The applications of artificial intelligence are approaching an economic shift profound enough to tear up everything we know about investing and growth.
Since the global financial crisis, market debates have been dominated by secular stagnation and the bleak acceptance that 2 per cent growth is the absolute best a mature economy can hope for. Now, a serious case is emerging that AI could deliver a sustained acceleration in US productivity, fundamentally rewriting those rules.
However, we must cut through the Silicon Valley hype.
For investors, the crucial distinction is between the value a technology creates for society and the profits shareholders capture.
Getting the technology right is only half the argument. Understanding who monetises it, and what price you pay for those future earnings, is the true test.
An economy becomes richer when it produces more or better output for the resources it consumes. Historically, this meant giving workers better physical tools or improving their education.
AI takes this further by tackling a stubborn cost embedded in almost every business: the human effort required to analyse information, exercise judgment and solve complex problems.
Training a professional takes decades, and their working hours are finite. Yet, once an AI model is trained, businesses can simply run additional copies on more servers.
Expanding a company’s capacity to do useful work is suddenly decoupled from the slow, expensive process of recruiting more people.
Between November 2022 and October 2024, the cost of using a model delivering GPT-3.5-level performance plummeted by a factor of more than 280. A collapse of that magnitude makes previously prohibitive analysis viable, opening the door to entirely new business models.
Yet, software alone cannot pour concrete or build houses. The explosive economic argument, championed by Elon Musk at Davos, explores what happens when cognitive AI merges with physical robotics.
Consider a restaurant with a single chef; buying a tenth oven adds nothing to the kitchen’s output. However, if the machine itself can cook autonomously, capital investment directly expands the physical capacity to do the work.
If capable machines take over physical tasks, global productive effort becomes untethered from human demographics.
We are already seeing evidence of this.
BMW recently reported that Figure 02 humanoid robots assisted in producing more than 30,000 X3 vehicles over 10 months, independently positioning heavy sheet-metal parts for welding.
Successfully applying that capability across the physical economy would trigger an unprecedented surge in output.
The ultimate prize is using AI to accelerate the process of scientific discovery.
Google DeepMind reports that AlphaEvolve developed a scheduling improvement recovering 0.7 per cent of Google’s worldwide computing capacity. Meanwhile, AlphaFold’s database contains more than 200 million predicted protein structures, radically expanding information for medical researchers.
If AI makes research faster and cheaper, we trigger a self-reinforcing cycle where better tools design superior tools, compounding growth well beyond initial automation gains.
Why is gross domestic product not exploding already? Redesigning established businesses requires massive upfront investment and extensive retraining.
Economic history demonstrates a productivity J-curve, where heavy capital expenditure required to reorganise obscures early efficiency gains. Building sovereign-level data centres fiercely competes for capital, steel and energy today, but the true macroeconomic payoff depends on eventually using that new capacity productively.
Source: https://www.afr.com/markets/equity-markets/ai-could-change-the-speed-limit-of-economic-growth-20260920-p60xk9