By Ernst Knacke, head of research at Shard Capital
For decades, semiconductors have been one of the most cyclical industries on the planet. Every surge in demand triggered a wave of new factories, oversupply, collapsing memory prices and years of disappointing returns. Investors learnt one lesson above all else: today’s boom is tomorrow’s bust.
But what if this time really is different?
The latest data from the Semiconductor Industry Association shows global semiconductor sales surged another 9.2% in May alone, extending what has become one of the strongest periods of demand the industry has ever experienced. Artificial intelligence has unleashed an unprecedented global race to build the infrastructure powering tomorrow’s economy.
Yet equity investors remain remarkably unconvinced. Samsung, SK Hynix and Micron – the companies supplying much of the world’s AI memory – continue to trade on single-digit forward earnings multiples. The market appears to believe history will repeat itself. Capacity will catch up, memory prices will fall, and another semiconductor bust will inevitably follow.
Perhaps.
Google’s recent TurboQuant breakthrough demonstrated that AI models can become dramatically more efficient, reducing the amount of compute and memory required to produce comparable results. If AI infrastructure becomes exponentially more efficient, perhaps demand for chips won’t grow forever.
But history offers another lesson.
In 1865, economist William Stanley Jevons observed that greater efficiency often increases consumption rather than reducing it. More efficient steam engines didn’t reduce coal demand; they made coal cheaper to use, creating entirely new industries. Economists have since called this Jevons Paradox.
See also: Knacke’s money maps: SpaceX and AI’s search for cheap energy
Artificial intelligence could follow the same path. As AI becomes cheaper to deploy, entirely new applications, from robotics and autonomous vehicles to healthcare and manufacturing, may emerge, creating even greater demand for semiconductors than today.
Which force wins? No-one knows.
Cycles have certainly not disappeared, and capital-intensive industries rarely escape the laws of economics. Yet AI may have fundamentally altered the scale and persistence of semiconductor demand.
The market is betting that history repeats. The AI bulls are betting history has changed.
Both can’t be right.
The chart below highlights the spectacular surge in semiconductor sales over the past 12 months, with year-on-year sales continuing to accelerate to levels not seen before.
