cut both ways

For bitcoin, the most relevant channel may be interest rates. In the BIS model, the natural rate of interest, the theoretical rate that neither stimulates nor restrains the economy, can rise during the investment boom as demand for capital strengthens, then fall once the bottleneck takes hold.

The paper's pre-AI reference rate is 1.65%. Its median estimate 20 years out is 1.2%, but the 90% model band is wide, ranging from -0.6% to 4.4%. That spread underlines how uncertain the long-term path remains.

Lower rates are often treated as supportive for risk assets, including bitcoin. But the reason rates fall matters. Declining rates that accompany a healthy productivity transition can bolster demand for assets. Declining rates driven by weak consumption, financial stress and contracting credit are another matter, and can coincide with deleveraging, when investors sell assets to pay down debt. This is the scenario that sits awkwardly with the assumption that monetary easing reliably translates into bitcoin demand.

Early signs in the data

Current economic evidence offers hints of the dynamics described, without confirming them. The Federal Reserve's Beige Book, released on 14 October, reported that businesses in five of its 12 districts were reducing headcount through attrition, postponing hiring or expecting to need fewer workers once AI-related productivity gains materialise.

Consumer spending was broadly flat, and more than half of the districts said higher-income households were responsible for driving it. Jobs held largely steady overall: three districts saw modest rises, five noted slight gains and four described conditions as flat.

The Beige Book is a qualitative survey, and this edition gathered observations up to 5 October. It describes uneven gains and unequal purchasing power, but it does not establish the self-reinforcing loop the BIS paper models.

Trade figures add another dimension. The US trade deficit widened to $105.6bn in August, as heavy purchases of foreign AI equipment pushed imports up. According to economist Dean Maki, roughly 70% of AI-related capital spending has gone on imports, meaning a large share of the investment boom's immediate economic benefit may be accruing outside the US.

What to watch

None of this invalidates the bullish case for bitcoin, and the BIS paper itself presents a high-growth alternative. But it suggests that the composition of AI-driven growth may matter as much as its pace. A boom concentrated in capital spending, with wages lagging and consumption narrowing, would produce a very different environment for scarce assets than one in which gains are broadly shared.

Indicators that may help distinguish between these paths include:

• Consumption's share of GDP relative to business investment
• Real wage growth and hiring intentions in AI-exposed sectors
• Whether falling interest rates coincide with credit expansion or contraction
• The share of AI capital spending flowing to imports

For bitcoin investors, the question raised by the BIS work is not whether AI will transform the economy, but who ends up with the income it generates, and whether enough of it reaches consumers to keep the cycle going.